Targeted deep sequencing reveals clinically relevant subclonal IgHV rearrangements in chronic lymphocytic leukemia

Transcription

1 Leukemia (2016), Macmillan Publishers Limited, part of Springer Nature. All rights reserved /16 ORIGINAL ARTICLE Targeted deep sequencing reveals clinically relevant subclonal IgHV rearrangements in chronic lymphocytic leukemia B Stamatopoulos 1,2,3,7, A Timbs 1,7, D Bruce 1, T Smith 4, R Clifford 1,3, P Robbe 1,3, A Burns 1,3, DV Vavoulis 3, L Lopez 5, P Antoniou 3, J Mason 1, H Dreau 1 and A Schuh 1,6 The immunoglobulin heavy-chain variable region gene (IgHV) mutational status is considered the gold standard of prognostication in chronic lymphocytic leukemia (CLL) and is currently determined by Sanger sequencing that allows the analysis of the major clone. Using next-generation sequencing (NGS), we sequenced the IgHV gene from two independent cohorts: (A) 270 consecutive patient samples obtained at diagnosis and (B) 227 patients from the UK ARCTIC-AdMIRe clinical trials. Using complementary DNA from purified CD19+CD5+ cells, we demonstrate the presence of multiple rearrangements in independent experiments and showed that 24.4% of CLL patients express multiple productive clonally unrelated IgHV rearrangements. On the basis of IgHV-NGS subclonal profiles, we defined five different categories: patients with (a) multiple hypermutated (M) clones, (b) 1 M clone, (c) a mix of M-unmutated (UM) clones, (d) 1 UM clone and (e) multiple UM clones. In population A, IgHV-NGS classification stratified patients into five different subgroups with median treatment-free survival (TFS) of 4280(a), 131(b), 94(c), 29(d), 15(e) months (Po0.0001) and a median OS of 4397(a), 292(b), 196(c), 137(d) and 100(e) months (Po0.0001). In population B, the poor prognosis of multiple UM patients was confirmed with a median TFS of 2 months (P = ). In conclusion, IgHV-NGS highlighted one quarter of CLL patients with multiple productive IgHV subclones and improves disease stratification and raises important questions concerning the pre-leukemic cellular origin of CLL. Leukemia advance online publication, 9 December 2016; doi: /leu INTRODUCTION Chronic lymphocytic leukemia (CLL) is the most common leukemia in the West, and follows a heterogeneous clinical course. Some patients present with indolent disease and do not need treatment for years, whereas others require early therapeutic intervention. 1 Using genome- or exome-wide next-generation sequencing (NGS) approaches, we and others previously demonstrated the presence of multiple subclones containing different somatic driver mutations in CLL and established their impact on treatment-free survival (TFS) and overall survival (OS). 2,3 The level of somatic mutation within the variable region of the immunoglobulin heavy chain (IgHV) is considered the gold standard of prognostication in CLL: patients with 98% of identity to the germline are considered as IgHV unmutated (UM) and have an inferior prognosis than those with more than 2% of mutations who are classified IgHV mutated (M) and who can survive decades without treatment intervention. 4,5 An exception are patients expressing the VH3-21 gene who have a poorer outcome regardless of the IgHV mutational status. 6,7 Further studies of IgHV rearrangements in CLL have identified stereotyping of the complementary-determining region 3 (ref. 8) that define CLL subsets with distinct clinical and biological features. 9 Taken together, these studies strongly suggest that leukemic clones are clonally selected in response to a particular and common antigens and might be critical in determining the clinical features and outcome for at least some CLL patients. In routine clinical practice, analysis of the IgHV locus remains technically challenging despite best practice guidelines. 10 Analysis is most commonly performed using multiplex PCR and Sanger sequencing (SSeq) meaning that without further manipulation such as gel separation/extraction or cloning only the dominant clone can be examined. In addition, it is not always possible to differentiate different clones in subclonal patients and the general failure rate ranges from 9 to 18% The cellular origin of CLL remains uncertain, but is generally thought to be a mature B cell. 14,15 However, this view is challenged by recent studies showing that driver mutations (for example, in SF3B1) can be detected in the progenitor B cells of patients with CLL. 16 Furthermore, studies in mice have shown that engrafting hematopoietic stem cells from patients with CLL into immunodeficient mice gives rise to clonal B cells with V(D)J rearrangements that are independent of the original clone. 17 Nonetheless, using conventional methods only ~ 5% of CLL are reported to have biclonal IgHV rearrangements However, using NGS, it was revealed that a significant proportion of patients with monoclonal B-cell lymphocytosis carry oligoclonal populations of CD5+CD19+ cells 21 and that distinct subclonal complementary-determining region 3 occurs in 13% of patients with CLL Oxford NIHR BRC Molecular Diagnostic Centre, John Radcliffe Site, Oxford University Hospitals, Oxford, UK; 2 Laboratory of Clinical Cell Therapy, Jules Bordet Institute, Université Libre de Bruxelles (ULB), Brussels, Belgium; 3 Nuffield Department of Laboratory Sciences, University of Oxford, Oxford, UK; 4 Department of Physiology, Anatomy and Genetics, Computational Genomics Analysis and Training Programme, MRC Functional Genomics Unit, University of Oxford, Oxford, UK; 5 Oxford Medical Genetics Laboratories, Oxford University Hospitals NHS Trust, Oxford, UK and 6 Department of Oncology, Translational Molecular Diagnostics Centre, University of Oxford, Oxford, UK. Correspondence: Professor A Schuh, Department of Oncology, Translational Molecular Diagnostics Centre, University of Oxford, Headington, Oxford OX3 7LJ, UK. anna.schuh@oncology.ox.ac.uk 7 These authors contributed equally to this work. Received 17 May 2016; revised 15 September 2016; accepted 28 September 2016; accepted article preview online 31 October 2016

2 2 We hypothesized that targeted deep NGS of the IgHV locus (IgHV-NGS) would potentially reveal multiple rearrangements and increase the prognostic precision of IgHV mutation analysis. In the present study, we, therefore, performed IgHV-NGS on two different CLL cohorts and correlated results with TFS and/or OS. Cohort (A) consisted of 270 highly purified CD19+CD5+ leukemia cells obtained at diagnosis, cohort (B) of 227 CLL patients from two fludarabine, cyclophosphamide, rituximab-based UK clinical trials. We identified multiple productive clonally unrelated IgHV subclones in 24.4% of CLL patients. Their presence raises important questions with regard to the cellular origin of CLL. MATERIALS AND METHODS Patients, sample collection and preparation, and DNA/RNA extraction This study was approved by the local Institute Ethics Committee and was based on samples collected from CLL patients after written informed consent. All CLL patients had a typical CD19+CD5+CD23+ phenotype and a Catovsky score of 4/5 or 5/5. For population A of 270 patients, samples were obtained at diagnosis before any treatment. This population is composed of unselected patients who either received treatment during disease course (55%) or patients who never required treatment (45%). Peripheral blood mononuclear cells were isolated by density gradient centrifugation over Linfosep (Biomedics, Madrid, Spain). B cells were purified with a CD19 + magnetic-bead system (MidiMACS, Miltenyi Biotec, Bergisch Gladbach, Germany) according to the manufacturers instructions. Mean B-cell purity was 499% and the mean percentage of CD5+/CD19+ cells after purification was 498%, as measured by flow cytometry. Total RNA was extracted from purified cells in a single step using TriPure Isolation Reagent (Roche Applied Science, Vilvoorde, Belgium). Complimentary DNA (cdna) was then generated by a retrotranscription using the qscript cdna Synthesis Kits (Quanta Biosciences/VWR International, Leuven, Belgium). Population B was composed of 227 CLL patients from two fludarabine, cyclophosphamide, rituximab-based UK clinical trial (AdMIRe and ARCTIC). All patients were treatment naive and samples were obtained at study enrollment. Peripheral blood mononuclear cells were isolated from CLL blood samples by ficoll gradient centrifugation. DNA was extracted from these cells using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer s protocol. Contrary to population A (45% of which did not receive treatment), patients in population B by definition all received treatment. Additional details on population A and B can be found in Supplementary Text 1, Supplementary Tables S1 and S2 and Supplementary Figure S1. Details about the assessment of prognostic factors, target sequencing and stereotyped receptors analysis are provided in Supplementary Text 2. Assessment of IgHV mutational status by Sanger and nextgeneration sequencing cdna ( ng) for population A and genomic DNA (gdna; ng) for population B were used for IgHV analysis. For SSeq, IgHV gene mutational analysis was performed using the IGH Somatic Hypermutation Assay v2.0 kit (Invivoscribe, San Diego, CA, USA) as previously described. 11,12 NGS analysis of IgHV was performed using the Lympho- Track IGH Somatic Hypermutation Assay Kit B MiSeq (Invivoscribe) according to manufacturer recommendations. Briefly, cdna/gdna was amplified using IGH LEADER (forward primers) and consensus JH (reverse primer) master mixes and PCR products were purified using magnetic beads, normalized and pooled to create a library for sequencing using the MiSeq v3 Reagent kit (600 cycles; Illumina, San Diego, CA, USA). Sequencing data were analysed using LymphoTrack bioinformatics software (Invivoscribe) and the percentage of clone was calculated using the first 200 most abundant unique sequences detected. The median of the total sequence count analysed per patient was (range, ). Clones with an abundancy of o2.5% were not considered to exclude background normal B-cell contamination and based on prognosis impact. The median count of the most abundant unique sequence was (range, ), whereas the median count of all clonotypes (the sum of all unique sequences with exactly the same VDJ rearrangement but with difference in length or percentage of mutations) was (range, ). The median coverage of VH region was 99.7% (range, ). The IMGT database was used to calculate the percentage of mutation compared with germline following ERIC recommendations. 10 Sequences with 2% deviation from any germline IgHV sequence were considered unmutated. 4,5 As NGS was able to detect different clones, we were able to create five different categories: patients with (a) multiple M clones, (b) 1 M clone, (c) a mix of M UM clones, (d) 1 UM clone (+VH3-21) and (e) multiple UM clones. Statistics TFS and OS distributions were plotted using Kaplan Meier estimates and were compared using the log-rank test. TFS and OS were calculated from the time of diagnosis until the date of first treatment and the date of death, respectively. All deaths included in this analysis were CLL-related. When the death was CLL-unrelated, the death event was not included in the OS analysis. Wilcoxon-matched pairs test was used to compare the percentage of clones between diagnosis and relapse samples. Univariate and multivariate Cox regression analysis evaluated the effects of the different prognostic variables on TFS and OS. All tests were two-sided. An effect was considered to be statistically significant if Po0.05. All analyses were performed with Graphpad Prism 5.0 software (La Jolla, CA, USA) or IBM SPSS 13.0 software (Chicago, IL, USA). RESULTS NGS: cut-off determination and exclusion of contamination with normal B cells To determine if polyclonal background normal B cells affected the data, we first sequenced purified normal B cells (mean of 97.1%, range ) from 10 age-matched normal controls (mean of 75 years old, range 59 90). This data was analyzed in the same way as the CLL samples to assess the potential effect of normal B-cell contamination, therefore only the first 200 most abundant reads were analysed. Each case demonstrated polyclonality, with the largest VJ clonotype representing 0.44% of the total number of sequenced reads and 49.9% of the 200 first most abundant (Supplementary Figure S2). Considering that our CLL samples do not contain more than 2% of CD5 /CD19+ (as determined by flow cytometry), potential contamination from normal B-cell rearrangements was, therefore, estimated at o1% of the analyzed reads. We chose 2.5% as a safe cut off for clonotype assignment. Next, we serially diluted purified CLL cells from one clonal and from one biclonal CLL sample with purified B cells taken from three healthy controls. Polyclonal rearrangements did not exceed the 2.5% cut off in any case even with a 50:50 dilution ratio and with all normal B-cell clonotypes taken into account (Figures 1a and b). We then considered the impact of using cdna and gdna from sorted and unpurified populations. Although purification had minimal effect on subclonal detection (Figures 1c and d, Supplementary Text 3), cdna had an increase sensitivity over gdna enabling the detection of further clones most likely due to the stability or the copy number difference between RNA and DNA (Figures 1e and f). However, using the cut off of 2.5%, we had a concordance of 87% (20/23) between cdna and gdna. Exclusion of PCR or sequencing errors To exclude whether the observed rearrangements resulted from sequencing errors or PCR bias we analyzed PCR fragments from 30 monoclonal patients and 30 subclonal patients using fluorescent capillary electrophoresis (Genescan, Thermo Fisher Scientific, Waltham, MA, USA): in all cases the subclonal rearrangements were confirmed (Supplementary Figure S3). In addition, we amplified IgHV using individual and specific VH family leader primer in 20 patients presenting with multiple rearrangements by NGS: for each patient, NGS-predicted VH family amplification was always observed even for low-frequency clones ( o5% of the total reads; Supplementary Figure S4). Furthermore, a comparison between singleplex IgHV-NGS and the Invivoscribe kit showed that results from both approaches were highly Leukemia (2016), Macmillan Publishers Limited, part of Springer Nature.

3 correlated (R 2 = , Supplementary Text 4 and Figure 1g). Finally for patients with differential Kappa/Lambda cell surface expression, we were able to distinguish the different clones phenotypically by flow cytometry. After sorting the different populations, SSeq was used to confirm that the presence of a different VH family rearrangement is linked to different clones (Figure 2). In addition, a recurrent mutation analysis on the sorted population clearly shows that these different populations carry different mutations (Supplementary Table S3). The LymphoTrack bioinformatics software provides all sequences of the 200 first clones that differ by size or sequence. In CLL patients, the majority of these clones belong to the same Macmillan Publishers Limited, part of Springer Nature. Leukemia (2016), 1 9

4 4 clonotype. The differences in sequence observed could, therefore, be due to error of PCR/sequencing. To control these potential PCR/sequencing errors, only the most abundant sequences with a median depth of reads (range, ) were used to calculate the IgHV mutational status. We compared the NGS sequence/mutational status for the most abundant clone with the Sanger sequences for population A: for the IgHV status determination, techniques are concordant in 99.6% of the cases (235/236). The only discordance was in a biclonal patient where the small unmutated IgHV clone was picked up by SSeq while the major clone was mutated. NGS and SSeq found exactly the same sequence in 97.9% of the cases (231/236). In five patients, we observed a discordance of one base. In four of the five cases, NGS sequences were the closest to the germline, suggesting that the error was most likely with SSeq. Further details on independent bioinformatics validation, reproducibility of NGS is given and discussed in Supplementary Text 5, 6 and Supplementary Table S4. IgHV mutational status by NGS refines IgHV SSeq classification and defines five different prognostic subgroups in an unselected population (population A) As NGS was able to identify different clones, we created five categories based on the number and the mutational status of clones: patients with (a) multiple M clones, (b) 1 M clone, (c) a mix of M UM clones, (d) 1 UM clone (+VH3-21) and (e) multiple UM clones. As productive and unproductive rearrangements could co-exist in the same cellular clone, 22 we excluded unproductive rearrangements and found 24.4% (66/270) of patients with multiple productive rearrangements. The median frequency of the second most abundant productive clone was 10.3% (range %). Using productive rearrangements only, our NGS-IgHV classification was able to classify this cohort into five subgroups with median TFS of 4280 (a), 131 (b), 94 (c), 29 (d) and 14.5 (e) months (Po0.0001; Figure 3a), and a median OS of 4397, 292, 196, 137 and 100 months (Po0.0001; Figure 3b). Next, we tested whether NGS-IgHV was able to predict treatment initiation within 4 years of diagnosis. NGS-IgHV had a positive predictive value of 80% in patients with multiple UM clones and a negative predictive value of 95% in patients with multiple M clones, subgroups that together represent 11% of patients in our cohort. NGS-IgHV could also predict an OS of o10 years from diagnosis with a positive predictive value of 75% for multiple UM and a negative predictive value of 100% for multiple M subgroups. For patients with only one clone (M or UM), similar positive predictive value and negative predictive value were observed between SSeq and NGS (Supplementary Table S5). Interestingly, 16% (42/270) of patients in population A and 18% (41/227) in population B had at least one unproductive rearrangement. We, therefore, analyzed the prognostic significance of unproductive rearrangements and demonstrated for the first time that they have a similar prognostic impact compared with productive rearrangements (Supplementary Text 7 and Supplementary Figure S5). On the basis of productive and unproductive rearrangements, NGS-IgHV status was able to identify 34.8% (94/270) patients with multiple rearrangements and to classify patients in five different groups with decreasing TFS (Po0.0001) and OS (P = ; Figures 3c and d). In Binet Stage A patients, NGS-IgHV classification remains a significant prognostic marker for TFS and OS (Supplementary Figure S6). In addition, we also verified that the heterogeneous received treatments did not introduce an OS bias (Supplementary Text 8 and Supplementary Figure S7). SSeq identified clonal rearrangements in 239 of 270 patients (89%), whereas NGS was able to identify clonal rearrangements in all patients, even those with multiple rearrangements. Using this new NGS classification including both productive and unproductive rearrangements, we found patients with differential prognosis among patients already classified by SSeq (Supplementary Figure S8). IgHV by NGS defines a very poor prognostic subgroup in a treatment-selected cohort (population B) In order to validate our new NGS model in a homogeneous clinical trial population, we applied this new classification in population B composed of 227 patients treated with fludarabine, cyclophosphamide, rituximab-based regimen. The number of multiple productive rearrangements detected was less than population A at 11.5% (26/227). As all of population B had received treatment (introducing a bias in patient selection), our five NGS-IgHV categories model was not confirmed in this population: indeed, the great majority of multiple-mutated patients (85% in population A) never required treatment and were, therefore, excluded by definition from this clinical trial population. When only treated patients were analyzed in population A, the same results were observed (data not shown). However, importantly, the poor prognosis of multiple UM patients observed in cohort A was confirmed: in population B, multiple UM patients had a median TFS of 2 months (P = ; Figure 4a). Multivariate survival Cox regression analysis Survival analyses with Cox regression in population A were used to determine hazard ratio (HR) of prognostic factors. CD38 and IgHV mutational status by SSeq were binarized using the previously reported cut off. 23 For IgHV mutational status by NGS, the five categories (a) multiple M; (b) 1 M; (c) mix M UM; (d) 1 UM (+VH3-21) and (e) multiple UM were named as 0, 0.25, 0.5, 0.75 and 1, respectively. In other words, HR represents the HR between group (a) (very favorable prognosis) and (e) (very unfavorable prognosis) taking into account the intermediate groups such that (a)o(b)o(c)o(d)o(e). 24 Using univariate analysis on patients with only productive rearrangements, we observed that NGS-IgHV mutational status displayed higher HR compared with all tested factors for TFS (26.84, 95% confidence interval , Po0.0001) and OS (20.37, 95% confidence interval , Figure 1. NGS is able to detect multiple specific VH family rearrangements independently of a polyclonal normal B-cell background. Monoclonal CLL (a) and biclonal CLL (b) were contaminated with increasing percentage of purified normal B cells from three healthy donors. DNA of the mix was extracted and subjected to NGS of the IgHV locus. Frequency of each clonotype was calculated for the 200 most abundant clones. Polyclonal B cells were not detectable in CLL samples with up to 50% normal B-cell contamination. (c) (cdna) and (d) (gdna) present the correlation between the frequency of the clonotype obtained using NGS-IgHV analysis from purified vs unpurified CLL cells. Purification has a limited impact on NGS-IgHV analysis. However, in case of cdna, the purification allows the detection of additional clones. (e) (CD19 purified) and (f) (unpurified) present the correlation between the frequency of the clonotype obtained using NGS-IgHV analysis from gdna vs cdna. Although the correlation is lower (but significant), using the cut off of 2.5%, we had a concordance of 87% (20/23) between cdna and gdna. (g) For six CLL patients, we performed seven individual and separate PCR reactions, each with one of the seven VH-leader forward primer and JH reverse primer. The seven reactions were pooled and sequenced by NGS. In parallel, the samples were also amplified using the Invivoscribe kit (a single multiplex PCR with all the seven VH-leader primers) and also sequenced. The frequency of major clonotypes between individual reaction and multiplex PCR was significantly correlated with a R 2 of and no additional clone was detected base on our 2.5% cut off. Leukemia (2016), Macmillan Publishers Limited, part of Springer Nature.

5 5 Figure 2. Phenotypic analysis validated the presence of different IgHV rearrangements. Peripheral blood mononuclear cells from CLL patients were analyzed by flow cytometry using Kappa/Lambda/CD5/CD19 fluorochrome-coupled antibody. After gating the leukemic cells (CD5 +/CD19+), we sorted clones based on their Kappa/Lambda expression, extracted RNA, produced a cdna and sequenced it by SSeq. SSeq confirmed the presence of different clones as well as the accuracy of the sequence provided by NGS. (a and b) showed two representative patients. Po0.0001). In a multivariate stepwise analysis (n = 180) that included five different factors (IgHV mutational status by SSeq, CD38, presence of a deletion 17p, presence of a deletion 11q, as well as the hierarchical classification of Döhner et al. 25 ) and IgHV mutational status by NGS, only IgHV mutational status by NGS (Po0.0001), CD38 (P = ) and the presence of a 17p deletion (P = ) were selected as independent and significant predictors of TFS. For OS multivariate analysis, CD38 (P = ) and IgHV mutational status by NGS (P = ) were selected (Supplementary Table S6). For 171 patients of population B, we also performed a Cox regression analysis including recurrent mutation analysis by target NGS (TP53, ATM, NOTCH1, SF3B1 and BIRC3), cytogenetic abnormalities (deletion 17p, 11q, 13q and trisomy 12) analysis, IgHV mutational status by SSeq (mutated or unmutated) and either multi-um or not by NGS: only this last one was considered 2016 Macmillan Publishers Limited, part of Springer Nature. Leukemia (2016), 1 9

6 6 Figure 3. IgHV by NGS defines five different prognostic subgroups in an unselected CLL population. TFS and OS, according to our new NGS- IgHV classification, were plotted with Kaplan Meier methods for the whole population A when only productive rearrangement was considered (a and b) and when productive and unproductive rearrangement were considered (c and d). Difference between curves was assessed by an univariate analysis using the log-rank test and median survival for each subgroup of patients was provided. as an independent prognostic factor for TFS prediction (P = , HR = 2.21, 95% confidence interval /Supplementary Data; Supplementary Table S7) and the NGS-IgHV classification showed no association with recurrent non-ighv mutations (Figure 4b). Productive IgHV rearrangements from the same patient are phylogenetically unrelated NGS-IgHV identified multiple productive rearrangements in 24.4 (66/270) and 11.5% (26/227) of patients in population A (cdna) and B (gdna), respectively. Within these 497 patients, the median variant allele frequency of the three most abundant clones was 76% (range, ), 8% (range, ) and 7% (range, ), respectively. Furthermore, NGS analysis of stereotyped receptors showed that intrapatient sequences were phylogenetically unrelated (Supplementary Text 9). We did not find significant differences in VDJ repertoire between patients with and without IgHV subclones (Supplementary Figure S9). All of these clones also had different D and J genes confirming they are unrelated as shown by a Circos plot visualization of the different VDJ association (Supplementary Figure S10 and Supplementary Tables S8 and S9). Treatment effects and natural clonal evolution of multiple productive NGS-IGHV clones To investigate the influence of treatment on IgHV subclonal composition, we screened 42 patients at diagnosis and after relapse. We did not find any significant difference in terms of clones and clone percentage between diagnosis and relapse. However, in subclonal patients, in two cases we observed the emergence of a new minority clone and consequently a decrease in the major clone (Supplementary Text 10, Supplementary Figures S11 and S12). An analysis of recurrently mutated non- IgHV genes in CLL confirmed that the diagnosis and relapse clone share similar mutations but that clonal selection can occur as shown by the appearance of new mutations after relapse (Table 1). DISCUSSION The definition of 'monoclonality' of malignant disease has recently been challenged by genome-wide analyses of cancers and hematological malignancies including CLL. Using conventional approaches, the percentage of CLL patients with more than one Leukemia (2016), Macmillan Publishers Limited, part of Springer Nature.

7 7 Figure 4. IgHV by NGS defines a very poor prognostic subgroups in treatment-selected population and is independent from other prognostic markers. Taking into account only productive rearrangements, the Kaplan Meier for the multiple UM patients against all other patients for clinical population B is provided in (a). Difference between curves was assessed by an univariate analysis using the log-rank test and median survival for each subgroup of patients was provided. (b) Heatmap showing the distribution of the recurrent mutation and cytogenetic abnormalities in the different NGS-IgHV subgroup. IgHV clone has been reported as rare (2.7, (ref. 19) and 5.0 (ref. 20) ). By contrast, the percentage of patients with multiple productive IgHV subclones identified by NGS in our unselected population is high (24%). This can be explained by increased sensitivity of NGS and its ability to amplify and sequence mixtures of DNA templates more reliably compared with SSeq. In a series of validation experiments, we exclude contamination with normal B cells and confirm our results in independent experiments using SSeq, total RNASeq and alternative bioinformatics approaches. Several reports have previously described the presence of multiple CLL clones 18,22,26 as well as multiple clones with a discordant IgHV mutational status. 19,27 The presence of multiple productive rearrangements in one quarter of CLL patients could be explained by a lack of allelic exclusion. 20 However, a recent study using single-cell sequencing demonstrated that the allelic exclusion is always present in CLL 22 suggesting that multiple rearrangements correspond to multiple leukemic subclones as was previously assumed. 18,19 Another explanation could be receptor editing, 28 but the nature of the rearrangements we observed also exclude this possibility. In 3 patients of 64 analyzed, we were able to demonstrate the presence of multiple subclones by flow cytometry based on kappa/lambda light-chain expression. Taken together, these results suggest that the rearrangements observed do not occur in the same cell and imply that leukemiainitiating events might occur prior to IgHV rearrangement in an un-rearranged early B-cell progenitor. Although our study suggests that these subclones are unrelated to each other, a recent work shows that immunoglobulin B-cell receptors of unrelated clonotypes may recognize common epitopes that were unpredictable from the molecular features of the immunoglobulin B-cell receptor, including the complementary-determining region 3 composition and length. 29 Further in-depth studies of the hematopoietic stem cell and early B-cell progenitor compartments in these patients will be required to confirm these observations. In the present study, we showed that IgHV-NGS was able to stratify patients in five different categories with respect to TFS and OS in an unselected CLL population: patients with multiple M rearrangements, had the better prognosis, even compared with patient with only one M clone, in line with Visco et al. 19 who observed that biclonal M patients had significantly lower CD38, more favorable cytogenetic lesions and a more indolent clinical course. In contrast, patients with multiple UM rearrangements had the worst prognosis and patients with a mix of M UM clones an intermediate prognosis. However, these authors did not find significant differences in prognosis between multiple M/1 M, or in multiple UM/1 UM. This might be due to the decreased sensitivity of the techniques used. Our new NGS-IgHV classification is able to precisely classify patients with multiple IgHV rearrangements for which SSeq is inconclusive and improved prognostication for 92 out of 270 patients. It remains to be seen how our classification into five categories (instead of two with SSeq) will impact on further improving the prediction of long remissions or even cure following chemoimmunotherapy in light of the recently published German CLL8 Study data. 30 Importantly, the poor prognosis of multiple UM patients detected by NGS-IgHV could be confirmed in an independent clinical trial population and showed a very short median TFS of 2 months. The presence of multiple IgHV rearrangements was the strongest independent prognostic indicator in regression analysis of 11 different variables including important driver mutations. Intriguingly, the inclusion of patients with unproductive rearrangements or patients treated with new therapies did not alter the significance of our model strongly highlighting the robustness of this new classification. Finally, we investigated treatment effect on clone frequencies as well as natural clonal evolution. Consistent with previous observations, our data demonstrated no change in clone type or in clone percentage between diagnosis and relapse: indeed the relapse clones had exactly the same VDJ rearrangements with the same IgHV hypermutations indicating that the original leukemic clone was able to escape treatment toxicity by the acquisition of 2016 Macmillan Publishers Limited, part of Springer Nature. Leukemia (2016), 1 9

8 8 Table 1. Target sequencing of recurrent mutations of sample at diagnosis and after relapse Gene Variant Chrom. Coordinate At diagnosis After relapse Altered VAF Read depth Alt read depth Altered VAF Read depth Alt read depth Patient 1 IgHV / % of identity / frequency VH3-9 a 01 /100%/ 100% VH3-9 a 01 / 100% / 100% NOTCH1 CAG4CAG/ C EGR2 T4T/C ATM a T4T/C Patient 2 IgHV / % of identity / frequency VH5-51 a 01 / 100% /84.2% VH5-51 a 01/100%/97.3% VH3-48 a 03 / 100% / 14.7% VH3-48 a 03 / 100% /2.6% SF3B1 C4C/T ATM a T4T/C ATM a C4C/G Patient 3 IgHV / % of identity / frequency VH4-34 a 01 / 95.56%/ 100% VH4-34 a 01 / 95.56% / 100% NFKBIE A4A/C POT1 a C4C/A ATM a A4A/C CHD2 a G4G/C Patient 4 IgHV / % of identity / frequency VH4-34 a 01 / 91.81% / 94.4% VH4-34 a 01 / 91.81% / 93.2% VH3-15 a 01 / 98.34% / 5.5% VH3-15 a 01 / 98.34% / 3.4% SETD2 a T4T/G Patient 5 IgHV / % of identity / frequency VH3-48 a 04 / 100% / 100% VH3-48 a 04 / 100% / 100% XPO1 C4C/G POT1 a C4C/G Patient 6 IgHV / % of identity / frequency VH1-69 a 02 / 100%/ 100% VH1-69 a 02 / 100%/ 100% SETD2 C4C/T ATM AT4AT/A SAMHD1 A4A/T SAMHD1 T4T/TA Patient 7 IgHV / % of identity / frequency VH3-30 a 02 / 100% / 100% VH3-30 a 02 / 100% /100% CHD2 a T4T/C RPS15 C4C/A Patient 8 IgHV / % of identity / frequency VH1-2 a 02 / 100% /100% VH1-2 a 02 / 100% / 100% ATM A4A/C ATM a A4A/C TP53 C4C/G RPS15 G4G/A Patient 9 IgHV / % of identity / frequency VH3-23 a 04 / 95.95% /100% VH3-23 a 04 / 95.95% / 100% SF3B1 T4T/C POT1 C4C/A TP53 C4C/T TP53 C4C/G Patient 10 IgHV / % of identity / frequency VH3-30 a 03 / 92.4% /100% VH3-30 a 03 / 92.4% /100% ATM a A4A/T ATM a A4G/G IRF4 AGC/GGA TP53 C4C/T Patient 11 IgHV / % of identity / frequency VH1-24 a 01 / 99.7% / 100% VH1-24 a 01 / 99.7% / 100% ITPKB a G4T/T TP53 C4C/T Abbreviation: VAF, variant allele frequency. a Validated germline polymorphism in dbsnp. VAFs are highlighted in bold. Leukemia (2016), Macmillan Publishers Limited, part of Springer Nature.

9 new recurrent mutations or by hiding in microenvironmental niches. 31,32 In conclusion, here we show for the first time that 24.4% of CLL patients present with multiple productive IgHV subclonal rearrangements by NGS that are likely to contribute to the biological heterogeneity of this disease and point toward leukemia-initiating events that pre-date IgHV rearrangement. From a practical perspective, NGS-IgHV resolves the diagnostic uncertainties in a significant number of patients, enables the detection of small subclones and refines the previous SSeq-based IgHV prognostic classification system. Finally, we propose a new NGS-IgHV classification with five different clinically relevant prognostic categories. CONFLICT OF INTEREST The authors declare no conflict of interest. ACKNOWLEDGEMENTS This research was supported by the 'Wallonie-Bruxelles International World' (WBI World), the IRIS-Recherche fund, and the Bekales Foundation and 'Les Amis de l Institut Bordet'. BS is supported by a Télévie grant (FRS-FNRS Fonds de la Recherche Scientifique FNRS). AS, PR, AB, DV, PA and JM are supported by the NIHR Biomedical Research Centre (Oxford). DISCLAIMER The views expressed are those of the authors and not necessarily those of the Department of Health. AUTHOR CONTRIBUTIONS BS and AT were the principal investigators and took primary responsibility for the paper; BS and AT performed research and wrote the manuscript. DB, HD and JM provided help for the next-generation sequencing experiments. DV, LL and PA performed stereotyped receptor analysis, coverage checking and verified the statistics. TS performed the bioinformatics validation. RC, PR and AB participated to the scientific discussion and revised the manuscript; AS supervised and revised the manuscript and the research. REFERENCES 1 Hamblin T. Chronic lymphocytic leukaemia: one disease or two? Ann Hematol 2002; 81: Schuh A, Becq J, Humphray S, Alexa A, Burns A, Clifford R et al. Monitoring chronic lymphocytic leukemia progression by whole genome sequencing reveals heterogeneous clonal evolution patterns. Blood 2012; 120: Landau DA, Carter SL, Stojanov P, McKenna A, Stevenson K, Lawrence MS et al. Evolution and impact of subclonal mutations in chronic lymphocytic leukemia. Cell 2013; 152: Hamblin TJ, Davis Z, Gardiner A, Oscier DG, Stevenson FK. Unmutated Ig V(H) genes are associated with a more aggressive form of chronic lymphocytic leukemia. Blood 1999; 94: Damle RN, Wasil T, Fais F, Ghiotto F, Valetto A, Allen SL et al. Ig V gene mutation status and CD38 expression as novel prognostic indicators in chronic lymphocytic leukemia. Blood 1999; 94: Tobin G, Thunberg U, Johnson A, Thorn I, Soderberg O, Hultdin M et al. Somatically mutated Ig V(H)3-21 genes characterize a new subset of chronic lymphocytic leukemia. Blood 2002; 99: Thorselius M, Krober A, Murray F, Thunberg U, Tobin G, Buhler A et al. Strikingly homologous immunoglobulin gene rearrangements and poor outcome in VH3-21-using chronic lymphocytic leukemia patients independent of geographic origin and mutational status. Blood 2006; 107: Agathangelidis A, Darzentas N, Hadzidimitriou A, Brochet X, Murray F, Yan XJ et al. Stereotyped B-cell receptors in one-third of chronic lymphocytic leukemia: a molecular classification with implications for targeted therapies. Blood 2012; 119: Bomben R, Dal Bo M, Capello D, Forconi F, Maffei R, Laurenti L et al. Molecular and clinical features of chronic lymphocytic leukaemia with stereotyped B cell receptors: results from an Italian multicentre study. Br J Haematol 2009; 144: Ghia P, Stamatopoulos K, Belessi C, Moreno C, Stilgenbauer S, Stevenson F et al. ERIC recommendations on IGHV gene mutational status analysis in chronic lymphocytic leukemia. Leukemia 2007; 21: Krober A, Seiler T, Benner A, Bullinger L, Bruckle E, Lichter P et al. V(H) mutation status, CD38 expression level, genomic aberrations, and survival in chronic lymphocytic leukemia. Blood 2002; 100: Austen B, Powell JE, Alvi A, Edwards I, Hooper L, Starczynski J et al. Mutations in the ATM gene lead to impaired overall and treatment-free survival that is independent of IGVH mutation status in patients with B-CLL. Blood 2005; 106: Burger JA, Keating MJ, Wierda WG, Hartmann E, Hoellenriegel J, Rosin NY et al. Safety and activity of ibrutinib plus rituximab for patients with high-risk chronic lymphocytic leukaemia: a single-arm, phase 2 study. Lancet Oncol 2014; 15: Seifert M, Sellmann L, Bloehdorn J, Wein F, Stilgenbauer S, Durig J et al. Cellular origin and pathophysiology of chronic lymphocytic leukemia. J Exp Med 2012; 209: Chiorazzi N, Ferrarini M. Cellular origin(s) of chronic lymphocytic leukemia: cautionary notes and additional considerations and possibilities. Blood 2011; 117: Damm F, Mylonas E, Cosson A, Yoshida K, Della Valle V, Mouly E et al. Acquired initiating mutations in early hematopoietic cells of CLL patients. Cancer Discov 2014; 4: Kikushige Y, Ishikawa F, Miyamoto T, Shima T, Urata S, Yoshimoto G et al. Selfrenewing hematopoietic stem cell is the primary target in pathogenesis of human chronic lymphocytic leukemia. Cancer Cell 2011; 20: Plevova K, Francova HS, Burckova K, Brychtova Y, Doubek M, Pavlova S et al. Multiple productive immunoglobulin heavy chain gene rearrangements in chronic lymphocytic leukemia are mostly derived from independent clones. Haematologica 2014; 99: Visco C, Moretta F, Falisi E, Facco M, Maura F, Novella E et al. Double productive immunoglobulin sequence rearrangements in patients with chronic lymphocytic leukemia. Am J Hematol 2013; 88: Rassenti LZ, Kipps TJ. Lack of allelic exclusion in B cell chronic lymphocytic leukemia. J Exp Med 1997; 185: Klinger M, Zheng J, Elenitoba-Johnson KSJ, Perkins SL, Faham M, Bahler DW. Nextgeneration IgVH sequencing CLL-like monoclonal B-cell lymphocytosis reveals frequent oligoclonality and ongoing hypermutation. Leukemia 2016; 30: Kriangkum J, Motz SN, Mack T, Beiggi S, Baigorri E, Kuppusamy H et al. Single-cell analysis and next-generation immuno-sequencing show that multiple clones persist in patients with chronic lymphocytic leukemia. PLoS One 2015; 10: e Stamatopoulos B, Van Damme M, Crompot E, Dessars B, El Housni H, Mineur P et al. Opposite prognostic significance of cellular and serum circulating microrna-150 in chronic lymphocytic leukemia patients. Mol Med 2015; 21: Stamatopoulos B, Meuleman N, Haibe-Kains B, Saussoy P, Van Den Neste E, Michaux L et al. microrna-29c and microrna-223 down-regulation has in vivo significance in chronic lymphocytic leukemia and improves disease risk stratification. Blood 2009; 113: Dohner H, Stilgenbauer S, Benner A, Leupolt E, Krober A, Bullinger L et al. Genomic aberrations and survival in chronic lymphocytic leukemia. N Engl J Med 2000; 343: Blachly JS, Ruppert AS, Zhao W, Long S, Flynn J, Flinn I et al. Immunoglobulin transcript sequence and somatic hypermutation computation from unselected RNA-seq reads in chronic lymphocytic leukemia. Proc Natl Acad Sci USA 2015; 112: Langerak AW, Davi F, Ghia P, Hadzidimitriou A, Murray F, Potter KN et al. Immunoglobulin sequence analysis and prognostication in CLL: guidelines from the ERIC review board for reliable interpretation of problematic cases. Leukemia 2011; 25: Chen C, Nagy Z, Prak EL, Weigert M. Immunoglobulin heavy chain gene replacement: a mechanism of receptor editing. Immunity 1995; 3: Mimmi S, Vecchio E, Iaccino E, Rossi M, Lupia A, Albano F et al. Evidence of shared epitopic reactivity among independent B-cell clones in chronic lymphocytic leukemia patients. Leukemia; e-pub ahead of print 20 September 2016; doi: /leu Fischer K, Bahlo J, Fink AM, Goede V, Herling CD, Cramer P et al. Long-term remissions after FCR chemoimmunotherapy in previously untreated patients with CLL: updated results of the CLL8 trial. Blood 2016; 127: Herishanu Y, Katz BZ, Lipsky A, Wiestner A. Biology of chronic lymphocytic leukemia in different microenvironments: clinical and therapeutic implications. Hematol Oncol Clin North Am 2013; 27: Burger JA. No cell is an island unto itself: the stromal microenvironment in chronic lymphocytic leukemia. Leuk Res 2007; 31: Supplementary Information accompanies this paper on the Leukemia website ( Macmillan Publishers Limited, part of Springer Nature. Leukemia (2016), 1 9