Clinical significance of TP53, BIRC3, ATM and MAPK-ERK genes in chronic lymphocytic leukaemia: data from the randomised UK LRF CLL4 trial

Despite advances in chronic lymphocytic leukaemia (CLL) treatment, globally chemotherapy remains a central treatment modality, with chemotherapy trials representing an invaluable resource to explore disease-related/genetic features contributing to long-term outcomes. In 499 LRF CLL4 cases, a trial with >12 years follow-up, we employed targeted resequencing of 22 genes, identifying 623 mutations. After background mutation rate correction, 11/22 genes were recurrently mutated at frequencies between 3.6% (NFKBIE) and 24% (SF3B1). Mutations beyond Sanger resolution (<12% VAF) were observed in all genes, with KRAS mutations principally composed of these low VAF variants. Firstly, employing orthogonal approaches to confirm <12% VAF TP53 mutations, we assessed the clinical impact of TP53 clonal architecture. Whilst ≥ 12% VAF TP53mut cases were associated with reduced PFS and OS, we could not demonstrate a difference between <12% VAF TP53 mutations and either wild type or ≥12% VAF TP53mut cases. Secondly, we identified biallelic BIRC3 lesions (mutation and deletion) as an independent marker of inferior PFS and OS. Finally, we observed that mutated MAPK-ERK genes were independent markers of poor OS in multivariate survival analysis. In conclusion, our study supports using targeted resequencing of expanded gene panels to elucidate the prognostic impact of gene mutations.


Introduction
The application of new technologies continues to reveal the biological basis for the clinical heterogeneity apparent within CLL [1][2][3]. In particular, next generation sequencing of large patient cohorts has led to the discovery of These authors contributed equally: Anna Schuh, Jonathan C. Strefford recurring genomic mutations that cluster into distinct biological signalling pathways. Mutations of specific genes including TP53 [4][5][6][7][8][9][10], ATM [9,[11][12][13][14], BIRC3 [9,15,16], SF3B1 [9,[17][18][19][20], NOTCH1 [1,9,15,17,[20][21][22][23], RPS15 [2,24], EGR2 [25,26] and KRAS [27,28] are associated with poorer outcome, especially shorter time to first treatment or overall survival (OS). However, numerous factors influence the clinical significance of a driver mutation in an individual patient. These include clinical status, immunogenetic background, clone size, the presence of biallelic abnormalities and co-existing driver mutations or copy number alterations (CNAs). The clinical importance of these potentially confounding factors is most easily established in context of large clinical trials with long follow-up and where data on numerous biomarkers are available. One such study is the phase III UK LRF CLL4 trial (NCT 58585610) that randomly assigned 777 patients to fludarabine (FDR) or fludarabine plus cyclophosphamide (FC) for six courses, or chlorambucil (CHL) for 12 courses, with the primary endpoint of OS, and secondary endpoints of response rates, progression-free survival, toxic effects and quality of life [29]. The trial demonstrated superior response rates and progression-free survival (PFS) for FC-treated patients compared with those patients treated with FDR or CHL. Previous genomic analysis of this trial has shown TP53 [8], SF3B1 [17], NOTCH1 (coding [17] and non-coding [21]), ATM plus del(11q) [12] and EGR2 [26] lesions to have prognostic significance in multivariate analysis (MVA) and of RPS15 [24] in univariate analysis. The importance of data from CLL4 may be questioned given the studies showing the superior efficacy of FC plus an anti-CD20 antibody (FCR) compared with chemotherapy alone, with the exception of patients with a NOTCH1 mutation [20], and emerging data suggesting the superiority of novel agents compared with chemotherapy-based regimens. However, the observation that TP53, SF3B1 and RPS15 mutations remain poor risk factors in the German CLL8 trial comparing FCR vs. FC [20] and the continuing global need for chemotherapy in CLL for the foreseeable future, indicate that genomic data from the UK CLL4 trial will continue to have clinical relevance.
Accordingly, we performed targeted resequencing on all available pre-treatment samples (n = 499) from the CLL4 trial to investigate the incidence, clinico-biological associations and prognostic impact of a panel of 22 genes recurrently mutated in CLL (study overview in Fig. S1). Important findings include the failure of <12% VAF TP53 mutations (1.97-11.18% variant allele frequency [VAF]) to influence PFS or OS, the importance of 11q deletions on PFS and OS in the context of ATM and BIRC3 mutations, and the reduced OS associated with mutations in the MAPK-ERK genes: BRAF, KRAS and NRAS.

Patients and molecular assays
We studied 499 patient samples taken at randomisation [29]. Patients were diagnosed using the iwCLL guidelines [30], with informed consent obtained in accordance with the declaration of Helsinki. This study was approved by national/ regional research ethics committees. The average lymphocyte percentage of the total white cell count in pre-treatment blood samples was 83.8%. To confirm high tumour load, CD19/ CD5 positivity from cases with available flow-cytometry data were compared with their matched average lymphocyte percentage (n = 233), with an agreement bias of −0.8% (Fig. S2). Our study cohort did not significantly differ from the entire trial cohort in terms of: treatment allocation, CNAs, age, gender, disease stage, ZAP70/CD38 expression, or IGHV status (Table S1). The assessment of established biomarkers was performed as described [31]. All published genetic and biological data on CLL4 patients for genes: TP53 [8], ATM [12,13], BIRC3 [12], NOTCH1 [17] (+3′UTR [21]) and SF3B1 [17], and CNAs: 13q deletion, 17p deletion, 11q deletion and trisomy 12 (5%, 10%, 5% and 3% clone size cut-offs, respectively [31]) were integrated into this study, as well as telomere length [32] and levels of prolymphocytes [33].  (Table S2). Libraries were generated from 250 or 50 ng (dependent on the amount of available starting material) of DNA according to manufacturer's instructions. The average sequencing yield after Illumina processing (MiSeq, paired-end, 2 × 150 bp) from 28 runs was 6.9 Gbp, with a mean read depth of >1000 × (range 502-7948) across all targeted genes, with only nine amplicons below a mean read depth of 1000 (range 502-987) (Fig. S3).
At this depth subclonal mutations can be detected at the 2% level, assuming a minimum observation of four sequencing reads containing the variant base, a Q50 phred like base quality score (p(detected) = 99.999) and a cumulative binomial distribution for n read depth ½ N! n! NÀn ð Þ! p n ð1 À pÞ NÀn . In addition, six variants below 2% were included, since the number of sequencing reads in the variant base were more than ten times the assumed minimum observation (range , and the total read depth exceeded 2000 reads in all cases (range 2582-6389). Bioinformatic data processing of variants was conducted as previously described [14].
All mutations included in this study are listed in Table S3. As the CLL4 cohort lacked germ-line DNA, we only considered variants previously observed as somatically acquired in CLL [1,2,14] or annotated in COSMIC (v70) [34], except for specific circumstances regarding TP53, ATM, BIRC3 and NOTCH1. For TP53, additional mutations annotated in IARC were re-introduced [35]. Pathogenic ATM variants were included if; they were observed in AT families as pathogenic (LOVD [https://databases.lovd.nl/sha red/genes/ATM]), they were evolutionary rare missense [36], or were somatically acquired in CLL [13] (Table S4). However, this variant strategy does not fully preclude ATM variants that exist in germ-line material. For BIRC3, only truncating mutations were included [9]. NOTCH1 PEST domain mutations not predicted to result in protein truncation were removed. All candidate variants were visually inspected in Integrated Genomics Viewer [37]. Genes were defined as recurrent using Tumour Portal (www.tumorporta l.org/power), with the background mutation rate for CLL stated on the website, and the number of cases in the study (n = 499) inputted. Mutations were stratified using Sanger sequencing threshold of 12% [5,9].

Statistical analysis
Fisher's exact tests were performed for co-occurrence analysis between mutated genes and clinical features. PFS and OS were assessed from randomisation using Kaplan Meier (KM) and Log rank analysis. PFS was defined as time from randomisation to progression (i.e. relapse needing further treatment) or death, or to last follow-up date (Oct 2010; final CLL4 PFS update). OS was defined as time from randomisation to death or to last follow-up date for survivors (August 2016, final CLL4 OS update). Multivariate Cox Proportional Hazard models were generated for OS and PFS using backwards selection (P < 0.05), to test the confounding effect of multiple prognostic variables. The Bland-Altman test was used to test agreement between multiple factors, reporting the agreement bias, which is the mean difference between two measurements. All reported P values were two-sided and results were considered significant at the 5% level, using multiple hypothesis testing when appropriate (Benjamini and Hochberg method [38]). Statistical analysis was conducted using R v3.3.0, SPSS v23 (IBM), and Prism v6.0 g (GraphPad).
Next, we assessed the genomic complexity of TP53mut cases. Both <12% VAF and ≥12% VAF TP53mut groups had increased mutation/CNA frequency in comparison to TP53wt cases (both P < 0.001) (Fig. 4b). To further understand the complexity of these two patient subgroups, we inferred the evolutionary history of TP53mut cases as previously described in CLL [2]. Both <12% VAF and ≥12% VAF cases exhibited the same heterogeneous pattern of co-existing mutations, where TP53 mutations were present at higher, or lower VAFS than concomitant driver mutations (Table S8, Figs. 4c and S14).

Biallelic BIRC3 deleted patients infer reduced OS in comparison to sole 11q deleted patients
Although neither ATM nor BIRC3 mutations, regardless of their VAF (Figs. S16 and S17), were associated with reduced PFS or OS in univariate survival analysis (Figs. S12 and S13), it has previously been demonstrated that the impact of these mutations may be dependent on the presence of a concomitant 11q deletion [12,39]. Therefore, we performed an integrated analysis of the clinical impact of ATM and BIRC3 mutations in the context of 11q deleted Fig. 1 Mutation landscape and co-occurrence associations of the CLL4 cohort. a Mutational landscape of CLL4. In the Waterfall plot, known recurrently mutated genes and copy number alterations are shown, hierarchically clustered by mutation frequency (vertical bar chart, right). The mutation burden captured by the study is shown in the bar chart above the heat map. Mutation types are depicted in the above key. The inset vertical bar chart represents the distribution of the number of mutated genes/CNAs per case. b Cooccurrence of all available clinico-biological features from the CLL4 clinical trial. The co-occurrence (red) or mutual exclusivity (green) is plotted per interaction in the graph based on the level of significance (from light to dark: P < 0.05, P < 0.01, Q > P [P < 0.05], Q > P [P < 0.01]).

Discussion
We report targeted resequencing analysis of 22 genes known to be recurrently mutated in CLL in the UK CLL4 clinical trial. CLL4 represents an ideal candidate for such an analysis, with expansive clinical and biological description [8, 12, 13, 17, 21, 24, 26, 29, 31-33, 40, 41] and protracted clinical follow-up. Our study confirms previous studies incorporating samples from this patient cohort showing the impact of TP53ab on PFS and OS in MVA, SF3B1, EGR2 [25,26], RPS15 [1,24] and NFKBIE [25,28,42] mutations on OS in univariate analysis, with SF3B1 and EGR2 mutations retained as independent markers of OS in MVA.
The literature suggests that patients with MAPK-ERK mutations represent a biologically distinct subgroup, where MAPK-ERK mutations are frequently mutually exclusive, are enriched for trisomy 12, unmutated IGHV genes and other adverse biological markers (e.g. CD38, ZAP-70, CD49d), and are linked to inferior time to first treatment in retrospective cohorts [41,[43][44][45]. We now show the MAPK-ERK genes, BRAF, KRAS and NRAS (collectively representing 12.2% of patients) are also independently associated with short OS in a cohort of patients requiring treatment. Vendramini et al. showed a similar frequency of mutations in these genes (14%) [44], while Giménez et al. found that 5.5% of CLL cases harbours functionally deleterious mutations in 11 genes involved in the MAPK-ERK pathway [45], the latter likely reflects the early-stage composition of the cohort. In support of the biological impact of these mutated genes in CLL, (1) Analysis of mutated patients exhibit an enrichment of gene sets associated with transcriptional activation of the MAPK-ERK pathway [44], (2) preliminary in vitro analysis suggests cells from these patients are prone to killing with ERK inhibitors [45], (3) BRAF mutations accelerated disease progression in Eµ-TCL1 mice [46], (4) mutant BRAF has been implicated in venetoclax resistance [47], and (5) KRAS mutated cases associated with poor response to chemoimmunotherapy [27] and lenalidomide [48].
Screening for TP53ab using FISH and Sanger sequencing has known prognostic value [6,8,20,31], and predicts for resistance to chemoimmunotherapy [49]. TP53 mutations that present at low VAFs, below the detection limit of conventional Sanger sequencing may also be positively selected by chemotherapy, and also predict inferior survival, at least in retrospective, institutional cohorts [3,5,9]. The TP53 Network of ERIC provide expansive guidelines on the most suitable approach for TP53 mutational analysis, but also conclude that the clinical importance of low-level TP53 clones remains an unresolved issue, requiring validation in clinical trials [49]. We demonstrated inferior PFS and OS only for those patients with ≥12% VAF TP53 mutations, but we could not demonstrate inferior survival associated with cases harbouring <12% VAF TP53 mutations, the inference perhaps is that these cases represent an intermediate-risk group. Given the unexpected nature of this finding, we also conducted stratified 17p deleted survival analysis, identifying the same result for <12% VAF TP53 mutations without 17p deletion. Furthermore, we proceeded to show that our observation was not associated with any differences in the type of TP53 mutation, their co-existence with other more clonal prognostically-important gene mutations or biological features, nor the enrichment of any specific treatment. As a consequence, we feel that our observation is technically sound, and warrants confirmation in further studies. There remains disagreement regarding the relative clinical significance of deletion and mutation of the BIRC3 and ATM genes, both mapping to the long arm of chromosome 11. The ATM gene is mutated in 30-40% of 11q deleted patients [11,13], where it results in biallelic inactivation of ATM, driving an impaired DNA damage response [50]. The prognostic impact of ATM mutations is controversial in unselected cohorts [9], with the strongest impact when the wild-type allele is lost. In our study, whilst we triaged ATM mutations based on their putative pathogenicity, several are reported in both somatic (i.e. COSMIC) and germline (i.e. dbSNP, EXAC, ClinVar) databases, lending uncertainty to their prognostic impact. The sequencing of matched germline material would provide additional clarity, but was not possible due to the historical nature of CLL4. Preliminary studies support a pathogenetic role of BIRC3 [16,39], more recent studies provide less certainty. For example, in the RESONATE clinical trial [51] and the large retrospective study coordinated by ERIC [52], BIRC3 mutations were not linked to inferior PFS or TTFT, respectively. Another comparator would be the RESONATE2 trial, which compared first line treatment with Ibrutinib vs. chlorambucil [53]. The 24 month PFS for 11q deleted patients in the Ibrutinib arm was 97%. Further studies are required to determine if the long-term outcome of biallelic BIRC3 cases is equally good under modern small molecule inhibition. In our previous CLL4 analysis, we demonstrated that BIRC3 dysfunction (defined as deletion AND/OR mutations of BIRC3) did not impact survival in 11q deleted CLL, while biallelic ATM lesions remained informative [12]. However, this analysis utilised Sanger sequencing, and hence only identified a small number of BIRC3 mutations. Our current study, therefore aimed to expand the analysis with a larger patient cohort with significantly improved technology. This approach permitted the identification of a meaningful number of cases with loss and mutation of BIRC3. As neither ATM nor BIRC3 mutations were linked to survival in univariate analysis, we performed a stratified analysis in 11q deleted cases. In so doing, we show that biallelic BIRC3 cases have a further reduction in survival in comparison to sole 11q deleted cases and were found to be independent prognostic markers for PFS and OS in MVA. Finally, ATM and BIRC3 mutated cases without 11q deletion have a similar survival to wild-type cases.
In conclusion, our study makes three main contributions to the field. We show an expansive analysis of the impact of clinico-biological disease features on the clinical importance of important gene mutations, including SF3B1, EGR2 and the MAPK-ERK genes. Our analysis suggests that <12% VAF TP53 mutations are an intermediate survival group. Finally, we show that biallelic BIRC3 aberrations identify a novel patient subgroup with poor survival, inferior to those with 11q deletions alone. Taken together, we demonstrate that a more expansive genomic screening approach provides additional clinical information, thereby helping to establish the precise importance of genetic alterations in the context of other established and emerging biomarkers. Furthermore, our work will facilitate the development of international standards for the detection and interpretation of somatic mutations in CLL.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons. org/licenses/by/4.0/.