(2014) 28, 129–137 & 2014 Macmillan Publishers Limited All rights reserved 0887-6924/14 www.nature.com/leu

ORIGINAL ARTICLE Monitoring of residual disease by next-generation deep-sequencing of RUNX1 can identify patients with resistant disease

A Kohlmann, N Nadarajah, T Alpermann, V Grossmann, S Schindela, F Dicker, A Roller, W Kern, C Haferlach, S Schnittger and T Haferlach

We studied the utility and clinical relevance of RUNX1 (runt-related 1) mutations and their application as residual disease detection markers using next-generation deep-sequencing. screening was prospectively performed in 814 acute myeloid leukemia patients. At diagnosis, 211/814 (25.9%) patients harbored mutations with a median clone size of 39% (range: 2–96%). Furthermore, in 57 patients paired samples from diagnosis and relapse were analyzed. In 47/57 (82.5%) cases the same alterations detected at diagnosis were present at relapse, whereas in 1/57 (1.8%) cases the mutation from the diagnostic sample was no longer detectable. Discrepancies were observed in 9/57 (15.8%) cases, also including the occurrence of novel RUNX1 mutations not restricted to those regions affected at diagnosis. Moreover, in 103 patients the prognostic impact of residual levels of RUNX1 mutations during complete remission was studied. Separation of patients according to median residual mutation burden into ‘good responders’ and ‘poor responders’ (median: 3.61%; range: 0.03–48.0%) resulted in significant differences of both event-free (median 21.0 vs 5.7 months, Po0.001) and overall survival (OS; median 56.9 vs 32.0 months, P ¼ 0.002). In conclusion, deep-sequencing revealed that RUNX1 mutations qualify as patient-specific markers for individualized disease monitoring. The measurement of mutation load may refine the assignment into distinct risk categories and treatment strategies.

Leukemia (2014) 28, 129–137; doi:10.1038/leu.2013.239 Keywords: RUNX1; mutation analysis; next-generation sequencing

INTRODUCTION example, PML–RARA, RUNX1–RUNX1T1 and CBFB–MYH11,isan 9 Alterations of RUNX1 (runt-related transcription factor 1) constitute established concept. In addition, also some molecular mutations, disease-defining aberrations in acute myeloid leukemia (AML).1 for example, NPM1 mutations, have been shown to be useful 10 Mechanistically, deregulations occur either through balanced targets for quantitative PCR-based MRD detection. However, the translocations, deletions, amplifications or molecular mutations. number of recurrently altered , which are potential targets 4,5,11,12 In particular, RUNX1 mutations were proposed as clinically useful for MRD assessment, as illustrated by mutations in RUNX1, 13,14 15–17 18,19 biomarkers to follow disease progression from myelodysplastic TP53, TET2, or ASXL1, is constantly increasing. syndromes to secondary AML,2 as well as to monitor minimal Moreover, it is expected that translating the wealth of know- residual disease (MRD).3 Moreover, RUNX1 mutations were ledge that is so rapidly being generated by whole-genome and demonstrated to be frequent in de novo AML with non-complex whole-exome sequencing studies into actionable panels will 20–22 karyotype and conferred an unfavorable prognosis,4 explained by have a profound impact on the molecular landscape in AML. an association with resistance to chemotherapy.5 This can already be exemplified by pivotal discoveries such as 23 24,25 26 The importance of detecting small subclones in hematological IDH1, DNMT3A, BCOR or cohesin complex (STAG2, SMC3, 27 malignancies has increased. This has become highly relevant RAD21 and SMC1A) gene mutations. As such, there is a great either in a setting of assessing MRD or identifying molecular unmet need for unbiased standardized objective methodologies mutations with prognostic or predictive relevance to direct to provide information on molecular alterations not only per se but treatment strategies.6 Indeed, applications of MRD monitoring also at a level of sensitivity and throughput necessary for have been proposed to include early assessment of response to diagnostics processes.28 therapy to improve risk stratification and to guide post-remission In principle, it seems that RUNX1 mutations qualify as a marker therapy. Further, post-treatment monitoring would be suited to for MRD detection, as RUNX1 is frequently mutated in AML. Overall detect impending relapse and guide pre-emptive therapy frequencies of RUNX1 mutations were reported to range from strategies.7 Particularly in AML, risk-adapted strategies can be 5.6 to 32.7% of AML patients, depending on the age group of established based on MRD data on an individual basis for each the patients and chromosomal aberrations.4,5,11,29 Therefore, here patient.8 we investigated for the first time whether amplicon Today, PCR techniques and multiparameter flow cytometry deep-sequencing has the potential to become such a have become standard methods to investigate MRD in AML.8 In standardized routine diagnostic methodology, enabling particular, the monitoring of MRD kinetics using quantitative PCR unprecedented levels of sequencing sensitivity and allowing an assays for the detection of leukemia-specific fusion genes, for individualized monitoring of disease progression and treatment

MLL Munich Leukemia Laboratory, Munich, Germany. Correspondence: Dr A Kohlmann, MLL Munich Leukemia Laboratory, Max-Lebsche-Platz 31, 81377 Munich, Germany. E-mail: [email protected] Received 31 July 2013; accepted 12 August 2013; accepted article preview online 20 August 2013; advance online publication, 17 September 2013 Residual disease in AML using NGS A Kohlmann et al 130 efficacy. However, in contrast to other candidate genes involved in Data analysis and detection of variants (ref. 10) (ref. 23) myeloid malignancies, such as NPM1 or IDH1 ,no Image and data processings were performed using default amplicon mutational hotspots are present in RUNX1. As such, offering pipeline settings of the GS FLX Sequencer Instrument software version 2.3 patient-specific quantitative real-time PCR assays to detect the or higher (Roche Applied Science). Subsequent sequence alignment and broad spectrum of mutations would require the need to apply variant detection was performed using default settings of the GS Amplicon individualized assays, which is not feasible on a per-patient basis Variant Analyzer software version 2.5.3 (Roche Applied Science). The results during routine diagnostics procedures. were further processed and visualized in R/Bioconductor using the package R453Plus1Toolbox (version 1.0.1) and the Sequence Pilot software Thus far, next-generation sequencing (NGS) has demonstrated version 3.4.0 (JSI Medical Systems, Kippenheim, Germany).34 For the its potential as a diagnostic platform and amplicon deep- detection of variants, filters were set to display sequence variants occurring sequencing assays have been envisioned to be applied in 41% of bidirectional reads per amplicon in at least one patient. The cut- 28,30–32 clinically. In the scenario of RUNX1 mutation analyses, off was set after sequencing a control cohort of 10 healthy individuals as seven amplicons are sufficient to test the complete coding region previously reported.35 Depending on the actual type of the mutation, that and to offer unprecedented levels of sensitivity in sequencing of is, single-nucleotide changes versus more complicated variants, higher 35 genomic alterations.33 Here we studied the use of amplicon deep- levels of sensitivity can be reached. Single-nucleotide polymorphisms sequencing in a routine diagnostics environment to characterize according to dbSNP build 131 or higher and alterations within introns with the spectrum of RUNX1 mutations in 814 cases with AML and to the exception of splice-site mutations, which were both validated on cDNA level for their effect on splicing, were not scored.36 assess the stability of RUNX1 mutations in 57 paired samples from An additional functional interpretation of the variants detected in RUNX1 diagnosis and relapse. Moreover, this deep-sequencing assay was performed using MutationTaster (www.mutationtaster.org),37 was applied to monitor the composition and mutation load PolyPhen (http://genetics.bwh.harvard.edu/pph2/index.shtml)38 and SIFT kinetics of RUNX1 mutations, following in detail 103 patients (http://sift.jcvi.org/) algorithms.39 during their course of disease, and the prognostic impact is presented allowing to identify AML patients responding poorly to Serial NGS analyses using deep-sequencing chemotherapy. For longitudinal studies, two additional cohorts were investigated. The overlap of cases from these cohorts with the cohort of 814 prospectively collected cases is shown in Supplementary Figure S1. A first cohort included 57 known RUNX1-mutated AML cases of whom MATERIALS AND METHODS samples were also available for a serial deep-sequencing study at relapse Patient cohorts stage. This cohort included 114 specimens in total, collected from We here report on prospective deep-sequencing molecular analyses in 814 November 2005 to May 2012. The median sampling interval from untreated AML cases at diagnosis, including 645 de novo AML, 109 diagnosis to relapse for these cases was 360 days (range: 56–1578 days). secondary AML and 60 therapy-related AML cases. The median age of the Of note, nine cases had previously been reported.4 patients was 69.6 years (range: 0.4–93.1 years), including 375 female and In a second independent cohort of 103 AML cases harboring RUNX1 439 male patients, respectively. According to cytogenetic data, 50.5% mutations, consecutive specimens were analyzed to monitor response to (411/814) of cases presented a normal karyotype, 38.8% (316/814) therapy (Supplementary Figure S1). This cohort included 206 specimens in harbored non-complex cytogenetic alterations, 9.6% (78/814) carried a total, collected from August 2005 to May 2012. In these consecutive complex aberrant karyotype and 1.1% (9/814) of patients were character- samples, the alterations detected at diagnosis were subsequently ized by favorable cytogenetics. Samples were sent to the MLL Munich investigated with high coverage (median depth: 844-fold) during the Leukemia Laboratory for diagnostic assessment during the period of July course of the disease. For the NGS analyses at the time of first follow-up, 2010 to April 2012 (Supplementary Figure S1). The study design adhered to the median sampling interval was 128 days, ranging from 60 to 198 days. the tenets of the Declaration of Helsinki and was approved by the All patients were treated with AML-specific intensive treatment protocols institutional Review Board before its initiation. comprising one or two courses of induction therapy with standard-dose or high-dose cytarabine and an anthracycline, as well as at least one course of consolidation therapy of identical intensity. RUNX1 mutational analysis and amplicon generation For molecular analyses, mononuclear bone marrow or peripheral blood Statistical analyses cells were enriched using a standard Ficoll density-gradient procedure. The Overall survival (OS) was defined as the time from diagnosis of AML to complete coding region of RUNX1 was amplified using the FastStart High death or last follow-up and event-free survival (EFS) as the time from Fidelity PCR System and GC-RICH PCR System kits (Roche Applied Science, diagnosis of AML to treatment failure, relapse, death or last follow-up. To Penzberg, Germany). The sequencing assay was developed using an eliminate the effect of allogeneic stem cell transplantation, OS was independent cohort to homogenously yield a sufficient coverage across all recalculated censoring patients at the day of transplantation (OSTXcens). amplicons and molecular barcodes used (Supplementary Figure S2).33 Each Survival curves were calculated according to Kaplan–Meier method and PCR product was individually purified using Agencourt AMPure XP beads were compared using the two-sided log-rank test. For all analyses, results (Beckman Coulter, Krefeld, Germany) and sequencing amplicon libraries were significant at a level of Po0.05 at both sides. Functional associations were quantified using the Quant-iT PicoGreen dsDNA kit (Invitrogen, were calculated using Fisher’s exact test. SPSS software version 19.0.1 (IBM Carlsbad, CA, USA). Information on primer sequences and the respective Corporation, Armonk, NY, USA) was used for statistical analyses. PCR cycler protocol are given online (Supplementary Tables S1–S3). Further information on assay development is provided online. RESULTS Landscape of molecular RUNX1 mutations in AML Next-generation amplicon deep-sequencing Overall, in this AML cohort including 814 cases, 211 patients The use of GS GType RUNX1 Primer Set 96-well plates for amplicon (25.9%) were detected to carry RUNX1 mutations. At diagnosis, the generation allowed multiplexing up to 12 patients per sequencing lane on median clone size was 39% and ranged from 2 to 96%. In all, a Genome Sequencer FLX System or GS Junior instrument (Roche Applied 73.9% (156/211) of mutated patients carried one, whereas 26.1% Science). Amplicon pools were prepared for the emulsion PCR step and (55/211) of cases harbored two (n ¼ 46) or more (n ¼ 9) RUNX1 final sequencing libraries were processed using the GS FLX Titanium Series mutations. Lib-A SV method (Roche Applied Science). Forward (A beads) and reverse (B beads) emulsion PCR reactions were carried out separately using We next restricted an analysis to the subgroup of AML with 2 000 000 capture beads per small volume emulsion oil tube. The normal karyotype. In this subgroup of 411 cases, a total of 114 amplification reaction, breaking of the emulsions and enrichment of patients (27.7%) harbored RUNX1 mutations. In the subgroup beads carrying amplified DNA was performed using the workflow as of RUNX1-mutated patients, full molecular mutation status recommended by the manufacturer and as previously published.33 characterization according to four additional molecular markers,

Leukemia (2014) 129 – 137 & 2014 Macmillan Publishers Limited Residual disease in AML using NGS A Kohlmann et al 131 that is, FLT3–ITD, NPM1, MLL–PTD and CEBPA was available for 110 second, independent clone harbored a p.Gly367Profs*203 altera- patients (110/114; 96.5%). In 66 (60.0%) of these 110 patients, tion in 11.55% of the sequencing reads. RUNX1 was detected to be mutated in a background of molecular As indicated in Figure 3, RUNX1 mutational analyses using wild-type status for FLT3–ITD, NPM1, MLL–PTD and CEBPA, whereas quantitative read counts allowed to closely monitor the composi- in 44/110 (40.0%) patients RUNX1 mutations were detected tion of clones. For example, in the first patient the p.Gln235* concomitantly with mutations in FLT3–ITD (n ¼ 27/44), MLL–PTD mutation was decreasing from 41% to only 30%. After transplan- (n ¼ 14/44), CEBPA (n ¼ 11/44) or NPM1 (n ¼ 1/44), respectively. As tation, this mutation was no longer detectable. In a second such, in AML with a normal karyotype, a high proportion of patient, two alterations with differing mutation load were patients negative for FLT3–ITD, NPM1, MLL–PTD and CEBPA are detectable at diagnosis. Although the p.Leu294Serfs*7 mutation likely to carry a RUNX1 mutation and, thus, RUNX1 mutation disappeared during therapy, the p.Arg444Profs*128 mutation screening in these patients for the first time allows a better increased. This mutation then was utilized for monitoring the prognostication and provides a potential marker for residual leukemic clone burden after transplantation. A third patient disease monitoring. still presented with detectable resistant clones (p.Arg139_Ser140- In detail, the 211 AML patients harbored a total number of 275 insPro), both before and after transplantation. Yet, after mutations in RUNX1. The 275 variants were distributed as follows: changing to a Sorafenib-based treatment scheme the initial 41.1% (113/275) frame-shift mutations, 34.6% (95/275) missense, clone no longer was detectable. As such, the deep- 13.8% (38/275) nonsense, 6.9% (19/275) exon-skipping/splicing, sequencing successfully allowed profiling patients before and 3.3% (9/275) in-frame insertion/deletion alterations and 0.4% after transplantation. (1/275) no stop changes. Recurrently altered codons were Ser114, Arg135, Gly138, Arg139, Asp171, Arg174 and Arg293, respectively (Figure 1). Serial NGS analyses and stability of RUNX1 mutations at relapse With respect to the functional domains, 213 of the 275 We first investigated the stability of individual RUNX1 mutations mutations were clustered either in the Runt homology domain between diagnosis and relapse. In 57 cases with RUNX1 mutations, (n ¼ 160, 58.2%) or transactivation domain (n ¼ 53, 19.3%). There the complete coding region of RUNX1 was studied at diagnosis was a positive association of missense mutations and their and at the time of relapsed disease by deep-sequencing. As location in the Runt homology domain (P 0.001) and, in addition, o demonstrated in Table 1, in 47/57 (82.5%) cases the same between frame-shift mutations and their location in the transacti- alterations detected at diagnosis were measurable at the time of vation domain (P 0.001). We also observed a positive association o relapse. In 1/57 (1.8%) cases, a p.Asp171Asn RUNX1 mutation was between nonsense mutations and the transactivation domain no longer detectable at relapse. Moreover, there were additional (P ¼ 0.007). Finally, we identified a positive association of 9/57 (15.8%) patients in whom discrepancies were observed. In frame-shift mutations and their location outside functional domains detail, in one patient with RUNX1 double mutations one of the two (P ¼ 0.04). We applied three different prediction tools (Mutation- mutations detected at diagnosis remained, whereas the other one Taster, PolyPhen2 and Sift) to determine the impact of single amino was lost. In another patient, the mutation detected at diagnosis acid substitutions on the RUNX1 structure and function. All was lost; however, a novel mutation was detected elsewhere in missense mutations were predicted to be damaging by at least two the same exon, thus targeted by the same deep-sequencing bioinformatics tools (Supplementary Data 1). amplicon. In three patients, the mutation detected at diagnosis was lost; however, a novel mutation was detected at a different Detection of RUNX1 mutations using deep-sequencing region of the gene and targeted by a different amplicon. Of note, As demonstrated in Figure 2, amplicon deep-sequencing enabled in one of these three patients two novel mutations did occur. the robust detection of molecular RUNX1 mutations. In this Finally, in four patients the mutation detected at diagnosis exemplary patient, 909 individual amplicon reads were sequenced remained stable, but additional mutations occurred in other in a massively parallel way as part of a single sequencing run. In regions of the gene. As such, the recommended screening of comparison with the reference genome sequence, two indepen- RUNX1 would require to be expanded to the complete coding dent clones were detected in this patient. The p.Ala382Profs*189 region at relapse and not necessarily be restricted to those regions alteration was present with 32.34% of mutated reads and a affected by RUNX1 mutations at diagnosis.

Figure 1. Landscape of RUNX1 mutations in AML. Two hundred and seventy-five RUNX1 mutations from 211 AML patients and their respective locations are depicted. Multiple identical mutations are given in circles. The exon nomenclature is based on Ensemble Transcript ID ENST00000344691. The nature of the alteration is color-coded. Variants affecting splicing of the mRNA were verified by sequencing of correlating cDNA products. Information on genomic DNA changes, mutation load and sequencing coverage is provided online (Supplementary Data 1). RHD, Runt homology domain; TAD, transactivation domain.

& 2014 Macmillan Publishers Limited Leukemia (2014) 129 – 137 Residual disease in AML using NGS A Kohlmann et al 132 • Clone #1: #1 909-fold coverage 32.34% mutated #2 c.1144_1150delGCCTCGGinsCC p.Ala382Profs*189

• Clone #2:

909-fold coverage 11.55% mutated c.1098_1106delAGGCCCGTTinsG p.Gly367Profs*203

Figure 2. Detection principle of RUNX1 mutations using deep-sequencing. In the top panel, an amplicon representing codons 297–386 of the RUNX1 gene is represented. The blue line indicates an absolute read count of 909 individual sequences (500 forward/409 reverse) for this sample. In comparison with the reference sequence, two molecular insertion/deletion (indels) events are detectable. The c.1144_1150delGCCTCGGinsCC alteration was found in 32.34% (294/909) of the reads and a c.1098_1106delAGGCCCGTTinsG was present in 11.55% (104/909) of the reads, respectively (bottom panel).

a

p.Gln235*

b

p.Leu294Serfs*7 p.Arg444Profs*128

c

p.Arg139_Ser140insPro

Figure 3. RUNX1 mutations and monitoring of allogeneic stem cell transplantation success rates in three exemplary patients. (a, b, c) Three patients are depicted carrying distinct RUNX1 alterations in a longitudinal analysis. All three patients underwent transplantation and the mutational analysis was used to molecularly monitor the decrease of the mutation load during therapy and after transplantation. On the right y-axis, the number of reads is indicated (grey bar graphs); on the left y-axis, the mutation load, that is, the percentage of reads harboring the mutation (red bar graphs) is given, respectively.

Serial NGS analyses and prognostic implication of residual RUNX1 patient-specific alterations that had been detected at the time mutations point of diagnosis were investigated with high sequencing We next focused on a series of 103 RUNX1-mutated AML cases coverage in these specimens. The median sequencing coverage with clinical data available and treated with standard curative for assessing this residual disease status was 844-fold, translating AML-specific regimens. In these cases with detectable RUNX1 into a sensitivity of B1:800. The samples analyzed in these follow- mutations at diagnosis, follow-up specimens were collected and up analyses were collected between 60 and 198 days (median: 128 also investigated using deep-sequencing. In particular, the days) after diagnosis.

Leukemia (2014) 129 – 137 & 2014 Macmillan Publishers Limited Residual disease in AML using NGS A Kohlmann et al 133 Table 1. Stability of RUNX1 mutations during the course of disease in 57 patients

Mutation lost at relapse Identical mutation at relapse One of two mutations Novel mutation lost at relapse

RUNX1 mutations from diagnosis (patients) 1 47 1a 1b,3c,4d Abbreviation: RUNX1, runt-related transcription factor 1. aPatient with RUNX1 double mutations and one of the two mutations detected at diagnosis remained, whereas the other one was lost. bMutation detected at diagnosis was lost; however, a novel mutation was detected elsewhere in the same exon, thus successfully targeted by the same deep-sequencing amplicon. cIn three patients, the mutation detected at diagnosis was lost; however, a novel mutation was detected at a different region of the gene and targeted by a different amplicon. Of note, in one of these three patients, two novel mutations did occur. dIn four patients, the mutation detected at diagnosis remained stable, but additional mutations occurred in other regions of the gene.

04/2008 35.3% Mutation at diagnosis Exon 08 (amplicon 1) Codons 334 -357 1,130-fold coverage

c.1026_1029dupGGCC // p.Met344Glyfs*230

08/2008 Mutation at 1st follow-up Exon 08 (amplicon 1) 5.4% Codons 334 -357 857-fold coverage

Figure 4. RUNX1 mutations in serial samples. In the top panel, an amplicon part representing codons 334–357 of the RUNX1 gene is presented. The blue line indicates an absolute read count of 1130 individual sequences (573 forward/557 reverse). In comparison with the reference sequence, one duplication was detectable. The c.1026_1029dupGGCC (p.Met344Glyfs*230) mutation was present in 35.3% (399/1130) of the reads. At the next consecutive follow-up analysis of this patient with 857-fold coverage, the same mutation is still detectable with a mutation load of 5.4% (46/857) of the reads (lower panel).

In this cohort, principally two distinct categories of patients to their median level of mutational burden. Cases o3.61% were apparent. In a first category, 48.5% (50/103) of patients mutation load were considered as ‘good responders’ (n ¼ 76), responded well to therapy and were characterized by a total whereas cases above the median were considered as ‘poor clearance of the mutated clone at this time point of follow-up responders’ (n ¼ 27) (Figure 5). A survival analysis (Figure 6) (0 mutated reads observed). Intriguingly, also in cases with two or demonstrated that this threshold resulted in significant differ- more alterations, these concomitant mutations all had responded ences in EFS (median 21.0 vs 5.7 months, Po0.001), OS (median in a similar manner. An additional 15.5% (16/103) of cases had 56.9 vs 32.0 months, P ¼ 0.002) and OSTXcens (median 37.1 vs 12.2 positive read counts that were around or below the deep- months, P ¼ 0.022). sequencing assay’s limit of detection that was set at 1.0%. In a second approach, we created a model that also takes the A second group comprised 25.2% (26/103) of patients. As mutation burden irrespective of the median mutation load into exemplarily demonstrated in Figure 4, in this group of patients account. As such, patients with low mutation burden of 0.03– mutations detected at diagnosis were still detectable using deep- 2.72% were included in this analysis if they had specific insertion/ sequencing in the first follow-up analysis. In this subset of cases, duplication events or harbored missense mutations in known none of these patients demonstrated a RUNX1 mutation clone size hotspot regions. The number of ‘good responders’ thus decreased reduction to a level below 4.22% of reads at the time point of to 57 cases and 46 patients were now considered as ‘poor follow-up analyses. In some instances, the burden of the RUNX1 responders’. This approach again showed differences in EFS mutation load had even increased. (median 21.3 vs 7.8 months, P ¼ 0.001), OS (median 56.9 vs 35.8 In six cases, the deep-sequencing assay clearly indicated a months, P ¼ 0.062) and OSTXcens (median not reached vs 37.1 molecularly confirmed residual disease, even though the mutation months, P ¼ 0.049; Figures 7a–c). burden was ranging only from 0.03 to 1.3%. This judgment was possible because of the specific nature of the alterations (for example, c.458_459insGGGGAGGA or c.1261_1268dupAACCA DISCUSSION GAG), in which even a very low read count would indicate a true In this study we investigated RUNX1 mutations, a pathobiologically mutation event. important molecular aberration in myelodysplastic syndromes and We next studied the prognostic implication of residual RUNX1 AML using next-generation amplicon deep-sequencing.1,40 mutations in more detail. Overall, in the 103 patients investigated A prospective routine diagnostics cohort of 814 AML cases at the median percentage of mutated reads was reduced to a level of diagnosis was analyzed over the course of 22 months. only 3.61% mutated reads at follow-up stage. Regarding assess- We demonstrated that amplicon-based NGS is a suitable ment of residual disease, patients were thus separated according method to accurately detect and quantify the variety of RUNX1

& 2014 Macmillan Publishers Limited Leukemia (2014) 129 – 137 Residual disease in AML using NGS A Kohlmann et al 134

Figure 5. RUNX1 mutations and residual mutation load at first follow-up. The panel depicts the mutation load of residual RUNX1 mutations at first follow-up across 103 samples. Samples are ordered in ascending orientation. In median for the 50 cases with detectable residual mutations, the load was 3.61%. Patients were grouped into ‘good responders’ and ‘poor responders’ based on this median residual mutation load as a threshold.

p<0.001 p=0.002

good responders good responders (n=76; median (n=76; median 21.0 months) 56.9 months)

poor responders (n=27; median poor responders 32.0 months) (n=27; median 5.7 months)

p=0.022

good responders (n=76; median 37.1 months)

poor responders (n=27; median 12.2 months)

Figure 6. Kaplan–Meier survival estimates according to median residual RUNX1 mutation load mutations of 3.61% at first follow-up. According to the median level of residual detectable RUNX1 mutations (3.61%; range 0.03–48.0%), cases were separated into ‘good responders’ (n ¼ 76) with RUNX1 mutation levels below 3.61% and ‘poor responders’ (n ¼ 27) with mutation levels starting at 3.61% at first follow-up. This resulted in significant differences in (a) EFS (median 21.0 vs 5.7 months, Po0.001), (b) OS (median 56.9 vs 32.0 months, P ¼ 0.002) and (c)OSTXcens, recalculated OS censoring patients at the day of transplantation (median 37.1 vs 12.2 months, P ¼ 0.022).

aberrations with high sensitivity and, therefore, robustly setting genes, has potentially increased the number of cases suitable for the baseline for an individualized monitoring of disease PCR-based MRD monitoring to 60–70%. Here, for the first time, we progression and treatment efficacy. addressed the clinically important question as to whether RUNX1 As recently summarized by Buccisano et al.,8 the discovery of mutations would represent useful markers to be added to a molecular mutations in AML cases, otherwise negative for fusion standard routine screening panel, including NPM1, FLT3, CEBPA

Leukemia (2014) 129 – 137 & 2014 Macmillan Publishers Limited Residual disease in AML using NGS A Kohlmann et al 135

p=0.001 p=0.062

good responders (n=57; median 56.9 months)

good responders (n=57; median 21.3 months)

poor responders (n=46; median 35.8 months)

poor responders (n=46; median 7.8 months)

p=0.049

good responders (n=57; median not reached)

poor responders (n=46; median 37.1 months)

Figure 7. Kaplan–Meier survival estimates according to very low residual RUNX1 mutation load at first follow-up. According to the residual level of detectable RUNX1 mutations (range 0.03–48.0%), cases were separated into ‘good responders’ (n ¼ 57) and ‘poor responders’ (n ¼ 46) with mutation levels starting as low as 0.03% at first follow-up. This resulted in significant differences in (a) EFS (median 21.3 vs 7.8 months, P ¼ 0.001), (b) OS (median 56.9 vs 35.8 months, P ¼ 0.062) and (c)OSTXcens (median not reached vs 37.1 months, P ¼ 0.049). and MLL–PTD. In our cohort of 814 AML patients, 411 (50.5%) detected at diagnosis were also measurable at the time of relapse. cases had a normal karyotype and a total of 114 (27.7%) patients However, discrepancies were detected in 15.8% of patients, in this specific subgroup harbored RUNX1 mutations. A full including, in some instances, also novel mutations gained in molecular characterization according to four additional markers, regions different from the location of the mutation at diagnosis. that is, FLT3, NPM1, MLL–PTD and CEBPA mutation status was Therefore, this data not only strengthen the utility of RUNX1 being available for 110 patients (110/114; 96.5%). It is therefore a suitable stable marker to assess residual disease but also important to note that 60.0% (66/110) of these comprehensively underline the need for the screening of RUNX1 being expanded to characterized cases would have had a wild-type status with the complete coding region at relapse, and not necessarily be respect to the four established genes NPM1, FLT3, CEBPA and restricted to those regions affected by RUNX1 mutations at MLL–PTD. As such, adding RUNX1 mutation screening to the diagnosis. standard workup of AML will not only advance the molecular In AML, RUNX1 mutations have particularly gained interest, as classification and prognostication of these patients,41 it might also they were demonstrated to have independent poor prognostic offer a novel opportunity of guiding the choice of post-remission relevance for OS. As reported by Tang et al.,11 the presence of a therapy, driven by the actual risk of relapse in cases with RUNX1 mutation was associated with inferior response in the 330 detectable RUNX1 mutations at the first follow-up analysis. adult de novo non-M3 AML patients examined, who were Along these lines, we were also interested in addressing the undergoing conventional induction chemotherapy. In both the question whether in cases with full relapse RUNX1 mutations total cohort and in a subset of patients with a normal karyotype, would be stable over the course of disease. As reported by Tang patients with RUNX1 mutations had significant poorer OS.11 et al.11 in their sequential analyses of six patients with distinct Further, their sequential study during the clinical course for 133 RUNX1 mutations and available samples for serial assessment, two patients showed that none of these patients acquired novel out of six patients regained the same mutations as those detected RUNX1 mutation at relapse, indicating that the mutation may have at diagnosis and another two out of six lost the mutation at a small role in disease progression.11 In the study by Schnittger relapse. Data from Schnittger et al.4 on 10 paired diagnosis and et al.,4 among 280 AML patients those carrying RUNX1 mutations relapse samples indicated a high stability of RUNX1 mutations had shorter OS and EFS compared with RUNX1 wild-type cases. during the course of the disease. Moreover, RUNX1 mutations had a strong adverse prognostic Here we included data on 57 patients with serial samples effect in AML with normal karyotype or non-complex available. The complete coding region of RUNX1 was studied chromosomal imbalances, especially in those that did not carry both at diagnosis and at the time of relapsed disease by CEBPA, NPM1, FLT3–ITD or MLL–PTD.4 Data from Gaidzik et al.5 deep-sequencing. Underlining the high stability of RUNX1 muta- underlined that RUNX1 mutations are characterized by distinct tions, in a high proportion of patients the same alterations biological and clinical features. In their cohort including 878

& 2014 Macmillan Publishers Limited Leukemia (2014) 129 – 137 Residual disease in AML using NGS A Kohlmann et al 136 non-APL AML cases, RUNX1 mutations were significantly risk stratification. It will be critical to validate this finding in an associated with resistance to induction chemotherapy. Lastly, independent cohort of AML patients so that recommendations on Mendler et al.29 reported on a poor outcome of RUNX1-mutated adapted dose schedules and/or transplantation procedures could primary AML with normal karyotype in both younger and older be given for RUNX1-mutated patients with detectable residual patients treated with intensive induction chemotherapy and not disease status. receiving allogeneic stem cell transplantation in first complete remission. A strong consideration of upfront novel therapies and/ or early allogeneic stem cell transplantation had been CONFLICT OF INTEREST 29 suggested. RUNX1 oligonucleotide primer plates were provided as part of the IRON-II study by Here we studied 103 cases with RUNX1 mutations at diagnosis Roche Diagnostics GmbH, Penzberg, Germany. WK, CH, SuS and TH are part-owners and also investigated the first available time point after induction of the MLL Munich Leukemia Laboratory GmbH. AK, NN, VG, TA, SoS, FD and AR are therapy by deep-sequencing in consecutive samples sent to our employed by MLL Munich Leukemia Laboratory. AK has received honoraria from laboratory for molecular analyses. In median, this sensitive Roche Diagnostics. sequencing assay was performed 128 days after the start of intensive chemotherapy treatment. Importantly, it was possible to distinguish between distinct categories of patients, that is, those that became ‘residual disease negative’ as assessed by AUTHOR CONTRIBUTIONS deep-sequencing read counts (‘good responder’) and cases with AK designed the study and wrote the manuscript. AK, VG and FD detectable residual RUNX1 mutation load (‘poor responder’). We performed research and generated data. AK, NN, TA, SoS, AR and hypothesized that in general, a better outcome would be WK analyzed and interpreted the data. WK, CH, SuS and TH observable for those cases that reached ‘deep-sequencing residual performed diagnostic testing on all patient samples. disease negativity’ at first follow-up. Indeed, an approach that segregated the cohort according to the median residual load revealed significant differences in both EFS (median 21.0 vs 5.7 REFERENCES months, Po0.001), OS (median 56.9 vs 32.0 months, P ¼ 0.002) 1 Osato M, Asou N, Abdalla E, Hoshino K, Yamasaki H, Okubo T et al. Biallelic and and a recalculated OS censoring patients at the day of heterozygous point mutations in the runt domain of the AML1/PEBP2alphaB gene transplantation OSTXcens (median 37.1 vs 12.2 months, P ¼ 0.022). associated with myeloblastic . Blood 1999; 93: 1817–1824. However, this first approach was next improved in a sense that 2 Dicker F, Haferlach C, Sundermann J, Wendland N, Weiss T, Kern W et al. Mutation analysis for RUNX1, MLL-PTD, FLT3-ITD, NPM1 and NRAS in 269 patients with MDS the initial model lacked biological knowledge about both the or secondary AML. Leukemia 2010; 24: 1528–1532. specific location of the residual mutation and the precise 3 Kohlmann A, Grossmann V, Harbich S, Dicker F, Alpermann T, Nadarajah N et al. molecular type of alteration. As such, a second model was tested, Monitoring of minimal residual disease using next-generation deep-sequencing which also integrated mutations with very low mutation burden in 460 acute myeloid leukemia cases identifies RUNX1 mutated patients with (0.03–2.72%), irrespective of the median mutation load in the total resistant disease. Blood 2011; 118: 747a. cohort. This approach again showed differences in EFS (median 4 Schnittger S, Dicker F, Kern W, Wendland N, Sundermann J, Alpermann T et al. 21.3 vs 7.8 months, P ¼ 0.001), OS (median 56.9 vs 35.8 months, RUNX1 mutations are frequent in de novo AML with noncomplex karyotype and P ¼ 0.062) and OSTXcens (median not reached vs 37.1 months, confer an unfavorable prognosis. Blood 2011; 117: 2348–2357. P 0.049). Taken together, this data indicated that RUNX1 5 Gaidzik VI, Bullinger L, Schlenk RF, Zimmermann AS, Rock J, Paschka P et al. ¼ RUNX1 mutations in acute myeloid leukemia: results from a comprehensive mutations would qualify as patient-specific markers for individua- genetic and clinical analysis from the AML study group. J Clin Oncol 2011; 29: lized disease monitoring. 1364–1372. In general, adjusted risk stratification based on pretreatment 6 Grimwade D, Vyas P, Freeman S. Assessment of minimal residual disease in acute cytogenetics and molecular genetics characteristics, as well as myeloid leukemia. Curr Opin Oncol 2010; 22: 656–663. MRD status at the end of consolidation, has the potential to refine 7 Dohner H, Estey EH, Amadori S, Appelbaum FR, Buchner T, Burnett AK et al. risk-classification schemes. Missing early identification of high-risk Diagnosis and management of acute myeloid leukemia in adults: recommenda- patients might hamper or delay the timely use of intensive tions from an international expert panel, on behalf of the European LeukemiaNet. treatments, such as allogeneic stem cell transplantation.42 As Blood 2010; 115: 453–474. reported by Gaidzik et al.,5 allogeneic hematopoietic stem cell 8 Buccisano F, Maurillo L, Del Principe MI, Del PG, Sconocchia G, Lo-Coco F et al. Prognostic and therapeutic implications of minimal residual disease detection in transplantation had a favorable impact on outcome in RUNX1- acute myeloid leukemia. Blood 2012; 119: 332–341. mutated AML. In fact, all RUNX1-mutated patients relapsed after 9 Ommen HB, Schnittger S, Jovanovic JV, Ommen IB, Hasle H, Ostergaard M et al. conventional consolidation therapy, including repetitive cycles of Strikingly different molecular relapse kinetics in NPM1c, PML-RARA, high-dose cytarabine and autologous hematopoietic stem cell RUNX1-RUNX1T1, and CBFB-MYH11 acute myeloid leukemias. Blood 2010; 115: transplantation, whereas the 4-year relapse-free survival after 198–205. allogeneic hematopoietic stem cell transplantation was 52%.5 10 Schnittger S, Kern W, Tschulik C, Weiss T, Dicker F, Falini B et al. Minimal residual Therefore, future studies will be necessary to address the question disease levels assessed by NPM1 mutation-specific RQ-PCR provide important whether an adjusted risk stratification might take into account prognostic information in AML. Blood 2009; 114: 2220–2231. that RUNX1-mutated patients might be selected for allogeneic 11 Tang JL, Hou HA, Chen CY, Liu CY, Chou WC, Tseng MH et al. AML1/RUNX1 mutations in 470 adult patients with de novo acute myeloid leukemia: prognostic hematopoietic stem cell transplantation due to ‘residual disease implication and interaction with other gene alterations. Blood 2009; 114: positivity’ based on the read count of residual clones at the time 5352–5361. point of first follow-up. 12 Roumier C, Eclache V, Imbert M, Davi F, MacIntyre E, Garand R et al. M0 AML, In conclusion, NGS accurately detects and quantifies RUNX1 clinical and biologic features of the disease, including AML1 gene mutations: a mutations in AML with high sensitivity. It has been postulated that report of 59 cases by the Groupe Francais d’Hematologie Cellulaire (GFHC) and whatever the method utilized, measurement of MRD will the Groupe Francais de Cytogenetique Hematologique (GFCH). Blood 2003; 101: potentially contribute to refining assignment to risk categories 1277–1283. in AML.8 The technique of deep-sequencing was observed to be 13 Haferlach C, Dicker F, Herholz H, Schnittger S, Kern W, Haferlach T. Mutations of superior to capillary routine sequencing methods and will advance the TP53 gene in acute myeloid leukemia are strongly associated with a complex aberrant karyotype. Leukemia 2008; 22: 1539–1541. the biological characterization of leukemias, in particular 14 Rucker FG, Schlenk RF, Bullinger L, Kayser S, Teleanu V, Kett H et al. TP53 during follow-up analyses and in detecting residual disease. alterations in acute myeloid leukemia with complex karyotype correlate with Amplicon deep-sequencing thus has the potential to enable an specific copy number alterations, monosomal karyotype, and dismal outcome. individualized monitoring of disease progression and to improve Blood 2012; 119: 2114–2121.

Leukemia (2014) 129 – 137 & 2014 Macmillan Publishers Limited Residual disease in AML using NGS A Kohlmann et al 137 15 Delhommeau F, Dupont S, Della VV, James C, Trannoy S, Masse A et al. Mutation in distinct gene and MicroRNA expression signatures. J Clin Oncol 2012; 30: TET2 in myeloid cancers. N Engl J Med 2009; 360: 2289–2301. 3109–3118. 16 Metzeler KH, Maharry K, Radmacher MD, Mrozek K, Margeson D, Becker H et al. 30 Kohlmann A, Grossmann V, Klein HU, Schindela S, Weiss T, Kazak B et al. TET2 mutations improve the new European LeukemiaNet risk classification of Next-generation sequencing technology reveals a characteristic pattern of acute myeloid leukemia: a Cancer and Leukemia Group B study. J Clin Oncol 2011; molecular mutations in 72.8% of chronic myelomonocytic leukemia by detecting 29: 1373–1381. frequent alterations in TET2, CBL, RAS, and RUNX1. J Clin Oncol 2010; 28: 17 Weissmann S, Alpermann T, Grossmann V, Kowarsch A, Nadarajah N, Eder C et al. 3858–3865. Landscape of TET2 mutations in acute myeloid leukemia. Leukemia 2012; 26: 934–942. 31 Kohlmann A, Klein HU, Weissmann S, Bresolin S, Chaplin T, Cuppens H et al. The 18 Metzeler KH, Becker H, Maharry K, Radmacher MD, Kohlschmidt J, Mrozek K et al. Interlaboratory RObustness of Next-generation sequencing (IRON) study: a deep ASXL1 mutations identify a high-risk subgroup of older patients with primary sequencing investigation of TET2, CBL and KRAS mutations by an international cytogenetically normal AML within the ELN Favorable genetic category. Blood consortium involving 10 laboratories. Leukemia 2011; 25: 1840–1848. 2011; 118: 6920–6929. 32 Thol F, Kolking B, Damm F, Reinhardt K, Klusmann JH, Reinhardt D et al. 19 Schnittger S, Eder C, Jeromin S, Alpermann T, Fasan A, Grossmann V et al. ASXL1 Next-generation sequencing for minimal residual disease monitoring in acute exon 12 mutations are frequent in AML with intermediate risk karyotype and are myeloid leukemia patients with FLT3-ITD or NPM1 mutations. Genes independently associated with an adverse outcome. Leukemia 2013; 27: 82–91. Cancer 2012; 51: 689–695. 20 Kohlmann A, Grossmann V, Nadarajah N, Haferlach T. Next-generation 33 Grossmann V, Kern W, Harbich S, Alpermann T, Jeromin S, Schnittger S et al. sequencing - feasibility and practicality in haematology. Br J Haematol 2013; 160: Prognostic relevance of RUNX1 mutations in T-cell acute lymphoblastic leukemia. 736–753. Haematologica 2011; 96: 1874–1877. 21 Patel JP, Gonen M, Figueroa ME, Fernandez H, Sun Z, Racevskis J et al. Prognostic 34 Klein HU, Bartenhagen C, Kohlmann A, Grossmann V, Ruckert C, Haferlach T et al. relevance of integrated genetic profiling in acute myeloid leukemia. N Engl J Med R453Plus1Toolbox: an R/Bioconductor package for analyzing Roche 454 2012; 366: 1079–1089. Sequencing data. Bioinformatics 2011; 27: 1162–1163. 22 The Cancer Genome Atlas Research Network. Genomic and epigenomic 35 Grossmann V, Roller A, Klein HU, Weissmann S, Kern W, Haferlach C et al. landscapes of adult de novo acute myeloid leukemia. N Engl J Med 2013; 368: Robustness of amplicon deep sequencing underlines its utility in clinical 2059–2074. applications. J Mol Diagn 2013; 15: 473–484. 23 Mardis ER, Ding L, Dooling DJ, Larson DE, McLellan MD, Chen K et al. Recurring 36 Sayers EW, Barrett T, Benson DA, Bolton E, Bryant SH, Canese K et al. Database mutations found by sequencing an acute myeloid leukemia genome. N Engl J Med resources of the National Center for Biotechnology Information. Nucleic Acids Res 2009; 361: 1058–1066. 2012; 40: D13–D25. 24 Ley TJ, Ding L, Walter MJ, McLellan MD, Lamprecht T, Larson DE et al. 37 Schwarz JM, Rodelsperger C, Schuelke M, Seelow D. MutationTaster DNMT3A mutations in acute myeloid leukemia. N Engl J Med 2010; 363: evaluates disease-causing potential of sequence alterations. Nat Methods 2010; 7: 2424–2433. 575–576. 25 Yan XJ, Xu J, Gu ZH, Pan CM, Lu G, Shen Y et al. Exome sequencing identifies 38 Adzhubei IA, Schmidt S, Peshkin L, Ramensky VE, Gerasimova A, Bork P et al. somatic mutations of DNA methyltransferase gene DNMT3A in acute monocytic A method and server for predicting damaging missense mutations. Nat Methods leukemia. Nat Genet 2011; 43: 309–315. 2010; 7: 248–249. 26 Grossmann V, Tiacci E, Holmes AB, Kohlmann A, Martelli MP, Kern W et al. 39 Kumar P, Henikoff S, Ng PC. Predicting the effects of coding non-synonymous Whole-exome sequencing identifies somatic mutations of BCOR in acute myeloid variants on protein function using the SIFT algorithm. Nat Protoc 2009; 4: leukemia with normal karyotype. Blood 2011; 118: 6153–6163. 1073–1081. 27 Welch JS, Ley TJ, Link DC, Miller CA, Larson DE, Koboldt DC et al. The origin 40 Mangan JK, Speck NA. RUNX1 mutations in clonal myeloid disorders: from and evolution of mutations in acute myeloid leukemia. Cell 2012; 150: conventional cytogenetics to next generation sequencing, a story 40 years in the 264–278. making. Crit Rev Oncog 2011; 16: 77–91. 28 Kohlmann A, Grossmann V, Haferlach T. Integration of next-generation 41 Grossmann V, Schnittger S, Kohlmann A, Eder C, Roller A, Dicker F et al. A novel sequencing into clinical practice: are we there yet? Semin Oncol 2012; 39: hierarchical prognostic model of AML solely based on molecular mutations. Blood 26–36. 2012; 120: 2963–2972. 29 Mendler JH, Maharry K, Radmacher MD, Mrozek K, Becker H, Metzeler KH et al. 42 Appelbaum FR. Incorporating hematopoietic cell transplantation (HCT) into the RUNX1 mutations are associated with poor outcome in younger and older management of adults aged under 60 years with acute myeloid leukemia (AML). patients with cytogenetically normal acute myeloid leukemia and with Best Pract Res Clin Haematol 2008; 21: 85–92.

Supplementary Information accompanies this paper on the Leukemia website (http://www.nature.com/leu)

& 2014 Macmillan Publishers Limited Leukemia (2014) 129 – 137