Introduction

Acute lymphoblastic leukaemia (ALL) with the MLL-AF4 fusion gene represents a very aggressive leukaemia subtype associated with poor prognosis1,2,3. The translocation involves the N-terminus of MLL (chromosome 11q23) that fuses in frame with the C-terminus of AF4 (chromosome 4q23) and leads to the balanced MLL-AF4 and AF4-MLL fusion genes4. Several studies investigated the leukaemogenic properties of both fusion genes in vitro and in vivo providing important insights into the pathogenesis of MLL-related leukaemia3,5,6. Nevertheless, in vivo studies provided discrepant results and mouse models resembling the features of the human MLL leukaemia are still missing3,7,8,9. In particular, the relative long disease latency7,8,9,10,11 in mice models argued against the concept that the MLL fusion gene alone is sufficient for full transformation and suggested that cooperating events are required to induce ALL. Whole genome sequencing12,13 as well as SNP analysis14,15 of MLL-AF4 positive patients reported very few additional genomic alterations suggesting that the MLL fusion per se predisposes cells to malignant transformation. However, patients with MLL rearranged leukaemia display high FLT3 expression16,17 and in vivo data showed that mutations in FLT3 accelerate MLL-induced leukaemogenesis18. Similarly, a transgenic mouse model simultaneously expressing MLL-AF4 and an activated KRAS mutated gene showed a shortened leukaemia latency11, which prompted for renewed KRAS and NRAS mutation screenings in MLL rearranged and wild type infant and paediatric patients.

Recently, several works showed the potency of next generation sequencing in revealing precise frequency of mutations even at very low frequencies19,20,21. In this study, we deeply investigated the RAS mutational status in infant and paediatric MLL-AF4+ patients and reported the results of ultra sensitive deep mutation sequencing of KRAS and NRAS hot spot regions in MLL-AF4+ paediatric patients at diagnosis. Moreover, to trace RAS mutations during the disease course we sequenced paired diagnosis and relapse samples. Furthermore, we studied the expression and proteomic profiles of RAS mutated (RASmut) patients looking for an activated RAS-related signature and we also evaluated the possibility to target the RAS pathway in MLL-AF4+ RASmut samples.

Results

Detection of KRAS and NRAS variants at diagnosis

Mutations with variant allele frequency (VAF) ≥1%20,22 were considered for analysis according to the calculated sensitivity threshold (Supplementary Figure 1). In total, we identified 23 mutated (63.9%) and 13 not mutated (36.1%) patients at diagnosis, Fig. 1A, Supplementary Tables 1 and 2. In particular, 16 out of 23 patients (69.6%) carried mutations in more than one clone, irrespective of the RAS mutated genes, whereas only 7 out of 23 (30.4%) carried mutations in one single clone. Moreover, clones harbouring different mutations at exactly the same genomic hotspot were also detected (ex. patients 13, 14, 22).

Figure 1
figure 1

RAS mutational profiles of infant and non-infant MLL-AF4 positive patients.

(A) Distributions of KRAS and NRAS mutations in the 36 analysed patients at diagnosis according to ultra deep sequencing analysis. The variant allele frequency (VAF) is reported as percentage using a colour code scale. (B) Schematic diagrams of all the identified mutations in KRAS and NRAS with VAF >1% with respect to proteins functional domains.

Full size image

Interestingly, 8 out of 23 patients (34.8%) simultaneously carried clones with both KRAS and NRAS mutations; on the other hand, 9 patients presented mutations only in KRAS whereas the remaining 6 patients (26.1%) harboured NRAS mutations only.

Overall, 29 mutations in KRAS exon 2 and 2 mutations in KRAS exon 3 were detected; only one KRAS mutation (A18D) and one silent KRAS mutation (A66A) were identified in a non hot spot region with a VAF ≥ 1%. The KRAS and NRAS mutations most frequently identified were G12D (11 clones) and G12D (7 clones) respectively, Fig. 1B,C.

RAS mutations and patients’ outcome

Clinical and outcome data were available for 34 out of 36 patients, 22 infants (aged 1 year or less at diagnosis) and 12 paediatric patients. Patients’ characteristics are described in Table 1. RAS mutations were significantly more frequent in infants as compared to older children: RASmut were 17/22 (77%) among infants, while only 4/12 (33%) in the paediatric cohort (p-value = 0.0248). No other significant association was observed between occurrence of any RAS mutation and relevant characteristics at diagnosis, i.e. sex, WBC count and PDN response. Types of events and outcome are described in Supplementary Table 3. All but one patient reached complete remission at the end of the induction phase. Most events were relapses and the 4-year CIR was markedly different according to RAS status, being 71.4 (SE 10.5) and 46.2 (SE 14.8) for RASmut and RASwt patients, respectively (p-value = 0.14). In agreement with previous data23, patients with a RAS mutation had a worse outcome than patients without RAS mutations, with a 4-year EFS of 14.3 (SE 10.9) vs. 38.5 (SE 13.8), although the difference was not statistically significant (p-value = 0.16), Supplementary Fig. 2.

Table 1 Patients characteristics.
Full size table

RAS mutations in paired diagnosis-relapse patients’ samples

To assess the development of RAS mutations along the disease course we analysed seventeen paired samples at diagnosis and relapse. Fourteen of these 17 paired samples (82.3%) presented RAS mutations at diagnosis; 3 patients did not show a RAS mutation at diagnosis and, similarly, did not acquire a RAS mutation at relapse. Seven of the fourteen mutated diagnostic samples (50%) still harboured a RAS mutation at relapse and 6/7 (85.7%) relapsed mutated patients presented at least one of the mutated clones detected at diagnosis also at relapse. Additionally, de novo mutated clones were also detected at relapse in patient 18 (3 clones) and patient 22 (1 clone), Supplementary Figure 3, Supplementary Table 1 and Supplementary Table 4. Analysing the data with respect to VAF, we observed that 4/5 patients (80%) (i.e. patient 9, 12, 26, 35) with a total VAF (i.e. VAF index) <10% at diagnosis lost their primary RAS mutated clones at relapse and only patient 18 presented four mutated clones at relapse, one from the diagnosis and three acquired de novo at relapse. On the contrary, 8/9 (88.9%) patients (i.e. patients 4, 10, 21, 22, 25, 27, 28, 29) with VAF index at diagnosis >10% lost some of their initial clones but, at the same time, 5/9 (55.5%) patients (i.e. patients 10, 25, 27, 28, 34) maintained clones already present at diagnosis or even acquired a de novo mutation (i.e. patient 22).

In general, four distinct behaviours of RASmut clones were identified, Fig. 2A: (i) clones at diagnosis disappearing at relapse, (ii) clones at diagnosis stably maintained at relapse, (iii) mutated clones not present at diagnosis appearing de novo at relapse, (iv) mutated clones at diagnosis evolving at relapse. The distribution of RAS mutated clones in paired diagnosis and relapse samples is shown in Fig. 2B,C and could be grouped in 3 major clusters: (i) mutations present at diagnosis and no longer detected at relapse (cluster 1), (ii) mutations stable from diagnosis to relapse or expanding at relapse (cluster 2) and (iii) de novo mutations at relapse (clusters 3).

Figure 2
figure 2

RAS clones dynamics over time.

(A) Plots exemplifying the four different behaviours of the mutated clones identified in paired samples at diagnosis (D) and relapse (R). (B) Distributions of KRAS and NRAS mutations in the 17 analysed patients at diagnosis (D) and relapse (R) according to ultra deep sequencing analysis. The variant allele frequency (VAF) is reported as percentage using a colour code scale. Straight lines separate individual patients and dotted lines divide data measured at diagnosis and relapse. (C) The variant allele frequency (VAF) of mutated clones is reported as percentage in matched samples at diagnosis and relapse allowing the identification of three major clusters. Cluster 1: mutations present at diagnosis only; cluster 2: mutations with a constant and/or increasing VAF over time; cluster 3: de novo mutations at relapse. Circle: KRAS mutation; triangle: NRAS mutation. (D) RAS mutated clones in first relapse (1st R), second (2nd R) relapse and in available control time points (C) in patient (Pt) 28 and patient (Pt) 35. (E) Detection of RAS mutated clones in DNA isolated from bone marrow cells of xenotransplanted NOD/SCID mice (N = 2) using patient 35 BM cells at diagnosis. 1st P: primary passage; 2nd P: secondary passage.

Full size image

Interestingly, the expansion of a clone at relapse was not related to the mutation load at diagnosis: e.g. in two patients (i.e. patients 18 and 22) the presence of the mutated clone at relapse was independent of the frequency of the variants of the clones at diagnosis, suggesting a positive clonal selection/survival over time. A similar selection process was seen in other three patients (i.e. patients 25, 27, 28) presenting at relapse only one of the clones that had been detected at diagnosis even if the clones at diagnosis had a similar mutation load.

In two cases (i.e. patient 28 and patient 35) we analysed also the DNA at second relapse (patient 28) and at two control time points (patient 35), Fig. 2D. In patient 28, the mutated clone identified at diagnosis and at primary relapse was maintained also at second relapse. Conversely, in patient 35, the KRAS G12S clone detected at diagnosis disappeared at subsequent time points.

The selection process acting on RASmut leukaemia clones that we postulated based on the analysis of patient samples at different time points was also confirmed in NOD/SCID mice (n = 2 mice/passage) serially transplanted with the BM cells of patient 35. The clone harbouring the KRAS G12S mutation present at diagnosis was negatively selected and not detectable in engrafted mice whereas a major clone carrying the NRAS Q61R mutation resulted to be the only one present in leukaemia cells isolated from leukaemia bearing mice at secondary passage, Fig. 2E. Of note, backtracking the NRAS Q61R clone in patient 35 specimens, we observed that this clone was present at all time points but with a VAF below the 1% cut-off -threshold, Fig. 2D.

Detection of an activated RAS-related profile

To investigate if the presence of RAS mutated clones was associated with a distinct leukaemia transcriptional profile, fifteen samples (n = 4 RASwt and n = 11 RASmut) with available gene expression data were analysed. We first examined the expression levels of HOXA9, HOXA10 and IRX1 genes that were previously shown to discriminate within MLL-AF4 leukaemia two distinct signatures24,25,26, Fig. 3A. Six patients with low HOXA9, HOX10 (HOXAlow) and high IRX1 (IRXhigh) expression were identified; all 6 HOXAlow IRXhigh patients were RASmut (no HOXAlow RASwt patients were available) whereas the 9 HOXAhigh IRXlow patients consisted of 5 RASmut and 4 RASwt patients. Aiming (i) to identify a gene signature reflecting the effect of an activated RAS signalling and (ii) to avoid comparison of HOXAlow and HOXAhigh RASmut patients versus HOXAhigh RASwt patients only as this results in the detection of genes belonging to the HOXAhigh/HOXAlow signature (i.e. HOXA5, HOXA9, HOXA10, IRX2, IRX1, Supplementary Figure 4), we decided to exclude the 6 HOXAlow IRXhigh patients from further analysis. The 9 samples with high HOXA9, HOXA10 (HOXAhigh) and low IRX1 levels (IRXlow) presenting both RASwt (n = 4) and RASmut (n = 5) clones had a total VAF (VAF index) ranging from 4.69% to 40% (mean value = 26%) and were used to identify transcriptional features related to the RAS mutation status. We identified a clear gene signature distinguishing the HOXhigh/IRXlow patients according to their RAS mutational status, Fig. 3B and Supplementary Table 5. Furthermore, the presence of a gene expression profile consistent with RAS pathway activation was confirmed by enrichment modules analysis of GSEA data. Indeed, RASmut patients showed a negative enrichment for down-regulated KRAS related genes and a positive enrichment for up-regulated MAP2K1, GTP-ases, phosphorylation and ligase-related signatures, Fig. 3C. Moreover, the connectivity map analysis on the RASmut patients’ signature identified within the top 15 ranked perturbagenes 4 compounds (i.e. lycorine, ouabain, enoxacin, cicloheximide) with negative enrichment score linked to the RAS activity modulation and potentially able to revert the RAS mutated phenotype27,28,29,30, Fig. 3D and Supplementary Table 6. To confirm the presence of an active RAS related signature also at protein level, we performed a reverse phase protein arrays (RPPAs) analysis on 2 RASwt samples and 8 RASmut samples with a VAF index ranging from 2.83% to 37.6% (mean value = 19.2%) assessing the phosphorylation status of several RAS-downstream targets in MLL-AF4 patients at diagnosis. The analysed molecules belong to the transcription/cell cycle regulation (MAPK3/MAPK1, MAPKAPK2), cell proliferation (PIK3CA-AKT1-MTOR) and the Ca2+ signalling (PRKCA and PRKCD) pathways. Moreover, we also included YAP1 that was recently shown to be stabilized by mutated RAS31 and to be crucial for proliferation of mutant KRAS neoplastic pancreatic cells32. RASwt and RASmut patients formed almost two distinct clusters in an unsupervised analysis with only two RASmut patients with VAF >10% clustering in the RASwt branch, Fig. 3E.

Figure 3
figure 3

Detection of a RAS related signature in RASmut patients.

(A) HOXA9 (209905_at and 214651_s_at), HOXA10 (213150_at and 213147_at) and IRX1 (230472_at) expression levels in 15 MLL-AF4 positive patients at diagnosis. *: RAS wild type (RASwt) patients. (B) Principal component analysis (PCA) of HOXAhighIRXlow/RASmut and HOXAhighIRXlow/RASwt patients. The samples were projected into a 2-dimensional space (PC1 and PC2) consisting of the differentially expressed genes (FDR q-val < 0.05) between HOXAhighIRXlow/RASmut (N = 5; label: 2) and HOXAhighIRXlow/RASwt (N = 4; label: 1) patients. Samples with similar characteristics will cluster together. (C) Enrichment map visualization of significant (FDR q-val < 0.05) enriched gene sets comparing HOXAhighIRXlow/RASmut and HOXAhighIRXlow/RASwt patients. Gene sets are depicted as circles (nodes) with edges indicating overlap between nodes. Node size is proportional to gene set size and the edges thickness shows the degree of overlap among the nodes. Red and blue colours indicate positive and negative enrichment in HOXAhighIRXlow/RASmut samples, respectively. (D) Bar-views according to connectivity map analysis. Each black line represents an individual treatment instance ordered according to its corresponding connectivity score (+1 and −1) with respect to the query signature based on genes differentially expressed between HOXAhighIRXlow/RASmut and HOXAhighIRXlow/RASwt patients with FDR q-val < 0.1. Instances at the bottom (connectivity score: −1) are more strongly anti-correlated with the query signature indicating the possibility to revert the RASmut phenotype. (E) The hierarchical clustering analysis of RASmut (N = 8) and RASwt (N = 2) patients samples according to proteins levels measured by RPPA reveals two clusters. One cluster contains all but 2 RASmut patients and the second one contains the 2 RASwt and the remaining 2 RASmut patients. Rows represent each analysed protein and columns represent patients. The red and green colours reflect high and low expression level, respectively. (F) MTT cell viability assay in the MLL-AF4+ RASmut cell line MI04 and in two RASwt MLL-AF4+ cell lines (i.e. RS4; 11 and SEM) treated for 72 h with increasing concentration of PD0325901. The myeloid cell line THP-1 with the NRAS G12D and the MLL-AF9 fusion gene was use as positive control. Data represent mean values ± s.d. of three independent experiments.

Full size image

Targeting RAS in MLL-AF4 leukaemia

The presence of RAS mutations at diagnosis detected in 63.9% of patients as well as gene and protein expression data supporting the presence of an active RAS-related profile in patients harbouring RAS mutations prompted us to explore the possibility of targeting the RAS pathway as a potential therapeutic approach in BCP-ALL MLL-AF4+ RASmut leukaemia. We choose to indirectly target RAS using PD0325901, a potent inhibitor of MAP2K1/2 kinase33 currently in phase II clinical trials34. The NRAS Q61K mutated MLL-AF4+ B cell line MI04 displayed high sensitivity at 72h post-treatment with PD0325901 even using nanomolar dosages, whereas the two RASwt MLL-AF4+ cell lines (RS4;11 and SEM) showed only a modest proliferative reduction even increasing the drug concentration. Of note, 10nM PD0325901 showed 60% reduction in growth capability of the RASmut MI04 cell line at 72h of treatment, Fig. 3F.

Discussion

The capability of the MLL-AF4 fusion gene to sustain leukaemogenesis without additional genetic alterations is still not completely clear. Several works showed that BCP MLL-AF4 rearranged leukaemia is characterized by a very low rate of genetic alteration13,14,15 and using Sanger sequencing a significant amount of data on the mutational status of RAS genes in infants and paediatric patients with MLL-AF4 leukaemia were previously obtained23,35,36,37. However, the percentage of RAS mutated patients showed a certain inconsistency among these studies ranging from 25 to 50%23,35,36,37. In the present study we analysed the mutational status of NRAS and KRAS hot spot regions in 36 MLL-AF4 patients taking advantage of the ROCHE 454 massive parallel sequencing approach that offers a powerful tool to investigate clonal heterogeneity and detect minor mutated clones19,20,21. Thanks to the ultra deep sequencing, we identified RAS mutations in 63.9% of patients at diagnosis. This higher recurrence of RAS mutations measured in our samples compared to previously published data can be attributed to the sample cohort as well as to the considerable increase in sensitivity of next generation sequencing (NGS) technology20,38,39 compared to Sanger-based sequencing (Supplementary Figure 5).

Using ultra deep sequencing analysis, we also dissected the clonal architecture of MLL-AF4+ B-cell leukaemia. Among RASmut MLL-AF4 rearranged leukaemia specimens at diagnosis, 69.5% were characterized by ≥2 different clones with a distinct RAS mutation in the mutation hot spot regions, which also means that KRAS and NRAS mutations are not mutually exclusive. These results further support previous data on the MLL-AF4 clonality based on the analysis of Ig/TCR rearrangements40, which overall depicts MLL-AF4+ leukaemia as a highly heterogeneous and subclonal disease.

Moreover, in line with previous data23, a higher frequency of RAS mutation was detected in infants rather than in older patients, corroborating the concept that aberrant RAS expression might shorten leukaemia latency11.

Nevertheless, all mutations detected at diagnosis had a VAF well below 50%, suggesting that only minor clones carry RAS genetic alterations and, in agreement with recent reports13,23,37,41, pointing to RAS mutations as secondary genetic events. However, the subclonal nature of KRAS and NRAS activating mutations is not just an unusual feature in cancer; low RAS mutation load was reported in colorectal cancer biopsies42, early T-cell precursor acute lymphoblastic leukaemia (ETP-ALL)43 and also in acute myeloid leukaemia (AML) with MLL rearrangements44.

Intriguing results were also obtained analysing the RASmut clones during the disease progression. The analysis of seventeen matched diagnosis-relapse samples revealed that 50% of the mutated patients at diagnosis carried also RAS mutations at relapse, pointing to RAS mutations as a recurrent genetic aberration not only at diagnosis but also at relapse. Even more interestingly, 6 out of 7 RASmut patients at relapse carried the same RAS mutated clone detected at diagnosis, which suggests that at least one clone resisted chemotherapy and survived the therapy-related selection pressure (Fig. 2B, cluster 2), although a fractions of mutated clones at diagnosis were targeted by the chemotherapy treatment and disappeared at relapse. De novo mutated clones (n = 4) (Fig. 2C, cluster 3) were also identified at relapse in 2 patients, further supporting the notion of RAS mutations as secondary events. Moreover, even though only 50% of the analysed paired samples presented a mutation also at relapse, a percentage that still raises some questions about the overall role of RAS mutations in MLL-AF4+ leukaemia, the mutated cells with a total VAF >10% seemed to show a certain dependency on the presence of an active RAS signalling, which may suggest that RAS mutations can sustain proliferative pathways already active in the leukaemic cells. MLL-AF4 leukaemia is known to be characterized by high FLT3 expression16 which results in the activation of its downstream proliferative pathways such as RAS/MAP2K1/MAPK1 and the PIK3CA/AKT1 pathways45. We did not identify a significant correlation between FLT3 expression and RAS VAF in RASmut patients (Supplementary Figure 6); however, we could not exclude the hypothesis that RAS mutations may still complement existing proliferative signalling pathways, thus providing support to cells’ survival.

Moreover, our analysis of RASmut patients at diagnosis highlighted the over-expression of RAS down-stream targets46 proving, in line with previous observations from transcriptome analysis13, the functionality of RAS signalling in these specimens, which is totally independent from the HOXA genes’ signature24.

Overall, the high percentage of RASmut patients at diagnosis, the presence of an active RAS-related signature at diagnosis as well as the detection of therapy resistant RAS mutated clones in 6/7 (85.7%) mutated samples at relapse indicated that RAS-pathway inhibition might represent a promising approach for a MLL-AF4+ RAS mutated patient-tailored treatment. Direct targeting of RAS has been evaluated in diseases coined “RASopathies” all associated with RAS mutations but without any successes so far47. Conversely, preclinical studies evaluating the inhibition of RAS-downstream molecules such as RAS/MAP2K1/2 and MAPK3/1 or PIK3CA/AKT1 have provided promising results47,48. Here, we successfully established for the first time, to our knowledge, a RASmut BCP-MLL-AF4 positive leukaemia cell line and we showed that the allosteric MAP2K inhibitor PD032590149 at nanomolar concentration reduced the RASmut MLL-AF4+ B cell line proliferation to 40% without any significant effect on RASwt MLL-AF4+ B cell lines even at micromolar concentration, suggesting to evaluate RAS-pathway inhibition by PD0325901 in MLL-AF4+ RASmut specimens in preclinical studies. These results are in line with published data on RAS mutant paediatric B-ALL samples (not MLL rearranged) that were treated with two different MAP2K1/MAPK1 inhibitors (i.e. PD98059 and U0126) showing higher cytotoxicity in the mutant samples and no effects in the RASwt specimens50.

In conclusion, taking advantage of next generation sequencing technology we demonstrated that RAS mutations are very frequent in MLL-AF4+ leukaemia especially in infant patients. Moreover, the analyses on paired diagnosis-relapse samples indicate that even if current therapy regimens can eliminate some of the RAS mutated clones, they do not remove all clones but rather allow the positive selection of the most resistant ones at relapse. Therefore, the potential use of RAS pathway inhibitors such as PD0325901 could be considered for precision of current therapy.

Materials and Methods

Patient samples, cell lines and xenotransplantation

DNA was isolated from bone marrow (BM) and/or peripheral blood (PB) of 36 paediatric patients diagnosed with MLL-AF4 positive B cell precursor (BCP) ALL and participating to the Interfant-99 (n = 10), Interfant-06 (n = 13) and AIEOP-BFM ALL-2000 (n = 13) treatment protocols. Patients’ characteristics are reported in Table 1. The present study has been carried out in accordance with ethical principles of the Declaration of Helsinki. The residual patients’ material (such as mononuclear cells from PB and BM, DNA and RNA) retained after the completion of the diagnostic screenings has been used for research purposes. The written informed consent was obtained from the patients’ parents or legal guardians. The use of primary patient samples was approved by the local Ethics Committee and by the scientific board of the Interfant Protocol (www.oncauvergne.fr/index.php?option=com_docman&task=doc_download&gid=944&Itemid). The methods were carried out in accordance with the approved guidelines and the samples used for analysis were selected according to material availability.

The MLL-AF4 positive cell lines RS4;11 and SEM were purchased from the German Collection of Microorganisms and Cell Cultures (DSMZ, Braunschweig, Germany) and were used as RAS wild type (RASwt) controls. The MLL-AF9 positive myeloid cell line THP-1 was also purchased from DSMZ and was used as a positive RASmut control. Information regarding the RAS mutational status of RS4;11, SEM and THP-1 cell lines was collected from the Cancer Cell Line Encyclopedia (CCLE) (http://www.broadinstitute.org/ccle/home). A RAS mutated (RASmut) BCP-ALL MLL-AF4 positive cell line (named MI04) was successfully established culturing the BM mononuclear cells of a patient diagnosed with MLL-AF4+ BCP ALL. Outgrowing cells were initially cultured in the presence of autologous BM stromal cells. After few weeks cell line growth became independent from the presence of the stromal cell layer. Cell culture conditions are reported below. The presence of the MLL-AF4 and AF4-MLL translocations and of the N-RAS Q61R mutation in the newly established cell line were assessed by FISH using a MLL/AFF1 dual fusion probe (Cytocell, United Kingdom) and Sanger Sequencing, respectively (Supplementary Figure 7). All cell lines were maintained as suspension cultures in RPMI 1640 (Invitrogen Life Technologies, The Netherlands) supplemented with 10% fetal calf serum (FCS), glutamine (2 mM/l; GIBCO, Invitrogen Life Technologies, Carlsbad, CA, USA), penicillin (100 U/ml; GIBCO) and streptomycin (100 μg/ml; GIBCO) and maintained at 37 °C in a humidified atmosphere with 5% CO2. Cells were treated with PD0325901 (Selleckchem, Houston, TX, USA) dissolved in dimethylsulphoxide (DMSO) for 72 hours and at different concentrations, or with DMSO alone.

Bone marrow cells from patient 35 at diagnosis were serially transplanted into sublethally irradiated NOD/SCID mice via tail vein injection, as previously described40. Upon signs of illness, mice were sacrificed and BM cells collected for DNA extraction. The in vivo procedures were performed in the Animal Facility at the University of Milano-Bicocca, under the protocol nr. 64/2014-PR approved by the National Ministry of Public Health in accordance with the relevant guidelines and in compliance with the National law (D. Lgs. n. 26/2014) and the European Directive (2010/63/UE) about the protection of live animals used for scientific purposes.

Amplicon ultra deep sequencing preparation

DNA was extracted using the Gentra Puregene DNA isolation kit (QIAGEN, Hilden, Germany); 20 ng/ul of genomic DNA were processed for the generation of PCR amplicons for ultra deep sequencing, according to manufacture’ protocol (Roche Applied Science, Penzberg, Germany). Fusion primers with 6 different molecular identifier barcode sequences (MID) were designed to amplify hot spot regions of exon 2 and exon 3 of KRAS and NRAS genes (Supplementary Table 7), for a total of 72 different amplicons for each gene. PCRs were conducted using FastStart High Fidelity PCR System kit (Roche Applied Science). Amplicon products were purified using Agencourt AMPure XP beads (Beckman Coulter, Krefeld, Germany), quantified using the Quant-iT PicoGreen dsDNA kit (Invitrogen, Carlsbad, CA, USA) and equimolar pooled together for the emPCR-Lib-A method (library at 1 × 107 molecules/ul). For each amplicon forward (A beads) and reverse (B beads) reactions were carried out using 5.000.000 beads per emulsion oil tube. The amplification reaction, breaking of the emulsions and enrichment of beads carrying amplified DNA were performed using the workflow as recommended by the manufacturer (Roche Applied Science).

Data analysis and variants detections

Data analysis was performed using the Roche proprietary software package for the GS Junior system (http://www.454.com/). Sequencing data was generated using GS Junior Sequencer Instrument software version 2.7 (Roche Applied Science). Image processing was performed using default settings of the GS RunBrowser software version 2.7 (Roche Applied Science). We generated a median of ~75.000 and ~62.000 sequencing reads for KRAS and NRAS obtaining a median of 1540- and 1762-fold coverage for KRAS and NRAS amplicons.

Sequence alignment and variant detection were performed using the GS Amplicon Variant Analyzer (AVA) software version 2.7 (Roche Applied Science). This software extracts sequences from the standard flowgram format (SFF) files generated after pyrosequencing and automatically assigns each read to the proper sample by looking for the MIDs located at both ends of amplicon. Only sequences with an average phred equivalent quality score >Q20 were conserved. The MIDs and primer sequences within the read have also to be complete without mismatch. Moreover, the reads have to match the full-length amplicon. AVA application computes the alignment of filtered reads from amplicon libraries to identify differences between the reads and the reference sequences (RefSeq NM_004985 and NM_002524, for KRAS and NRAS, respectively) to call variants. The Variant/Consensus parameters of AVA software were set to minimum read percentage of 0.01% (per read direction), minimum read count of “1” per orientation and appearing in both forward and reverse directions and dynamic N-mer thresholding for homopolymers. Variant Allele Frequencies (VAFs) of variant i were measured as the number of reads carrying the variation i, divided by the number of all reads spanning the position of the variant.

In order to establish a bioinformatic approach to identify the best threshold for KRAS and NRAS subclonal mutations out of the background error noise of ultra-deep-NGS, a triplicate experiment was performed for a subset of patients. Variants across the entire sequence of amplicons were analysed and evaluated for recurrence in all triplicate samples. Only variants presented in all triplicate were considered as true positive call. Variants were divided in different intervals thresholds (i.e. 0.01–0.1; 0.1–03; 0.3–0.5; 0.5–1; 1–5; >5) to estimate the best cut-off with minor false positive call. Variants overlapping the primers or neighbouring them by one nucleotide and variants mapping within homopolymeric tracts that specifically suffer from errors in 454 NGS51,52 were excluded from the analysis. The same approach was used considering only variants occurring in hot spot loci. Variants with a VAF of 1% and more in bidirectional reads per amplicon were considered.

Different variants for a specific genomic position in the same amplicon were considered as different clone. The mutated clones’ dynamics between diagnosis and relapse were defined in clusters as follows: Cluster 1: variants detected at diagnosis and disappeared at relapse; Cluster 2: all variants detected both at diagnosis and at relapse with similar or increased VAF; Cluster 3: de novo variants not present at diagnosis but appearing at relapse.

Gene expression analysis

Gene expression profiles (GEPs) of patients included in this study were available as a part of the MILE study53. For the gene expression analysis only MLL-AF4 positive patients with available RAS mutational status information and high HOXA9 and HOXA10 expression levels were used. Data analysis was performed in R (http://www.R-project.org/ version 3.0.2) using Bioconductor and R packages. Probe level signals were normalized and converted to expression values using the robust multi-array averaging (RMA) algorithm54. After normalization, batch effects were removed due to different protocols using ComBat method55. Differentially expressed genes (FDR q-val < 0.05) were identified using the Shrinkage t-statistic56. Gene set enrichment analysis (GSEA) was done comparing the expression profiles of RASmut and RASwt samples using the “C5 molecular function” and “C6 oncogenic signatures” gene sets within the molecular signatures databases (MSigDB) collection57. The signal to noise metric and the gene-set permutation were used to identify statistical enrichment of the selected gene sets. Enrichment maps showing relationships between gene sets with significant enrichment (p-value cutoff < 0.005; FDR q-val < 0.05; overlap coefficient: 0.5) were generated using the Enrichment Map plug-in in Cytoscape 3.2.158. Only Gene sets with FDR <0.05 derived from the “C6 oncogenic signatures” and the “C5 molecular function” within the MSigDB were used to build the network. To generate gene-set relationships, we used the overlap coefficient cut-off parameter set to 0.5. A node represents a functional gene set and the size is proportional to size of gene set. Edges represent the degree of gene overlap between two gene sets and the thickness is proportional to the overlap between the gene sets. The intensity of the node colour is proportional to the enrichment score. Connectivity map analysis59 was performed using genes differentially expressed (FDR q-val < 0.1) between RASmut and RASwt samples.

Reverse Phase Protein Arrays (RPPA)

Reverse Phase Protein Arrays (RPPA) were performed as previously described60. Briefly, for this analysis 10 patients (8 RASmut and 2 RASwt) were lysed and printed on nitrocellulose coated glass slides using the 2470 Aushon Arrayer (Aushon BioSystems, Billerica, MA, USA). Slides were stained with the following antibodies on an automated slide stainer (Dako Autostainer Plus, Dako Cytomation, Carpinteria, CA, USA) using the CSA kit (Dako Cytomation): AKT1 S473 (1:100), AKT1 T308 (1:100), MTOR S2448 (1:100), RPS6KB1 T389 (1:200), PDPK1 S241 (1:100), EIF4G1 S1108 (1:100), GSK3A/B S21/9 (1:100), MAP2K1/2 S217/221 (1:500), MAPK3/MAPK1 T202/Y204 (1:500), PKCD T505 (1:75) (all from Cell Signaling Technology Inc, Danvers, MA, USA), PKCA S657 (1:2500) (Merck Millipore, Darmstadt, Germany), YAP1 S127 (1:100) and EIF4EBP1 S65 (1:250) (Abcam, Cambridge, UK) and AKT1S1 T246 (1:1000) (Thermo Fischer Scientific, Waltham, MA, USA). Stained slides were analysed using the MicroVigene software (VigeneTech Inc, Boston, MA, USA), normalizing for total protein content quantified by Fast Green FCF (Sigma-Aldrich, St Louis, MO, USA) staining. Normalized protein values were used for an unsupervised hierarchical clustering analysis using the Ward’s method in R (version 2.14.1).

MTT assay

Cell proliferation was assessed by MTT ((3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide) assay after treatment. Equal concentrations of cells were plated in quadruplicate in a 96-well plate and incubated with 10 μl MTT (Sigma-Aldrich) for 4h. Absorbance was measured at 560 nm using Victor3TM 1420 Multilabel Counter (PerkinElmer, Waltham, MA, USA).

Statistical analysis

Event-free survival (EFS) was defined as the time from diagnosis to first event, i.e. resistance, relapse, death from any cause, or second malignant neoplasm. Observation periods were censored at time of last contact when no events were reported. EFS curves were estimated with the Kaplan-Meier method and standard errors according to Greenwood and compared with the log-rank test. Cumulative incidence of relapse (CIR) curves were estimated adjusting for competing risks and compared with Gray’s test. Fisher’s exact test was used to assess the association between clinical characteristics and occurrence of RAS mutations. All tests were two sided. Analyses were performed using SAS 9.2. Correlation analysis and Mann-Whitney test were performed using Prism 5 for MAC Version 5.0a; p- values < 0.05 were considered significant.

Additional Information

Accession Codes: Gene expression data are deposited in the GEO database under accession number GSE77416.

How to cite this article: Trentin, L. et al. Deciphering KRAS and NRAS mutated clone dynamics in MLL-AF4 paediatric leukaemia by ultra deep sequencing analysis. Sci. Rep. 6, 34449; doi: 10.1038/srep34449 (2016).