Introduction
Acute lymphoblastic leukemia (ALL) is the most frequent childhood malignancy with an assumed worldwide annual incidence of more than 50,000 cases [
1,
2]. Fortunately, many clinical trials recently demonstrated 5-year survival rates above 90% in pediatric ALL, whereas only a few decades ago, survival rates were below 20% [
1,
3]. The great improvement in prognosis is essentially a result of iterative cycles of long-term clinical trials. These trials helped to identify biologic subtypes of ALL, to assess their distinct response to treatment and risk of relapse, and to develop individual risk-adapted therapies [
4].
Risk stratification within the recently completed AIEOP-BFM ALL 2009 trial, an international collaborative treatment protocol for children and adolescents with acute lymphoblastic leukemia, was based on immunophenotype, specific chromosomal aberrations, and response to treatment assessed by minimal residual disease (MRD). The two chief chromosomal aberrations relevant to risk stratification in AIEOP-BFM ALL 2009 were the
AFF1-
KMT2A gene fusion and a hypodiploid karyotype. The fusion
BCR-
ABL1 represented an exclusion criterion and those cases were treated within the EsPhALL study [
5]. Cases showing none of these genetic markers were treated based on their response to treatment and MRD has proven to be an effective measure to identify patients with a high risk of relapse [
6]. However, it is known that other genetic aberrations, some of them recurrent, exist in the MRD high-risk group and great efforts are being made to identify and characterize these aberrations, as this knowledge will be key to improve risk stratification, to introduce targeted treatment, and hopefully to increase long-term survival rates [
7]. Therefore, the diagnostic workflow of the current AIEOP-BFM ALL 2017 study investigates for an extended set of stratification relevant gene fusions (i.e.,
ETV6-
RUNX1,
AFF1-
KMT2A, and
TCF3-HLF) as well as the
IKZF1plus subtype [
8,
9]. It is noteworthy that
IKZF1plus is not defined by a single genetic aberration, but by deletions in six different genetic loci (i.e., deletion of
IKZF1 complemented by at least one additional deletion in
CDKN2A,
CDKN2B,
PAX5, or
CRLF2-
P2RY8, and no deletion in
ERG) [
8]. The
IKZF1plus profile of gene deletions points at the necessity that future routine diagnostic workflows, which are currently often limited to specific genes of interest using fluorescence in situ hybridization (FISH) and polymerase chain reaction (PCR) analyses, will have to assess a wider range of targets in order to detect novel and yet undefined profiles relevant to risk stratification.
We here report on our results from implementing analysis of the leukemia transcriptome using panel-based RNA sequencing and genome-wide copy-number analysis using array CGH into the diagnostic workflow of the German study group of the international AIEOP-BFM ALL 2017 trial. In addition to assessing detection of IKZF1plus in array CGH, the major aim of this study was to compare the RNA-seq-based detection of above mentioned gene fusions with well-established techniques such as karyotyping, FISH, and PCR analyses. We also discuss expanded use of RNA-seq data beyond stratification relevant gene fusions, which in some cases identified potentially targetable lesions. Prospectively, the consistent collection of genome-wide CGH+SNP array as well as RNA-seq data within the AIEOP-BFM ALL 2017 trial is a valuable resource for elucidation of new prognostic markers of pediatric ALL.
Methods
Patients and samples
This study was performed on a consecutive cohort of patients with pediatric ALL sent to our genetic reference laboratory for initial cytogenetic and molecular genetic diagnostics. All patients were treated in German centers and were enrolled in the AIEOP-BFM ALL 2009 trial (EudraCT no. 2007-004270-43). Written informed consent was obtained from the legal representatives (and the patients, if applicable). The study cohort consisted of n = 117 patients (n = 65 male, n = 52 female) with a median age of 4.8 years. Of these patients, 6 (5.1%) were classified with an immunophenotype of pro-B-ALL (B-I), 72 (61.5%) with a common-ALL (B-II), and 39 (33.3%) with a pre-B-ALL (B-III).
Bone marrow (BM) aspirate was obtained at the time of initial diagnosis. BM from heparinized tubes was used for cytogenetic analyses (karyotyping and FISH). RNA was preferably isolated from EDTA BM, or if necessary from heparin BM, on a Chemagic 360 instrument as per the manufacturer’s instructions (PerkinElmer, Baesweiler, Germany). RNA integrity (RIN value) was measured by Bioanalyzer 2100 (Agilent Technologies, Waldbronn, Germany). Isolation of DNA from EDTA BM was performed using the QIAamp DNA Blood Midi Kit (Qiagen, Hilden, Germany).
Genetic markers relevant for risk stratification
This study assisted to optimize the diagnostic standard workflow of the German study group of the AIEOP-BFM ALL 2017 trial. Using the methods listed below, the following genetic markers relevant for risk stratification were evaluated: gene fusions (ETV6-RUNX1, AFF1-KMT2A, and TCF3-HLF), hypodiploidy (< 45 chr.), and IKZF1plus. BCR-ABL1 is an exclusion criterion for the AIEOP-BFM ALL trial, as these cases are treated within the EsPhALL study instead.
Karyotyping and FISH analysis
For all patients, short-term cultures (24–48 h) were set up from BM aspirates. Chromosome preparation, fluorescence R-banding, and FISH were performed as previously described [
10]. Karyotypes were described according to the International System for Human Cytogenetic Nomenclature (ISCN, 2016). In addition to classical banding analysis, the diagnostic standard protocol included FISH analyses for
BCR-ABL1 and
ETV6-RUNX1 fusions as well as for translocations affecting the genes
KMT2A and
TCF3 (FISH probes: Vysis LSI
BCR/
ABL Dual Color, Abbott, Wiesbaden, Germany; Vysis LSI
ETV6(
TEL)/
RUNX1(
AML1) ES Dual Color, Abbott; Vysis LSI
MLL(
KMT2A) Dual Color, Abbott; SureFISH 19p13.3
TCF3 DF, Agilent) [
10].
Reverse transcriptase PCR
Reverse transcriptase PCR (RT-PCR) was performed in parallel to cytogenetic analyses whenever EDTA-material was available (74/117 cases) for the qualitative detection of fusion transcripts of
BCR-ABL1,
ETV6-RUNX1, and
AFF1-
KMT2A as published elsewhere [
11,
12]. In cases where only heparinized material was available, no RT-PCR was performed due to possible inhibitory effects of heparin on amplification. cDNA quality was verified by quantification of the
ABL1 gene.
Panel-based RNA sequencing and data analysis
RNA-seq was performed using the TruSight RNA Pan-Cancer Panel (Illumina, San Diego, CA, USA) according to the manufacturer’s instruction. The panel targets 1385 cancer genes (see online Tab.S1) and detects fusions between genes targeted by the panel but also fusions of targeted genes with non-target genes. Prior to this study, the RNA-seq method was successfully validated on commercially available reference material (SeraSeq® Myeloid Fusion RNA Mix, Hiss Diagnostics GmbH, Freiburg, Germany), external reference samples with defined fusions provided by the group of Giovanni Cazzaniga (University of Milano-Bicocca, Fondazione MBBM, Monza, Italy), and genetically characterized cell lines and samples from our laboratory (see Tab.S2 for tested fusions). Gene-pseudogene “fusions” and recurrent technical artifacts were defined during this validation process and excluded from later data analysis. Each RNA-seq run contained a pooled cDNA library of eight patient samples and was executed on an Illumina MiSeq machine using MiSeq® Reagent Kit v3 (150 cycles) with a PE MiSeq® Flow Cell. The anticipated minimum number of reads per sample was 3 million.
Data analysis was performed using the Illumina BaseSpace apps TopHat Alignment (version 1.0.0, read mapping on hg19 reference genome by TopHat2 [
13], fusion calling by TopHat-Fusion [
14]) and RNA-seq Alignment (version 1.1.0, read mapping on hg19 reference genome by STAR [
15], fusion calling by Manta [
16]) using standard settings (
https://basespace.illumina.com/apps). Fusion transcripts with a low number of split-reads (< 10) were excluded as likely false-positives, unless the fusion was verified with a second method. The turn-around time from RNA isolation to detection of fusion transcripts using this workflow is 4 days.
Validation of single gene fusions detected in RNA-seq was carried out using the HemaVision®-28N multiplex RT-PCR Kit if applicable and according to the manufacturer’s instructions (DNA Diagnostic, Risskov, Denmark).
Array CGH
Array-based comparative genomic hybridization (array CGH) was performed using a 400 k SurePrint G3 Custom CGH Human Genome Microarray (e-Array design 84704, Agilent Technologies, Waldbronn, Germany) according to the manufacturer’s protocol by hybridizing 500 ng of patient DNA isolated from EDTA material against sex-matched human reference DNA (Megapool Reference DNA, Kreatech, Amsterdam, The Netherlands). Regions of genomic copy number change were analyzed and interpreted with Agilent Genomic Workbench 7.0 software under the following analysis setting: aberration detection algorithm ADM-2, threshold 6.0, aberration filter of at least four aberrant probes (resulting in an average genome-wide resolution of ~ 30 kb), and average log2 ratios of + 0.3 for gains and − 0.3 for losses.
The original description of the IKZF1plus gene deletion profile by Stanulla et al. has been based on MLPA analyses targeting IKZF1, CDKN2A, CDKN2B, PAX5, and CRLF2-P2RY8, as well as multiplex PCR for the detection of ERG deletions. Prior to this study, reproducibility of IKZF1plus profiles in our high-resolution array CGH was verified for selected cases (data not shown).
Cases indicative of masked hypodiploidy (i.e., presence of multiple tetrasomies) were analyzed on CGH+SNP array (e-Array design 85320, Agilent Technologies) using sex-matched reference DNA (Human Reference DNA, Agilent Technologies).
Discussion
This study aimed to evaluate the applicability of panel-based RNA-seq and array CGH for detecting relevant gene fusions as well as chromosomal deletions for risk stratification of pediatric BCP-ALL patients. This knowledge is intended to support optimization of the diagnostic workflow of the German study group of the international AIEOP-BFM ALL 2017 trial by reducing the number of necessary diagnostic tests, while assuring robustness of aberration detection. As discussed later, the here presented results conclusively show that panel-based RNA-seq and CGH+SNP array are powerful tools for the detection of fusion genes and genome-wide copy number aberrations. However, these techniques still need to be supported by FISH analysis in order to guarantee reliable detection of relevant fusions for all cases. The second and more broadly defined aim of this study was to examine the detection of genetic lesions beyond established markers, as recent publications have demonstrated that patients of the B-other subtype frequently harbor fusion transcripts of possible therapeutic relevance [
19‐
21]. Prospectively, successful integration of genome-wide CGH+SNP array and RNA-seq targeting almost 1385 transcripts into the AIEOP-BFM ALL 2017 diagnostic workflow could help to generate the data essential for revealing new genetic markers of pediatric ALL in the future.
The
IKZF1plus subtype itself is a good example for a new risk marker, which has been identified due to simultaneous consideration of multiple genetic loci. Notably, without analysis for
IKZF1plus, ten cases of this study would have been assigned as B-other and two cases as hyperdiploid ALL, with no further genetic risk stratification in both groups. One needs to be aware that occasional
IKZF1plus profiles might be present in cases not tested by array CGH in our workflow (e.g.,
ETV6-RUNX1). Projected onto the whole cohort of 117 BCP-ALLs, we therefore observed
IKZF1plus in at least 10% of the cases, which is higher than the originally published share of 6% by Stanulla et al. [
8]. Use of array CGH instead of MLPA and PCR-based analysis does not seem to be the reason for this discrepancy, as all deletions relevant for of
IKZF1plus (Tab.S4) would be detectable by MLPA analyses as described by Stanulla et al. Furthermore, the
ERG locus is covered by 37 array CGH probes, of which seven probes are located in the region assessed by
ERG deletion PCR in Stanulla et al. Despite the satisfactory probe coverage of
ERG in array CGH, the higher sensitivity of PCR for deletion detection needs to be acknowledged, which might influence
IKZF1plus detection. However, the high number of
IKZF1plus cases in our study is more likely to be explained by the small cohort size in our analysis as our extended array CGH data set on more than 600 cases comprises a lower number of
IKZF1plus within the B-other and hyperdiploid cases (data not shown). In line with Stanulla et al., the age of onset of
IKZF1plus cases was considerably higher than observed in the complete cohort (median age 8.7 vs 4.8 years, Table
1). It is important to mention that in the AIEOP-BFM ALL 2017 trial, the final risk stratification within
IKZF1plus is dependent on MRD-levels and not all cases with an
IKZF1plus deletion profile receive high-risk treatment. Within our cohort, nine of the twelve
IKZF1plus cases were MRD positive or inconclusive on treatment day 33 and therefore would be assigned to the high-risk treatment group of the AIEOP-BFM ALL 2017 trial.
In summary, array CGH analysis is a well-established method in many laboratories and in our study, its implementation in the diagnostic workflow was successful without difficulty. One important limitation of the method is the amount of leukemic blasts in BM aspirate obtained at the time of diagnosis, which should be at least 20% in order to reliably detect leukemia-associated genetic aberrations. On the other hand, CGH+SNP array proved to be more sensitive in detecting hyperdiploidy, which is in line with the previous observation that hyperdiploid leukemic cells frequently fail to proliferate in culture, thus resulting in an absence of aneuploidy in analyzed metaphases or deficiency of metaphases in general (Tab.S3 cases 83, 85, 90, 91, 94, 116) [
22]. Furthermore, array analysis identified other highly relevant genetic aberrations beyond
IKZF1plus, e.g., masked hypodiploidy and aberrations in line with fusions observed in RNA-seq analysis. For these reasons, copy number analysis using CGH+SNP arrays will be a central diagnostic tool in the AIEOP-BFM ALL 2017 trial to detect the current stratification relevant markers as well as to generate high-resolution genome-wide data for the definition of future genetic markers.
Detection of fusion genes using RNA-seq is a comparably new approach and not yet established in routine diagnostics for most malignancies. Our data demonstrate consistent results between RNA-seq and our routine diagnostic methods (karyogram, FISH, and RT PCR analyses). Considering the stratification relevant aberrations BCR-ABL1 and ETV6-RUNX1 as well as rearrangements involving either KMT2A or TCF3, which are relevant for stratification if resulting in an AFF1-KMT2A or TCF3-HLF fusion, only two discrepant results were observed (cases 26 and 29). For case 26, RNA-seq outperformed the routine FISH and RT-PCR methods by identifying an ETV6-RUNX1 fusion, which could later be confirmed with a second RT-PCR kit. The reason for the failure of the routine methods to detect this fusion is still unclear. Case 29, with a small insertion of KMT2A into the AFF1 gene, was the only example of a clear failure in fusion detection by RNA-seq. Even though present as multiple, identical reads at the RNA-seq raw data level, both applied alignment algorithms did not report the AFF1-KMT2A fusion, as they require the presence of at least three non-identical split reads for fusion calling. In our study, these two discrepant cases are the only two examples for false negative results. However, further influences exist, which can cause relevant fusions to remain undetected. These include a low percentage of leukemic blasts, loss of blasts during cell culture, cryptic or variant translocations, poor RNA quality, and notably the sensitivity of bioinformatics algorithms. Importantly, not all methods (karyogram, FISH, RT-PCR, and RNA-seq) are equally affected by these influences and a combination of methods would be able to rectify their respective weaknesses in most cases. Therefore, the diagnostic workflow of the German study group of AIEOP-BFM ALL 2017 trial will use FISH and panel-based RNA-seq as standard techniques for the detection of fusion genes. Karyotyping and RT-PCR will still be available but restricted to inconclusive cases only.
As a technical prerequisite, both alignment tools tested in this study will be applied in future RNA-seq analyses, as some fusions, like P2RY8-CRLF2, were detected by one of the algorithms only. Furthermore, the minimum number of sequencing reads per sample will be set to 3 million reads, even though our current data indicate that some fusion transcripts, like ETV6-RUNX1, are detectable at coverages below 2 million reads. However, a higher number of reads per sample will also facilitate detection of lowly expressed fusion genes as well as detection of fusions in samples with low blast counts. We detected no correlation between RNA-seq coverage and RIN value or ABL1 copy numbers in RT-PCR analysis (data not shown). Thus, variance in RNA-seq coverage in our study might rather be a result of differences in library preparation and quantification, which can possibly be addressed by standardization and automatization of RNA-seq protocols. Another important technical aspect of our study is that RNA-seq was conducted on RNA isolated from BM in EDTA tubes sent to our laboratory within 24 h. Our results indicate that use of RNA stabilizing solutions, at least for the fusion genes of interest, seems expendable.
It needs to be acknowledged that this workflow, despite panel-based RNA-seq and genome-wide array CGH, will not detect all aberrations present in the here defined B-other group. For example, fusions affecting the
IGH locus as well as translocations resulting in promoter or enhancer rearrangements (e.g., translocations of the
MYC or
MECOM locus) cannot be detected and these aberrations are being followed up on by FISH analysis on a research basis for selected cases. On the same note and in line with previous reports, our RNA-seq analysis detected additional fusion transcripts only in the B-other subtype, including cases 32–37, which all would have been deemed B-other until recently if not stratified as
IKZF1plus in our study [
20,
21]. Interestingly, the previously described “targetable” kinase fusions
EBF1-PDGFRB,
ETV6-NTRK3, and
TERF2-JAK2 were all observed within the
IKZF1plus subtype [
19,
23,
24]. Other detected fusions may provide prognostic information (
P2RY8-CRLF2,
MEF2D-BCL9) [
25‐
27]. Many fusion transcripts have already been described in pediatric ALL (e.g.,
HOOK3-FGFR1,
PAX5-DACH1,
EP300-ZNF384,
PAX5-
ETV6, or
ZNF618-
NUTM1), even though their potential therapeutic or prognostic impact remains to be addressed in more detail [
28‐
33]. To the best of our knowledge, some fusions identified in this study have not been described in the literature yet. A more detailed follow-up on the relevance of these fusions will be necessary, as our data on orientation and on in frame fusions of the coding sequences (CDS) gives only a first perception on potential functionality. Based on this limited data,
PAX5-MYO1G,
PAX5-NCOA6, and possibly
PAX5-LINC01251 seem to be promising new fusions, whose influence on the course of the disease warrants further consideration [
34,
35]. Just recently, the clinical implication of chimeric in frame
PAX5 fusions has been highlighted, as well as the potentially initiating, subtype-defining character of
PAX5 alterations in B-ALL [
33]. In this respect, particularly the fusions with
MYO1G and
NCOA6, first identified in this study, appear to leave the paired box DNA-binding domain of
PAX5 operational.
In conclusion, this study demonstrated that panel-based RNA-seq, supplemented by FISH analysis, and CGH+SNP arrays are a reliable combination of methods to detect the genetic markers BCR-ABL1, ETV6-RUNX1, hypodiploidy, IKZF1plus as well as rearrangements of KMT2A and TCF3 necessary for risk stratification in pediatric BCP-ALL. Foremost, acquisition of genome-wide copy number and panel-wide fusion data within the standard diagnostic workflow is expected to support identification of new genetic markers relevant for treatment in the future.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.