PAX5 Biallelic Genomic Alterations Define a Novel Subgroup of B-Cell

Leukemia https://doi.org/10.1038/s41375-019-0430-z

ARTICLE

Acute lymphoblastic leukemia PAX5 biallelic genomic alterations deﬁne a novel subgroup of B-cell precursor acute lymphoblastic leukemia

1,2,3,4,5 1 3,4,6 1 1 Lorenz Bastian ● Michael P. Schroeder ● Cornelia Eckert ● Cornelia Schlee ● Jutta Ortiz Tanchez ● 1 2,7,8 1 1 Sebastian Kämpf ● Dimitrios L. Wagner ● Veronika Schulze ● Konstandina Isaakidis ● 1,3,4 5 1,3,4 9 1 Juan Lázaro-Navarro ● Sonja Hänzelmann ● Alva Rani James ● Arif Ekici ● Thomas Burmeister ● 1 10 11 3,4,12 13 Stefan Schwartz ● Martin Schrappe ● Martin Horstmann ● Sebastian Vosberg ● Stefan Krebs ● 13 14,15 3,4,12 3,4,16 5 Helmut Blum ● Jochen Hecht ● Philipp A. Greif ● Michael A. Rieger ● Monika Brüggemann ● 3,4,16 1,3,4,5 1,2,3,4,5 Nicola Gökbuget ● Martin Neumann ● Claudia D. Baldus

Received: 6 September 2018 / Revised: 29 January 2019 / Accepted: 4 February 2019 © Springer Nature Limited 2019

Abstract Chromosomal rearrangements and specific aneuploidy patterns are initiating events and define subgroups in B-cell precursor acute lymphoblastic leukemia (BCP-ALL). Here we analyzed 250 BCP-ALL cases and identified a novel subgroup (‘PAX5- ’ n = fi

1234567890();,: 1234567890();,: plus , 19) by distinct DNA methylation and gene expression pro les. All patients in this subgroup harbored mutations in the B-lineage transcription factor PAX5, with p.P80R as hotspot. Mutations either affected two independent codons, consistent with compound heterozygosity, or suffered LOH predominantly through chromosome 9p aberrations. These biallelic events resulted in disruption of PAX5 transcriptional programs regulating B-cell differentiation and tumor suppressor functions. Homozygous CDKN2A/B deletions and RAS-activating hotspot mutations were highly enriched as cooperating events in the genomic profile of PAX5-plus ALL. Together, this defined a specific pattern of triple alterations, exclusive to the novel subgroup. PAX5-plus ALL was observed in pediatric and adult patients. Although restricted by the limited sample size, a tendency for more favorable clinical outcome was observed, with 10 of 12 adult PAX5-plus patients achieving long-term survival. PAX5-plus represents the first BCP-ALL subgroup defined by sequence alterations in contrast to gross chromosomal events and exemplifies how deregulated differentiation (PAX5), impaired cell cycle control (CDKN2A/B) and sustained proliferative signaling (RAS) cooperatively drive leukemogenesis.

Introduction have been identified in pre-leukemic clones [4]characterizing chromosomal translocations or ploidy disorders Gross chromosomal events, such as specific aneuploidy as hallmark initiating events in leukemogenesis [5]. patterns or oncogenic gene fusions induced by chromo- A subset of these oncogenic gene fusions are already somal translocations, define molecular subgroups in B-cell targets for therapeutic interventions (e.g., ABL1-class precursor acute lymphoblastic leukemia (BCP-ALL) [1]. fusions) [1]. They inform biological programs as observed by Further steps of leukemic transformation include dele- distinct gene expression [2] and DNA methylation profiles tions of hematopoietic transcription factors (IKZF1, PAX5, [3], show an age-dependent prevalence pattern and and ETV6), which are co-occurring aberrations in 40% of impact treatment outcomes [1]. Some of these aberrations BCP-ALL cases, indicating that impairment of hematopoietic differentiation promotes BCP-ALL development [6]. PAX5 has been characterized as haploinsufficient tumor Supplementary information The online version of this article (https:// suppressor in this context, both in BCP-ALL patient sam- doi.org/10.1038/s41375-019-0430-z) contains supplementary material, ples [6, 7] and mouse models [8–10]. Additional mutations which is available to authorized users. in genes, such as cell cycle regulators (CDKN2A/B) and * Claudia D. Baldus RAS/MAPK regulators (NRAS, KRAS, and NF1), are [email protected] acquired during leukemic development and contribute to an Extended author information available on the last page of the article individual disease phenotype [5]. L. Bastian et al.

Up to 30% of BCP-ALL patients, especially at adult age, Variant calling and differential gene expression lack described initiating lesions, thus precluding the allo- analysis cation to molecular subgroups with specific functional and clinical characteristics and hampering the development of All sequences were aligned to the human genome build targeted therapeutic strategies. GRCh37.75 using the bcbio-nextgen pipeline v0.9.1a- To identify novel driver candidates and to characterize 7da8dce (https://github.com/bcbio/bcbio-nextgen) for WES their functional context, we analyzed 250 BCP-ALL first and STAR-aligner for RNA-Seq data [11]. Somatic muta- diagnosis samples across age groups for gene fusions and tions were detected in WES and corresponding target gene gene expression profiles, DNA methylation profiles, copy panel sequencing using bcbio-nextgen employing Mutect number alterations, single nucleotide variants (SNVs), and [12], Freebayes [13], VarDict [14], and VarScan [15]. small insertions/deletions (Indels). Integration of these Mutations were required to be reported by two of the four multi-omics data enabled us to identify a novel BCP-ALL callers with a variant allele frequency (VAF) of >10%, a subgroup defined by PAX5 sequence variants with different read depth of at least 10 with at least three reads supporting modes of inactivation of the second allele, CDKN2A/B the mutation and a VAF below 5% in the remission control. deletions, and RAS-activating mutations. For the 206-gene target panel, mutations were detected by Varscan [15] or Platypus [16]. Ensemble Variant Effect predictor [17] was used to annotate variants. Mutations Patients and methods were required to have a read depth of at least 30 and a VAF > 10%. Population variants were excluded based on dbSNP Patient materials and standard diagnostics 135, COSMIC v75 and gnomAD annotations. Gene panel sequencing achieved an average target coverage of >300×. Bone marrow or peripheral blood samples at first diagnosis Only mutations confirmed by gene panel sequencing were of BCP-ALL were obtained from adult and pediatric patients reported and included in the manuscript. Gene fusions were enrolled into trials of population-based German study detected from RNAseq data using deFuse v0.7.0 [18] and cohorts (GMALL, AIOP-BFM, and COALL). Patients and/ FusionCatcher v0.99.7c [19]. Expression quantification was or their families gave informed consent for biological obtained with Stringtie [20]. Differential expression analy- research on their banked samples. Studies were approved by sis was performed by ANOVA based on TPM. GSEA the institutional ethic review boards of the study groups (v3.0) software [21] was used in preranked mode for gene and the institutional ethic review board of the Charité - set enrichment analysis. For input, ranked gene lists were Universitätsmedizin, Berlin, Germany. Bone marrow sam- obtained by median log-fold change between each mole- ples obtained during remission (minimal residual disease cular subgroup and all other samples of the cohort. level < 10−2) were used as germline control. Further sample processing and clinical standard procedures are detailed in Copy number analyses Supplementary patient materials and methods. DNA copy number variations (CNV) were assessed by Sequencing analyses DNA hybridization to SNP-Arrays (CytoScan HD Array; Life Technologies, Carlsbad, CA) and analysis using the For whole-exome sequencing (WES) three samples per lane Chromosome Analysis Suite (ChAS; Life Technologies, were processed on an Illumina HiSeq4000 platform using Carlsbad, CA). Virtual karyotypes were derived from WES Low input Exome-Seq Human v5 + UTRs (Agilent, Santa data using CopywriteR [22] to calculate log-ratio counts in Clara, California) with an average coverage of 141.6 Mio bins of 20–50 kb between diagnosis and remission control. mapped reads/sample (MMRS). Panel sequencing was CNVkit [23] was used to call copy number variations from performed using one of two customized biotinylated RNA CopywriteR output. Copy number status in nine regions oligo pools (both SureSelect, Agilent, Santa Clara, Cali- most frequently affected by CNVs in BCP-ALL were fornia). For samples with WES data, target regions were assessed by Multiplex Ligation dependent Probe Amplifi- selected to validate mutations form WES (362 kbp target cation (MLPA) using the SALSA MLPA P335 ALL-IKZF1 size). For the remaining samples, target genes were selected probemix and Coffalyser Software (both: MRC Holland, to cover genes with central functions in normal or malignant Amsterdam, The Netherlands). hematopoiesis and frequent mutations in leukemia (Sup- plementary Table S1; 206 target genes). Panel sequencing DNA methylation profiling was performed on HiSeq2000 or HiSeq1500 platforms. For RNA sequencing (RNA-Seq), six samples per lane were DNA methylation status was obtained by DNA methylation sequenced with an average of 64 MMRS. chip array (Infinium Human Methylation 450 K or EPIC PAX5 biallelic genomic alterations define a novel subgroup of B-cell precursor acute lymphoblastic. . .

BeadChip, Illumina; San Diego, CA) and analyzed within fusion (n = 19), Near haploid - High Hyperdiploid (NH- Genomics Suite (Partek, St. Louis, USA) using SWAN HeH, n = 16), Low Hypodiploid - Near Triploid (LH–NT, normalization to obtain corrected beta values. Differentially n = 14), KMT2A fusions (n = 14), TCF3-PBX1 (n = 6), methylated CpGs were analyzed by one-way ANOVA-tests. MEF2D fusions (n = 5), ETV6-RUNX1 (n = 4), PAX5- ETV6 (n = 4), CEBP transcription factor family fusion (n = Statistical analysis 5), and IGH-MYC/IGH-BCL2 (n = 2). Within the remaining samples (n = 39), we identified a novel cluster including 19 Subgroup-specific differences in gene expression were patients (RNA-Seq: n = 18, one additional sample identified analyzed by employing one-way ANOVA-tests and by DNA methylation) by a shared gene expression profile Benjamini–Hochberg correction for multiple testing. distinct to established molecular subgroups (‘Novel cluster’, Unsupervised clustering was done using eucledian distance Fig. 1). t-SNE analysis confirmed this cluster as distinct with average linkage. Group differences in immunological subgroup (Supplementary Figure S1). surface marker expression where analyzed by Mann– Whitney U test. Fisherʼs exact test was used to analyze the The novel subgroup consistently harbors PAX5 association of molecular and clinical features with sub- sequence mutations groups. All statistical tests were two-tailed. Clustering analysis, ANOVA statistics and heatmap visualization was All 19 patients of the novel cluster harbored sequence performed using Gitools [24]. variants in the B lymphoid transcription factor PAX5,as observed by WES and/or targeted gene panel sequencing Primary data (Fig. 2a, Table 1). PAX5 is a member of the paired box family of transcription factors, sharing a paired box Genome data has been deposited at the European DNA-binding domain, which was affected by mutations Genome-phenome Archive (EGA) under accession number in all 19 cases. A proline to arginine change at codon EGAS00001003209. 80 was present as hotspot in 16/19 cases (Fig. 2b) and was exclusive for this patient cluster (p < 10E-5, Fisher’s exact test). PAX5 sequence mutations were somatic as Results analyzed by WES together with matched germline control in eight out of eight patients studied. Protein integrity Unsupervised gene expression analysis identifies a scores (SWIFT, Polyphen, Condel) indicated a deleterious novel BCP-ALL subgroup effect on PAX5 protein integrity for all these sequence variants. 3D modelling based on crystallographic analyses We studied a total of 250 BCP-ALL patients to dissect revealed that the two PAX5 mutations recurrent in molecular subgroups, and performed unsupervised hier- the subgroup (p.P80R, p.R38H) affected amino acid archical clustering of top variably expressed protein coding residues in close spatial proximity after protein folding genes (n = 800; 4%) from transcriptome sequencing of 208 (Supplementary Figure S2) suggesting a common func- patient samples with available high-quality RNA (Fig. 1). tional consequence. Unsupervised hierarchical clustering revealed 14 distinct Variant allele frequencies of PAX5 mutations were >0.6 clusters. In 11 clusters a known founding alteration [1] was in 13 of the 19 patients. Three of the remaining patients identified by gene fusion calling from transcriptome harbored PAX5 sequence variants in two independent sequencing or digital karyotypes from whole exome positions, together indicating loss of heterozygosity or sequencing (WES) and SNP-Arrays. Patients without suf- acquisition of compound heterozygosity as frequent char- ficient mRNA (n = 42) for RNAseq were assigned to acteristics of this subgroup. subgroups either by sharing a subgroup-specificDNA- PAX5 sequence variants were also identified in 14 methylation signature (n = 12) or presence of the BCR- patients of the remaining cohort (n = 144; Fig. 2a, b; Sup- ABL1 or KMT2A-AFF1 gene fusions as determined by plementary Tables S3–S5). Yet, 12 of the 14 cases were break-point specific qRT-PCR (n = 30, Supplementary assigned to established molecular subgroups based on the Table S2). Two novel transcriptional clusters were defined presence of a founding lesion: these PAX5 sequence var- by previously described fusions involving the CEBP tran- iants occurred most prominently in Ph-like cases (n = 7/48 scription factor family and co-occurrence of IGH-MYC/ Ph-like) and specifically in combination with CRLF2- IGH-BCL2 fusions. Thus, we could assign 211 patients fusions (n = 7/23 CRLF2-fusion; one patient without the of the entire BCP-ALL cohort (n = 250) to 13 subgroups corresponding Ph-like gene expression profile defined as with described founding alterations: Ph-like (n = 47), ‘unknown’), characterizing PAX5 alterations as cooperating DUX4 fusion (n = 44), Ph-positive (n = 31), ZNF384 event in these cases. L. Bastian et al. PAX5 biallelic genomic alterations define a novel subgroup of B-cell precursor acute lymphoblastic. . .

Fig. 1 A novel BCP-ALL subgroup is identified by unsupervised clus- PAX5-plus group. One additional PAX5-plus patient har- tering of top variably expressed genes. Gene expression profiles were bored a hotspot mutation in the FLT3 tyrosine kinase obtained by transcriptome sequencing of BCP-ALL samples obtained at domain (p.N676S), indicating another mode of JAK/STAT first diagnosis (n = 208). Unsupervised clustering (Euclidean distance, Average Linkage) of top 800 variable expressed protein coding genes is pathway activation. It seems possible that these events shown together with the spanning read count of subgroup-defining gene could further enforce RAS activation in PAX5-plus patients fusions. Aneuploid subgroups were identified by virtual karyotypes or that they can substitute for absent RAS mutations in two derived from WES or SNP-Arrays (detailed in Fig. 3). Samples without PAX5-plus patients. No other gene harbored significantly available virtual karyotypes and a gene expression profile specificforthe corresponding aneuploid subgroup were classified accordingly. The enriched sequence mutations in PAX5-plus patients. following definition was used for age groups: child (1–15 years), adolescent (16–20 years), young adult (21–39 years), adult (40–59 years), Chromosome 9 alterations promote loss of + t fi elderly (60 years). -SNE analysis con rmed the cluster separation and heterozygosity for PAX5 sequence variants independence of the novel subgroup (Supplementary Figure S1)

To investigate underlying chromosomal mechanisms for PAX5 sequence mutations, CDKN2A/B deletions, and PAX5 loss of heterozygosity indicated by variant allelic fre- RAS-activating mutations define PAX5-plus BCP-ALL quencies >0.6, we analyzed virtual karyotypes obtained from WES and SNP-Arrays (Fig. 3;Table1). MLPA analysis was In all but one of the 19 patients from the novel subgroup, the used to deduce structural chromosome 9 alterations in cases, INK4-ARF locus including the cell cycle regulator genes where no complete virtual karyotype was available (n = 4; CDKN2A and CDKN2B wasaffectedbycopynumberlosses Table 1). These analyses identified chromosome 9p deletions (Fig. 2a, Table 1). CDKN2A/B deletions were homozygous in (n = 9) or uniparental disomy of chromosome 9 (n = 2) or 9p 14 of the 19 patients. One of three samples with a hemizygous (n = 1) together affecting 12 of 19 PAX5-plus patients (63%, CDKN2A deletion harbored a disruptive frameshift mutation Table 1)ascomparedto6/31patientswithgrosschromosome on the second CDKN2A allele, while another one also har- 9 aberrations in the remaining cohort (19%, p = 0.003), bored a RB1 deletion. Together, the CDKN2A/B—RB1— excluding aneuploid cases. In PAX5-plus patients without E2F cell cycle control axis was impaired in 18 of the 19 gross chromosome 9 alterations (n = 7), disruption of the patients (95%) of this novel subgroup in contrast, to the second PAX5 allele occurred through a focal PAX5 deletion remaining cohort were 97 of 231 patients were affected in a (n = 1) or a PAX5-disruptive chromosomal translocation (n = similar way (42%, p <0.001;Fig.2a). 1; t(9;22)(p13;q13)). Four patients harbored PAX5 sequence Furthermore, an enrichment of activating hotspot muta- variants in independent positions, suggesting compound het- tions in the proto-oncogenes NRAS (n = 9) or KRAS (n = 6) erozygosity. Together, we observed mechanisms for biallelic was identified in the novel PAX5 subgroup (Fig. 2a, Table 1). PAX5 events in 18 of 19 PAX5-plus patients. One patient without NRAS/KRAS mutations showed a dele- Additional recurrent chromosomal events in PAX5-plus tion of NF1, a key negative regulator of RAS signaling. Thus, patients were chromosome 7 deletions and deletions of the vast majority of patients (84%) harbored an activating chromosome 20q each affecting three of 14 patients with alteration in RAS pathway regulators. In the evaluable available virtual karyotypes (Fig. 3). Deletions of 20q were patients of the remaining cohort (n = 144), NRAS/KRAS associated with 9p deletions, including PAX5, in all three mutations were seen in only 31 cases (22%, p <0.001). cases. Analysis of the corresponding break point suggested Together, the combinational pattern of disruptive PAX5 the presence of a dicentric chromosome 9;20 (dic (9;20)). In sequence variants (100%), predominantly homozygous line with previous reports of this rare cytogenetic finding CDKN2A/B deletions (95%), and RAS pathway activating [26], RNA-Seq analysis provided no evidence of a func- mutations (84%)—in the absence of established founding tional fusion transcript. Application of the PAX5-plus gene alterations—represents a highly specific pattern of triple expression signature on a large published cohort of aneu- alterations exclusive for the novel subgroup (p < 0.001, ploid BCP-ALL [27], identified three cases with dic(9;20) Supplementary Figure S3), therefore further on named and a PAX5-plus mutation profile (Supplementary Fig- ‘PAX5-plus’. ure S4), supporting the hypothesis that dic(9;20) is not a Analysis of additional mutations enriched in PAX5-plus driver alteration by itself but represents one of different patients identified in 4/19 patients (21%) frameshift muta- means of 9p inactivation in PAX5-plus leukemogenesis. tions in the IL-7 receptor (IL7R) transmembrane domain (Fig. 2a, c) as compared to 2/142 patients (2%, p = 0.0019) DNA methylation profiling confirms distinct PAX5- of the remaining cohort. These alterations have been plus epigenetic phenotype described to drive BCP-ALL via JAK/STAT activation and were associated with increased CRLF2 expression levels To assess the epigenetic profile of PAX5-plus ALL, we [25], which we confirm in the IL7R mutated patients of the performed DNA methylation profiling on 91 samples L. Bastian et al.

Fig. 2 The novel subgroup is defined by a distinct profile of sequence PAX5 sequence mutations, CDKN2A deletions and RAS-activating alterations. a Sequence alterations (SNVs / Indels) were obtained from mutations in the absence of established founding lesions. Asterix (*) WES using MRD-negative remission samples as germline controls and indicates an NF1 deletion identified by SNP-Array. b Gene schematic targeted gene panel re-sequencing (n = 57) or target panel sequencing of PAX5 depicting the functional protein domains and all PAX5 of 206 genes involved in normal and malignant hematopoiesis alone sequence mutations identified. Dots represent the frequency of (n = 104). CNVs in nine genomic regions most frequently affected by alterations and colors indicate the molecular subgroups. Mutation copy number alterations in BCP-ALL were analyzed by WES (n = 49) calling for two patients (Berlin_S63507, BerlinLB6600) was based on or MLPA (n = 104) using the SALSA MLPA P335 ALL-IKZF1 RNA-Seq alone due to the absence of sufficient amounts of high- probemix or both (n = 8). The most frequently occurring alterations quality DNA for gene panel sequencing. These patients were not affecting genes within four major functional groups are shown. The included in (a). c Gene schematic of IL7R, depicting the protein novel subgroup (PAX5-plus ALL) is defined by an exclusive profile of domains and motifs of all IL7R sequence mutations identified PAX5 biallelic genomic alterations define a novel subgroup of B-cell precursor acute lymphoblastic. . .

Table 1 Overview of subgroup-speciﬁc genomic alterations in the PAX5-plus cohort (n = 19)

VAF variant allele frequency, UPD uniparental disomy, del deletion from our cohort, including PAX5-plus samples (n = 5). GMALL cohorts [29]. Although the analysis of MRD Unsupervised hierarchical clustering of top 1% (n = 8400) response in molecular subgroups was limited by low patient variably methylated CpGs identified PAX5-plus samples by numbers, a tendency for lower molecular response rate was an exclusive cluster, which was confirmed by t-SNE ana- seen in Ph-like (n = 8/18 molecular remissions; 44%) and lysis (Fig. 4, Supplementary Figure S5, Supplementary ZNF384-rearrangend cases (n = 7/14, 50%), confirming pre- Table S6). A total of 34436 CpGs representing 4.1% of all vious reports [30, 31]. In contrast, MRD negativity was CpGs represented on the array were differentially methy- achieved in the majority of PAX5-plus patients (n = 7/10; lated in PAX5-plus patients (FDR < 0.05), confirming the 70%) and DUX4-rearrangend cases (n = 11/13; 85%). The independent regulation of PAX5-plus ALL on the epige- limited sample size restricts definite conclusions on the netic level. Moreover, 12 samples with insufficient RNA overall survival of theses molecular subgroups. Nevertheless, quality for RNAseq could be assigned to molecular sub- the more favorable response to induction treatment of PAX5- groups due to shared DNA methylation profiles. plus patients (n = 10) translated also to a tendency of increased 10 years overall survival probability (80% (±13%)) PAX5-plus patients have a rather favorable compared to 61% (±5%), for patients in the remaining cohort treatment outcome and a characteristic (n = 118; p = 0.45, Supplementary Figure S6, Supplementary immunophenotype Table S7). This trend was also observed when PAX5-plus patients were compared to patients of the remaining large PAX5-plus ALL occurred at all age groups with a preference molecular subgroups (Supplementary Figure S6). for adolescents/young adults (median: 21 years, range: 9–64 PCR-based MRD analyses use immunoglobulin rear- years, Table 1, Supplementary Table S7) and with an equal rangements as clonal marker. Complete absence of PAX5 sex distribution (male: 11 / female: 8). A total of 126 adult function has been found to heavily skew the immunoglo- patients treated on GMALL protocols with age-adapted bulin rearrangement repertoire in murine pro-B cells intensity (GMALL 06/99, 07/2003, 08/2013, Elderly through exclusion of the largest and most distal VHJ gene 1/2003) were evaluable for outcome analyses. Of these 98% family (VHJ588) from rearrangements [32]. In PAX5-plus (n = 123) achieved complete hematologic remission after patients, we observed an overall VH gene family usage induction therapy and 103 of the 123 patients were evaluable comparable to large BCP-ALL cohorts (Supplementary for response evaluation by minimal residual disease (MRD) Tables S8, S9) indicating that this non-transcriptional PAX5 assessment using patient-specific PCR markers for immu- function was retained despite biallelic PAX5 alterations. noglobulin/T-cell receptor gene rearrangements [28]. Overall, With the exception of one pro-B ALL, all PAX5-plus 68% of patients (n = 70/103, Fig. 5a) achieved a complete patients showed a common ALL immunophenotype molecular remission, which is in the range of representative as identified by immunophenotyping of evaluable patients L. Bastian et al.

Fig. 3 Chromosome 9 aberrations promote homozygosity for sequence if these alterations affected at least the major part of a chromosome alterations in PAX5-plus ALL. Virtual karyotypes were obtained from arm or whole chromosomes. Copy number changes are represented WES (n = 47), SNP-Arrays (n = 12), or both (n = 7). Shown are gross as fold changes from the normal diploid (0.5 - hemizygous deletion, chromosomal gains and losses as well as uniparental di- and trisomies 1.0 - normal diploid, 1.5 - trisomy, 2.0 - tetrasomy, 4.0 - octasomy)

(n = 193; Supplementary Figure S7A, B). Absence of Adefinition of ‘common ALL immunophenotype with CD34 intracellular IgM µ heavy chain expression (pre-B ALL) in ≤85% and CD52 ≤70% surface marker expression’ identified PAX5-plus patients is consistent with an impaired or 13 of 14 PAX5-plus samples from the entire cohort (n = 173 absent IgM expression described in PAX5-mutated B-cell with CD34 and CD52 measurements), while only 12 of 159 models [6]. Thus, PAX5 sequence alterations in PAX5- non–PAX5-plus samples met these criteria (Supplementary plus patients retained the ability to perform immunoglo- Figure S7C, p < 0.0001) confirming a distinguishing PAX5- bulin rearrangements but most likely impair the assembly plus profile also on the immunophenotypic level. of IgM µ heavy chains. Comparison of PAX5-plus to common ALL samples from Transcriptional regulation of B-cell differentiation, the other major subgroups indicated a similar proportion of proliferative control, and energy metabolism is blasts (CD19 positive; Fig. 5b, Supplementary Figure S7B), perturbed in PAX5-plus patients while PAX5-plus patients had a markedly reduced surface marker expression of CD34 (median: 39% vs. 88.5%; p = The hematopoietic transcription factor PAX5 is a critical 0.0024) and CD52 (median: 46% vs. 91%; p = 0.0002). regulator of development and maintenance of B-cell identity PAX5 biallelic genomic alterations define a novel subgroup of B-cell precursor acute lymphoblastic. . .

Established Subgroup PAX5 -plus Novel Subgroup Immunophenotype Age Group Sex

Patients (n=91)

DNA methylation Immunophenotype: common pre-B pro-B 0 0.5 1 Age Group: Childhood Adolescent Young Adult Adult Elderly Sex: Female Male

Subgroups: Ph-like ZNF384 KMT2A DUX4 NH - HeH MEF2D Ph-positive LH - NT unknown

Fig. 4 PAX5-plus ALL has a distinct DNA methylation profile. The (n = 8400) variable methylated CpGs identified PAX5-plus samples DNA methylation status of 850.000 CpGs was analyzed after hybri- (n = 5) by a cluster exclusive to this subgroup. t-SNE analysis con- dization of DNA from first diagnosis samples of the cohort (n = 91) firmed the cluster separation and independence of PAX5-plus sub- to Infinium MethylationEPIC BeadChips. Unsupervised hierarchical group (Supplementary Figure S5) clustering (Euclidean distance, Average Linkage) of the top 1% L. Bastian et al.

a MRD response to induction therapy 100

80 molecular failure complete molecular remission 60

40 patients (%) 20

DUX4 (n=13) Ph-like (n=18)ZNF384 (n=14) PAX5-plus (n=10) overall total (n=103)

b CD19 CD34 CD52 n.s. ** *** 100 100 100 ) 80 80 80

60 60 60

40 40 40

positive cells (%) 20 positive cells (% 20 positive cells (%) 20

0 0 0

PAX5-plus PAX5-plus PAX5-plus (n=15) (n=15) (n=13) other subgroups other subgroups other subgroups (n=104) (n=104) (n=93)

Fig. 5 PAX5-plus ALL has a standard MRD response to induction patients (n = 193; Supplementary Figure S7) was assessed by flow therapy and a distinct immunophenotype. a Outcome-available adult cytometry during routine immunophenotyping at BCP-ALL first patients received induction chemotherapy with an age adapted inten- diagnosis in the central GMALL reference laboratory. Expression of sity on subsequent GMALL protocols (n = 103). Minimal residual CD19, CD34, and CD52 is shown exclusively in samples with com- disease was analyzed by qRT-PCR of clone-specific markers after mon ALL immunophenotype in order to compare evaluable PAX5- completion of induction treatment. Shown are the proportions of plus common ALL samples (n = 15) to the common ALL samples of patients with complete molecular response (no MRD detectable with a the nine largest subgroups of the remaining cohort (n = 104; Supple- sensitivity of at least 10–04) or molecular failure (MRD level > 10−04) mentary Figure S7). Statistical analysis was performed by Mann– in the four largest molecular subgroups for this analysis and the total Whitney U test. n.s. not significant; **p < 0.01; ***p < 0.001) evaluable cohort. b Surface marker expression in adult BCP-ALL through induction of lineage specific and suppression development [9]. Pax5 re-expression in this model acti- of lineage inappropriate gene expression. Analysis of vated a transcriptional program reminiscent of normal PAX5-plus ALL gene expression profiles showed differ- B-cell differentiation and induced BCP-ALL regression. ential mRNA expression of 1555 protein coding genes (up: Genes downregulated upon Pax5 re-expression in this 483/down: 1072, FDR < 0.05, Fig. 6a; Supplementary model were highly upregulated in PAX5-plus patients, Table S10). Among these, known targets of PAX5 tran- while Pax5-induced genes in the model were downregulated scriptional activation in B cells [33] were downregulated in PAX5-plus patients (Fig. 6c). This provides a cross- (e.g., VPREB3, CD72) and targets of PAX5 transcriptional species validation of a PAX5 tumor suppressor function in repression were upregulated (e.g., ITGB7, TLR7). Gene set BCP-ALL and indicates that this tumor suppressive role enrichment analysis (GSEA) confirmed in PAX5-plus was impaired in PAX5-plus patients. patients downregulation of PAX5 target genes activated For an unrestricted exploration of oncogenic pathways during the early steps of B-cell lymphopoiesis [33] (Fig. 6b) involved in PAX5-plus ALL, we performed GSEA for indicating a PAX5 loss-of-function phenotype in PAX5- the MSigDB Hallmark [34] and KEGG pathway [35] plus patients. gene sets comparing each molecular BCP-ALL subgroup PAX5 has been characterized as tumor suppressor in a to the background of the remaining cohort. Gene sets mouse model, where constitutive Stat5 activation and RNAi with the highest enrichment in PAX5-plus samples knockdown of Pax5 cooperatively promote BCP-ALL (FDR < 0.05; NES > 2.0) included the proto-oncogene PAX5 biallelic genomic alterations define a novel subgroup of B-cell precursor acute lymphoblastic. . .

MYC-E2F signaling axis as well as DNA replication/ model [9], which had also identiﬁed Myc activation and cell cycle progression and oxidative phosphorylation upregulation of DNA replication and cell cycle regulation (Fig. 6d). These ﬁndings paralleled pathway analyses of genes as consequence of a lost PAX5 tumor suppressor the murine Stat5 activation/Pax5 knockdown BCP-ALL function. L. Bastian et al.

Fig. 6 PAX5 transcriptional regulation of B-cell differentiation and more similar to the P80R PAX5-plus cases than to the energy metabolism is perturbed in PAX5-plus patients. a The tran- PAX5-mutated cases of the remaining cohort, supporting a scriptional regulation present in PAX5-plus ALL was analyzed by common functional consequence of PAX5 mutations in RNAseq. Genes differentially expressed (FDR < 0.05; log2-fold change > 3.0) in PAX5-plus patients were obtained by one-way PAX5-plus ALL. ANOVA and Benjamini–Hochberg correction comparing PAX5-plus In all but one PAX5-plus patient, inactivation of the patients (n = 18) to the remaining cohort (n = 190). Hierarchical second PAX5 allele could be deduced from a second clustering of genes and patient samples was performed using Eucle- PAX5 b–d disrupting sequence mutation or was observed dian distance and Average linkage. Gene set enrichment analysis PAX5 of transcriptional profiles obtained from PAX5-plus patients and the 10 through focal deletions involving the locus or a largest other subgroups of the cohort. For application in GSEAPrer- variety of chromosomal aberrations resulting in chromo- anked (v1), protein coding genes were ranked according to the log2- some 9p loss of heterozygosity. This observation parallels fold difference of expression between each subgroup and the hetero- reports of rare cases of familial BCP-ALL susceptibility geneous background of the remaining cohort. Gene sets were derived from analysis of b PAX5 target genes during normal hematopoiesis where heterozygous PAX5 sequence variants (p.G183S) [33], c genes differentially expressed after PAX5 restoration in a were passed on as germline variants, while individuals murine BCP-ALL model driven by constitutive STAT5 activation and affected by BCP-ALL had acquired sporadic 9p losses d PAX5 knockdown [9], and MSigDB [34] Hallmark and KEGG retaining only the mutated PAX5 allele [38, 39]. In pathway [35] gene sets with the strongest enrichment in PAX5-plus ALL. Human genes were mapped to murine orthologues using Bio- PAX5-plus patients of our cohort, sequencing of germline Mart for the analysis of gene sets in b and c, which were originally controls indicated a somatic origin of the defining derived from mouse models. NES normalized enrichment score, FDR sequence alterations. Further evidence for a leukemogenic false discovery rate role of biallelic PAX5 alterations comes from Pax5 haploinsufficient mouse models, which spontaneously PAX5 has been identified as metabolic gatekeeper acquired sequence mutations in the Pax5 paired domain of towards malignant transformation through glucose and the retained allele upon viral or chemical carcinogenesis energy restriction [7]. Analyzing related gene expression in [10] or upon infection exposure [40]. Notably, addition- PAX5-plus patients, we observed upregulation of enzymes ally acquired alterations in these models were similar to with central functions in citrate cycle (MDH1, MDH2)as the mutation pattern in PAX5-plus patients, including well as mitochondrial oxidative phosphorylation (succinate- JAK/STAT-/RAS-/MAPK-activating sequence mutations ubiquinone oxidoreductase complex: SDHA, SDHB, SDHD, and Cdkn2a deletions. and cytochrome c oxidase complex (COX8A, COX5A, Most recently, other large-scale transcriptome analyses COX5B, and COX6A1). These findings indicate an upre- [36, 41–43] have also established PAX5 alterations as gulation of catabolic energy metabolism in PAX5-plus hallmark of novel BCP-ALL subgroups. All studies patients, which was confirmed by GSEA of the corre- observed a specific subgroup-defining gene expression pat- sponding pathways (Fig. 6d). tern in PAX5 P80R mutated cases, while the largest analysis [36] pointed out that samples with other homozygous or compound heterozygous PAX5 sequence mutations had a Discussion distinct transcriptional profile, which was shared with cases harboring PAX5 gene fusions (PAX5alt). Consistently The highly exclusive mutation pattern and distinct gene with our analysis, CDKN2A deletions and RAS-activating expression, DNA methylation, and immunophenotype pro- mutations were highly enriched both, in the described files, as well as shared karyotype aberrations characterize novel P80R mutated subgroups, and the ‘PAX5alt’ PAX5-plus ALL as a novel and unique biological entity subgroup. The high frequency of this defined pattern within the landscape of BCP-ALL subgroups. All PAX5- of cooperating events argues in favor of a shared functional plus patients harbored PAX5 sequence variants including p. basis of PAX5-driven leukemias, while individual PAX5 P80R (n = 16/19 PAX5-plus), p.R38H (n = 2/19) as well as mutations - as also previously described [6, 36, 37] - result in including one patient with p.T75I (n = 1/19) amino acid heterogeneous transcriptional regulations. Based on our change. Functional studies have highlighted that different data, we have proposed the definition of ‘PAX5-plus’ ALL, single amino acid changes in the PAX5 DNA-binding emphasizing this functional context. Nevertheless, it domain show a heterogeneous impact ranging from slightly remains to be established, if there is a functional equivalent reduced to completely abolished PAX5 function [6, 36, 37]. of absent CDKN2A- or RAS-alterations in some cases or if Thus, PAX5 p.R38H/p.R140L or p.T75I mutations in these represent a different type of regulation. Larger pro- PAX5-plus patients could have specific functional con- spective studies are needed to determine how different sequences, which are distinct to other PAX5 sequence PAX5 alterations in PAX5-driven ALL translate into dis- mutations of the remaining subgroups. In these lines, gene tinct clinical phenotypes and to substantiate a subgroup expression profiles of non-P80R PAX5-plus cases were definition for further clinical application. PAX5 biallelic genomic alterations define a novel subgroup of B-cell precursor acute lymphoblastic. . .

Biallelic events as observed in PAX5-plus patients CDKN2A loss-of-function and MEK inhibitors to target represent a typical case of the Knudson two-hit hypothesis RAS activation. Both inhibitor types have shown pre- [44], consistent with a loss-of-function phenotype. In line clinical activity in BCP-ALL [47, 49]. with this, reports on functional modeling of PAX5 mutations Although limited by the sample size, our data indicates a recurrent in PAX5-plus patients had shown previously higher incidence of PAX5-plus ALL in adolescents/young either a disruption of DNA-binding capacity (p.R38A) [37] adults and a more favorable outcome. This finding is in line or loss of capability to rescue IgM expression—a hallmark with very recent descriptions of PAX5-mutated ALL sub- of PAX5 activation—in PAX5 deficient B cells (p.P80R) groups, which also report a very good [41] or better out- [6, 36, 38]. In PAX5-plus patients, gene set enrichment come for these patients when compared to other molecular analyses confirmed an impairment of PAX5 transcriptional subgroups [36]. PAX5-plus ALL thus contributes to a regulation of B-cell differentiation [33] as well as a dereg- specific profile of outcome-related disease drivers including ulation of PAX5 tumor suppressor functions [9] and its role DUX4-[50] and MEF2D-[51] rearrangements, which share in restricting oncogene-induced energy requirements [7]. a peak incidence in adolescents/young adults, characterizing Nevertheless, PAX5 function was partly retained in this as a potentially distinct clinical entity with specific PAX5-plus ALL patients, as indicated by a CD19-positive therapeutic targets. B-cell immunophenotype, normal distribution of VH gene Driver subgroups in BCP-ALL were so far defined either segment usage during immunoglobulin rearrangement and by gene fusions or aneuploidy patterns. Our data indicate retained expression of a subset of PAX5 target genes. This for the first time that a specific triple-combination of could be explained as target sequence recognition by PAX5 sequence alterations in major hematopoietic regulators depends on asymmetric contributions of the two DNA- defines a molecular subgroup with distinct transcriptional binding subdomains [37, 45]. Sequence mutations in all and epigenetic regulations, corroborating the long-standing PAX5-plus patients affected a linker sequence proximal to assumption [52] that only a limited number of mutations are the N-terminal subdomain or the subdomain itself, which required for malignant transformation. Our findings extend mainly contributes to the specificity of PAX5 DNA binding the well-established function of PAX5 as a haploinsufficient [45]. The less specific C-terminal subdomain remained tumor suppressor in BCP-ALL by a novel role of biallelic unaffected, most likely retaining some less specific PAX5 PAX5 alterations, which cooperate with impaired cell cycle functionality, which would be in line with the general control (CDKN2A/B) and sustained proliferative signaling absence of a complete PAX5-‘null’-phenotype in B-cell (RAS) to drive leukemogenesis. malignancies. Beyond the specific PAX5 sequence mutations, Acknowledgements This study was funded by the German Cancer PAX5-plusALLwasdefined by homozygous deletions Aid (Deutsche Krebshilfe; grant 111533) and by the Deutsche Jose CDKN2A/B Carreras Leukämie Stiftung (grant DJCLS 01 R/2016). Dr. L. Bastian of the locus and RAS-activating alterations. is participant in the BIH CharitéClinician Scientist Program funded by CDKN2A/B loss confers self-renewal capacities and the Charité- Universitatsmedizin̈ Berlin and the Berlin Institute of proliferation of hematopoietic progenitors and BCP-ALL Health. cells [46]. In line with this, we observed an upregulation of hallmark pathways related to DNA replication and cell Compliance with ethical standards cycle transition in PAX5-plus patient samples. A similar fl fl regulation was also observed in a Pax5 deficient Stat5 Con ict of interest The authors declare that they have no con ict of interest. driven murine BCP-ALL model [9], suggesting that CDKN2A/B loss synergizes with PAX5 alterations to Publisher’s note: Springer Nature remains neutral with regard to extend proliferative capacities in PAX5-plus leukemo- jurisdictional claims in published maps and institutional affiliations. genesis. Sustained proliferative signaling through activating RAS mutations is a hallmark of various References malignancies, which we and others have also identified in BCP-ALL[47]. Remarkably, also a cooperating effect 1. Iacobucci I, Mullighan CG. Genetic basis of acute lymphoblastic – of constitutive RAS signaling and CDKN2A/B inacti- leukemia. J Clin Oncol J Am Soc Clin Oncol. 2017;35:975 83. 2. Lilljebjörn H, Henningsson R, Hyrenius-Wittsten A, Olsson L, vation has been described in a murine BCP-ALL Orsmark-Pietras C, von Palffy S, et al. Identification of ETV6- model [48], together indicating that multiple levels RUNX1-like and DUX4-rearranged subtypes in paediatric of proliferative control were disrupted in PAX5-plus B-cell precursor acute lymphoblastic leukaemia. Nat Commun. ALL through converging pathways. This cooperative 2016;7:11790. fi 3. Nordlund J, Bäcklin CL, Zachariadis V, Cavelier L, Dahlberg J, effect might represent a speci c vulnerability in Öfverholm I, et al. DNA methylation-based subtype prediction PAX5-plus ALL, which could be exploited by combined for pediatric acute lymphoblastic leukemia. Clin Epigenetics. treatments with CDK inhibitors to substitute for 2015;7:11. L. Bastian et al.

4. Greaves M. Pre-natal origins of childhood leukemia. Rev Clin 23. Talevich E, Shain AH, Botton T, Bastian BC. CNVkit: genome- Exp Hematol. 2003;7:233–45. wide copy number detection and visualization from targeted DNA 5. Mullighan CG. The genomic landscape of acute lymphoblastic sequencing. PLoS Comput Biol. 2016;12:e1004873. leukemia in children and young adults. Hematol Am Soc Hematol 24. Perez-Llamas C, Lopez-Bigas N. Gitools: analysis and visualisa- Educ Program. 2014;2014:174–80. tion of genomic data using interactive heat-maps. PLoS ONE. 6. Mullighan CG, Goorha S, Radtke I, Miller CB, Coustan-Smith E, 2011;6:e19541. Dalton JD, et al. Genome-wide analysis of genetic alterations in 25. Shochat C, Tal N, Bandapalli OR, Palmi C, Ganmore I, te Kronnie acute lymphoblastic leukaemia. Nature. 2007;446:758–64. G, et al. Gain-of-function mutations in interleukin-7 receptor-α 7. Chan LN, Chen Z, Braas D, Lee J-W, Xiao G, Geng H, et al. (IL7R) in childhood acute lymphoblastic leukemias. J Exp Med. Metabolic gatekeeper function of B-lymphoid transcription fac- 2011;208:901–8. tors. Nature. 2017;542:479–83. 26. Zachariadis V, Gauffin F, Kuchinskaya E, Heyman M, Schou- 8. Heltemes-Harris LM, Willette MJL, Ramsey LB, Qiu YH, Neeley mans J, Blennow E, et al. The frequency and prognostic impact of ES, Zhang N, et al. Ebf1 or Pax5 haploinsufficiency synergizes dic(9;20)(p13.2; q11.2) in childhood B-cell precursor acute lym- with STAT5 activation to initiate acute lymphoblastic leukemia. J phoblastic leukemia: results from the NOPHO ALL-2000 trial. Exp Med. 2011;208:1135–49. Leukemia. 2011;25:622–8. 9. Liu GJ, Cimmino L, Jude JG, Hu Y, Witkowski MT, McKenzie 27. Holmfeldt L, Wei L, Diaz-Flores E, Walsh M, Zhang J, Ding L, MD, et al. Pax5 loss imposes a reversible differentiation block in et al. The genomic landscape of hypodiploid acute lymphoblastic B-progenitor acute lymphoblastic leukemia. Genes Dev. 2014;28: leukemia. Nat Genet. 2013;45:242–52. 1337–50. 28. Brüggemann M, Kotrova M. Minimal residual disease in adult 10. Dang J, Wei L, de Ridder J, Su X, Rust AG, Roberts KG, et al. ALL: technical aspects and implications for correct clinical PAX5 is a tumor suppressor in mouse mutagenesis models of interpretation. Blood Adv. 2017;1:2456–66. acute lymphoblastic leukemia. Blood. 2015;125:3609–17. 29. Gökbuget N, Kneba M, Raff T, Trautmann H, Bartram C-R, 11. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Arnold R, et al. Adult patients with acute lymphoblastic leukemia et al. STAR: ultrafast universal RNA-seq aligner. Bioinforma Oxf and molecular failure display a poor prognosis and are candidates Engl. 2013;29:15–21. for stem cell transplantation and targeted therapies. Blood. 12. Cibulskis K, Lawrence MS, Carter SL, Sivachenko A, Jaffe D, 2012;120:1868–76. Sougnez C, et al. Sensitive detection of somatic point mutations in 30. Hirabayashi S, Ohki K, Nakabayashi K, Ichikawa H, Momozawa Y, impure and heterogeneous cancer samples. Nat Biotechnol. Okamura K, et al. ZNF384-related fusion genes defineasubgroupof 2013;31:213–9. childhood B-cell precursor acute lymphoblastic leukemia with a 13. Garrison E, Marth G. Haplotype-based variant detection from characteristic immunotype. Haematologica. 2017;102:118–29. short-read sequencing. arXiv, 2012. preprint arXiv:1207.3907 31. Herold T, Schneider S, Metzeler KH, Neumann M, Hartmann L, [q-bio.GN]. https://arxiv.org/abs/1207.3907. Roberts KG, et al. Adults with Philadelphia chromosome-like 14. Lai Z, Markovets A, Ahdesmaki M, Chapman B, Hofmann O, acute lymphoblastic leukemia frequently have IGH-CRLF2 and McEwen R, et al. VarDict: a novel and versatile variant caller for JAK2 mutations, persistence of minimal residual disease and poor next-generation sequencing in cancer research. Nucleic Acids Res. prognosis. Haematologica. 2017;102:130–8. 2016;44:e108. 32. Cobaleda C, Schebesta A, Delogu A, Busslinger M. Pax5: the 15. Koboldt DC, Chen K, Wylie T, Larson DE, McLellan MD, guardian of B cell identity and function. Nat Immunol. 2007; Mardis ER, et al. VarScan: variant detection in massively parallel 8:463–70. sequencing of individual and pooled samples. Bioinformatics. 33. Revilla-I-Domingo R, Bilic I, Vilagos B, Tagoh H, Ebert A, Tamir 2009;25:2283–5. IM, et al. The B-cell identity factor Pax5 regulates distinct tran- 16. Rimmer A, Phan H, Mathieson I, Iqbal Z, Twigg SRF, WGS500 scriptional programmes in early and late B lymphopoiesis. EMBO Consortium. et al. Integrating mapping-, assembly- and haplotype- J. 2012;31:3130–46. based approaches for calling variants in clinical sequencing 34. Liberzon A, Birger C, Thorvaldsdóttir H, Ghandi M, Mesirov JP, applications. Nat Genet. 2014;46:912–8. Tamayo P. The Molecular Signatures Database (MSigDB) hall- 17. McLaren W, Gil L, Hunt SE, Riat HS, Ritchie GRS, Thormann A, mark gene set collection. Cell Syst. 2015;1:417–25. et al. The ensembl variant effect predictor. Genome Biol. 2016; 35. Kanehisa M, Furumichi M, Tanabe M, Sato Y, Morishima K. 17:122. KEGG: new perspectives on genomes, pathways, diseases and 18.McPhersonA,HormozdiariF,ZayedA,GiulianyR,HaG, drugs. Nucleic Acids Res. 2017;45:D353–61. Sun MGF, et al. deFuse: an algorithm for gene fusion 36. Gu Z, Churchman ML, Roberts KG, Moore I, Zhou X, Naki- discovery in tumor RNA-Seq data. PLoS Comput Biol. 2011;7: tandwe J et al. PAX5-driven subtypes of B-progenitor acute e1001138. lymphoblastic leukemia. Nat Genet. 2019. https://doi.org/10.1038/ 19. D Nicorici, M Satalan, H Edgren, S Kangaspeska, A Murumagi, O s41588-018-0315-5 Kallioniemi, et al. FusionCatcher—a tool for finding somatic 37. Czerny T, Schaffner G, Busslinger M. DNA sequence recognition fusion genes in paired-end RNA-sequencing data, bioRxiv, 2014, by Pax proteins: bipartite structure of the paired domain and its https://doi.org/10.1101/011650 binding site. Genes Dev. 1993;7:2048–61. 20. Pertea M, Pertea GM, Antonescu CM, Chang T-C, Mendell JT, 38. Shah S, Schrader KA, Waanders E, Timms AE, Vijai J, Miething Salzberg SL. StringTie enables improved reconstruction of a C, et al. A recurrent germline PAX5 mutation confers suscept- transcriptome from RNA-seq reads. Nat Biotechnol. 2015; ibility to pre-B cell acute lymphoblastic leukemia. Nat Genet. 33:290–5. 2013;45:1226–31. 21. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, 39. Auer F, Rüschendorf F, Gombert M, Husemann P, Ginzel S, Izraeli Gillette MA, et al. Gene set enrichment analysis: a knowledge- S, et al. Inherited susceptibility to pre B-ALL caused by germline based approach for interpreting genome-wide expression profiles. transmission of PAX5 c.547G>A. Leukemia. 2014;28:1136–8. Proc Natl Acad Sci USA. 2005;102:15545–50. 40. Martín-Lorenzo A, Hauer J, Vicente-Dueñas C, Auer F, González- 22. Kuilman T, Velds A, Kemper K, Ranzani M, Bombardelli L, Herrero I, García-Ramírez I, et al. Infection exposure is a causal Hoogstraat M, et al. CopywriteR: DNA copy number detection factor in B-cell precursor acute lymphoblastic leukemia as a result from off-target sequence data. Genome Biol. 2015;16:49. of Pax5-inherited susceptibility. Cancer Discov. 2015;5:1328–43. PAX5 biallelic genomic alterations define a novel subgroup of B-cell precursor acute lymphoblastic. . .

41. Passet M, Boissel N, Sigaux F, Saillard C, Bargetzi M, Ba I, et al. 47. Irving J, Matheson E, Minto L, Blair H, Case M, Halsey C, et al. PAX5 P80R mutation identiﬁes a novel subtype of B-cell pre- Ras pathway mutations are prevalent in relapsed childhood acute cursor acute lymphoblastic leukemia with favorable outcome. lymphoblastic leukemia and confer sensitivity to MEK inhibition. Blood. 2019;133:280–4. Blood. 2014;124:3420–30. 42. Li J-F, Dai Y-T, Lilljebjörn H, Shen S-H, Cui B-W, Bai L, et al. 48. Sewastianik T, Jiang M, Sukhdeo K, Patel SS, Roberts K, Kang Transcriptional landscape of B cell precursor acute lymphoblastic Y, et al. Constitutive Ras signaling and Ink4a/Arf inactivation leukemia based on an international study of 1,223 cases. Proc Natl cooperate during the development of B-ALL in mice. Blood Adv. Acad Sci USA. 2018;115:E11711–20. 2017;1:2361–74. 43. Zaliova M, Stuchly J, Winkowska L, Musilova A, Fiser K, Sla- 49. Nemoto A, Saida S, Kato I, Kikuchi J, Furukawa Y, Maeda Y, mova M et al. Genomic landscape of pediatric B-other acute et al. Speciﬁc Antileukemic Activity of PD0332991, a CDK4/6 lymphoblastic leukemia in a consecutive European cohort. Hae- Inhibitor, against Philadelphia Chromosome-Positive Lymphoid matologica. 2019. https://doi.org/10.3324/haematol.2018.204974 Leukemia. Mol Cancer Ther. 2016;15:94–105. 44. Knudson AG. Mutation and cancer: statistical study of retino- 50. Yasuda T, Tsuzuki S, Kawazu M, Hayakawa F, Kojima S, blastoma. Proc Natl Acad Sci USA. 1971;68:820–3 Ueno T, et al. Recurrent DUX4 fusions in B cell acute lym- 45. Perez-Borrajero C, Okon M, McIntosh LP. Structural and phoblastic leukemia of adolescents and young adults. Nat Genet. dynamics studies of Pax5 reveal asymmetry in stability and DNA 2016;48:569–74. binding by the paired domain. J Mol Biol. 2016;428:2372–91. 51. Gu Z, Churchman M, Roberts K, Li Y, Liu Y, Harvey RC, et al. 46. Mullighan CG, Williams RT, Downing JR, Sherr CJ. Failure of Genomic analyses identify recurrent MEF2D fusions in acute CDKN2A/B (INK4A/B-ARF)-mediated tumor suppression and lymphoblastic leukaemia. Nat Commun. 2016;7:13331. resistance to targeted therapy in acute lymphoblastic leukemia 52. Vogelstein B, Kinzler KW. The path to cancer—three strikes and induced by BCR-ABL. Genes Dev. 2008;22:1411–5. you’re out. N Engl J Med. 2015;373:1895–8.

Afﬁliations

1 Department of Hematology/Oncology, Charité - Erlangen-Nürnberg, Erlangen, Germany Universitätsmedizin Berlin, Campus Benjamin Franklin, 10 Berlin, Germany University Hospital Schleswig-Holstein, Campus Kiel, Department of Pediatrics, Kiel, Germany 2 Berlin Institute of Health, Berlin, Germany 11 Research Institute Children’s Cancer Center, Dept. of Pediatric 3 German Cancer Consortium (DKTK), Heidelberg, Germany Hematology and Oncology, University Medical Center Hamburg, Hamburg, Germany 4 German Cancer Research Center (DKFZ), Heidelberg, Germany 12 Department of Medicine III, University Hospital, LMU Munich, 5 Department of Medicine II, Hematology and Oncology, University Munich, Germany Hospital Schleswig-Holstein, Kiel, Germany 13 Laboratory for Functional Genome Analysis, Gene-Center, LMU 6 Department of Pediatric Oncology/Hematology, Charité - Munich, Munich, Germany Universitätsmedizin Berlin, Campus Virchow Klinikum, Berlin, Germany 14 Centre for Genomic Regulation (CRG), The Barcelona Institute of Science and Technology, Barcelona, Spain 7 Institute for Medical Immunology, Charité - Universitätsmedizin Berlin, Berlin, Germany 15 Universitat Pompeu Fabra (UPF), Barcelona, Spain

8 Berlin-Brandenburg Center for Regenerative Therapies (BCRT), 16 Department of Medicine II, Hematology/Oncology, Goethe Charité - Universitätsmedizin Berlin, Berlin, Germany University Hospital, Frankfurt/M, Germany

9 Institute of Human Genetics, Friedrich-Alexander University of