Copy number alterations in B-cell development genes, drug resistance, and clinical outcome in pediatric B-cell precursor acute lymphoblastic leukemia

Pediatric B-cell precursor acute lymphoblastic leukemia (BCP-ALL) is associated with a high frequency of copy number alterations (CNAs) in IKZF1, EBF1, PAX5, CDKN2A/B, RB1, BTG1, ETV6, and/or the PAR1 region (henceforth: B-cell development genes). We aimed to gain insight in the association between CNAs in these genes, clinical outcome parameters, and cellular drug resistance. 71% of newly diagnosed pediatric BCP-ALL cases harbored one or more CNAs in these B-cell development genes. The distribution and clinical relevance of these CNAs was highly subtype-dependent. In the DCOG-ALL10 cohort, only loss of IKZF1 associated as single marker with unfavorable outcome parameters and cellular drug resistance. Prednisolone resistance was observed in IKZF1-deleted primary high hyperdiploid cells (~1500-fold), while thiopurine resistance was detected in IKZF1-deleted primary BCR-ABL1-like and non-BCR-ABL1-like B-other cells (~2.7-fold). The previously described risk stratification classifiers, i.e. IKZF1plus and integrated cytogenetic and CNA classification, both predicted unfavorable outcome in the DCOG-ALL10 cohort, and associated with ex vivo drug cellular resistance to thiopurines, or L-asparaginase and thiopurines, respectively. This resistance could be attributed to overrepresentation of BCR-ABL1-like cases in these risk groups. Taken together, our data indicate that the prognostic value of CNAs in B-cell development genes is linked to subtype-related drug responses.

PAX5. CNAs of the transcription factor PAX5 were observed in 28% of the BCP-ALL cases (Fig. 1C). CNAs were detected throughout all BCP-ALL subtypes, although the frequency was relatively high in BCR-ABL1 (50%) and BCR-ABL1-like (46%) cases (Fig. 1C). The strong co-occurrence of PAX5 and CDKN2A/B CNAs (Fig. 2, OR = 6.36, p < 0.001) is likely caused by the high frequency of chromosome 9p deletions observed in these cases 28 ; the chromosome arm on which PAX5 and CDKN2A/B are located. In correspondence, chromosome 9p deletions were observed in 51.8% (44/85) of these cases. Despite of the strong association between PAX5 CNAs and IKZF1 deletions (Fig. 2, OR = 2.94, p < 0.001), CNAs in PAX5 were not predictive for high MRD levels (Fig. 3C) nor a poor prognosis in pediatric BCP-ALL cases (Fig. 4A). Strikingly, leukemic cells harboring CNAs in PAX5 showed an increased sensitivity (~5.1 fold, p = 0.008) to prednisolone compared to PAX5-wildtype cells (Fig. 5C). This difference in sensitivity was only significant (p = 0.031) in ETV6-RUNX1 cases, but a similar pattern was also observed in the remaining BCP-ALL subtypes (Fig. 5C). Interestingly, this association depended on the IKZF1 status: cells with both an IKZF1 deletion and a CNA in PAX5 were equally sensitive to prednisolone as IKZF1 and PAX5 wildtype cells, whereas cells with only an IKZF1 deletion were more resistant to prednisolone ( Supplementary Fig. 8B). These results suggest that CNAs in PAX5 might compensate for prednisolone resistance induced by loss of IKZF1. ETV6. Deletions of the transcription factor ETV6 were detected in all BCP-ALL subtypes, but were especially enriched in ETV6-RUNX1 cases (71%; Fig. 1D). After induction therapy (TP1), ETV6-deleted cases more often showed low (<10 −4 ) MRD levels compared to ETV6-wildtype cases ( Fig. 3D; OR = 2.6, p = 0.02). However, this association was subtype dependent: in BCR-ABL1-like and B-other cases an adverse association between ETV6 deletions and MRD levels was observed ( Supplementary Fig. 3). Prognosis of cases with loss of ETV6 was not different compared to ETV6-wildtype cases (Fig. 4). ETV6-deleted cells appeared to be more sensitive www.nature.com/scientificreports www.nature.com/scientificreports/ to prednisolone (~3.2 fold, p = 0.046), but more resistant to vincristine (~1.8 fold, p < 0.01) and daunorubicin (~1.9 fold, p = 0.028). Remarkably, loss of the wildtype ETV6 allele in ETV6-RUNX1-positve cells associated with a high sensitivity to vincristine instead of resistance (p = 0.013, Fig. 5D), suggesting that associations of vincristine resistance differ between genetic subtypes of ALL. Moreover, deletion of ETV6 was associated with L-asparaginase resistance in high hyperdiploid cells and high 6-thioguanine sensitivity in ETV6-RUNX1 cells ( Supplementary Fig. 8C).
RB1. The cell cycle regulator RB1 was deleted in a minority (~7%) of the BCP-ALL cases (Fig. 1F) and deletions were detected in all BCP-ALL subtypes. Within the DCOG-ALL10 cohort, RB1-deleted cases showed a trend towards a poor event free survival (5-year EFS: 68.2% versus 86.7%, p = 0.057), which was caused by an unfavorable response in the medium risk (MR) treatment group (5-year EFS: 46.9% versus 88.3%, p = 0.003; Supplementary Fig. 7A). No association with MRD levels or cellular resistance to the tested drug panel was observed (Fig. 5).
BTG1. The anti-proliferative gene BTG1 was deleted in a minority (~8%) of the BCP-ALL cases. No deletions were detected in KMT2A-rearranged or TCF3-PBX1 cases, whereas the highest frequency was observed in BCR-ABL1 (30%) and ETV6-RUNX1 (16%) cases (Fig. 1G). Four out of five BTG1-deleted BCR-ABL1-like and B-other cases also harbored an IKZF1 deletion. These four cases all experienced an event and only the patient with wildtype IKZF1 remained in remission ( Supplementary Fig. 7B). This finding underlines an earlier report, in which a cooperative effect of BTG1 and IKZF1 lesions in leukemogenesis was observed 27 . CNAs in cytokine receptors. PAR1. Deletions in the pseudoautosomal region 1 (PAR1) were the least prevalent (~4%) in this pediatric BCP-ALL cohort. CNAs in this region indicate presence of interstitial deletions or a translocation, which both induce overexpression of CRLF2 29 . Deletions of the PAR1 region were detected in BCR-ABL1-like (11%), B-other (10%), and high hyperdiploid cases (2%), but not in remaining BCP-ALL subtypes (Fig. 1H). Unfortunately, power was lacking to reliable study the association between deletions in the PAR1 region, MRD levels, clinical prognosis, and cellular drug resistance.  www.nature.com/scientificreports www.nature.com/scientificreports/ Taken together, with the exception of loss of the IKZF1 gene, none of the CNAs in the remaining B-cell development genes strongly associates with clinical outcome and cellular drug resistance as single marker. Our results show that the clinical value of CNAs in B-cell development genes is highly context dependent and differs between the diverse oncogenic drivers of pediatric BCP-ALL. www.nature.com/scientificreports www.nature.com/scientificreports/ Risk stratification classifiers. In recent studies, IKZF1 plus and integrated cytogenetic and CNA classification were shown to be prognostic classifiers 18,20 . In the DCOG-ALL10 cohort 12 of the 210 cases were classified as IKZF1 plus20 . The prognosis of IKZF1 plus cases was unfavorable compared to cases with wildtype IKZF1 (Supplementary Fig. 9). Strikingly, no prednisolone resistance was observed in IKZF1 plus cells, which could be explained by underrepresentation of high hyperdiploid cases in this group (n = 1, Fig. 5A). However, IKZF1 plus cases did show ex vivo resistance to 6-thioguanine and 6-mercaptopurine, mainly caused by the high proportion of BCR-ABL1-like and B-other cases in this group.
Integration of cytogenetic and CNA data as reported by Moorman et al. 18 identified cases with a genetic good and poor risk. Cases that were classified as poor risk showed an unfavorable 5-years EFS and CIR compared good risk cases, as shown in Supplementary Fig. 10A. These genetically poor risk cases showed high MRD levels after induction therapy and the first block of consolidation therapy, indicating a poor response to drugs that are used during these treatment phases ( Supplementary Fig. 10B). Indeed, ex vivo cellular drug response data showed resistance of poor risk cells to L-asparaginase, 6-thioguanine, and 6-mercaptopurine (Fig. 5A, Supplementary  Fig. 10C). Enrichment of BCR-ABL1-like cases could attribute to the thiopurine and L-asparaginase resistance observed in the poor risk group 4 .

Discussion
BCP-ALL cases harbor various genetic aberrations in genes involved in lymphoid maturation, cell cycle regulators, tumor suppressors, and tyrosine kinases. We performed an explorative study to gain insight in the association between CNAs in B-cell development genes, MRD levels, long-term prognosis, and cellular drug resistance. Interestingly, the distribution and clinical relevance of these CNAs was subtype-dependent. A high   The association between CNAs in all risk groups and cumulative incidence of relapse (CIR) and event-free-survival (EFS) was examined. BCP-ALL patients (n = 210) were treated according to DCOG-ALL10 protocol. CIR was estimated using a competing risk model. Relapse and non-response were considered as event, and death as competing event. To test equality of the CIRs, the Gray's test was applied. Non-response, relapse, and death were considered as events for EFS. EFS rates were determined using Cox regression, and compared using the Wald test. For reliable test results, groups should contain at least 5 cases. (B) CIR and EFS curves of cases without or with an IKZF1 deletion. Curves contain either all risk groups, or the medium risk arm only.

BCR-ABL1-like/ B-other
High Hyperdiploid ** ** Figure 5. The association between CNAs and the ex vivo cellular drug response. (A) Leukemic cells were incubated for four days with a concentration range of prednisolone (µg/ml), vincristine (µg/ml), L-asparaginase (IU/ml), daunorubicin (µg/ml), 6-mercaptopurine (µg/ml), and 6-thioguanine (µg/ml), after which cell viability was measured using an MTT assay. The Mann-Whitney U test was applied to compare LC50-values. No association is depicted in grey, resistance in blue (p < 0.05, fold induction (FI) > 1), sensitive in green (p < 0.05, FI < 1), and not determined in white. The number of cases that were tested for prednisolone is depicted, and represent the maximum number of cases. For reliable test results, groups should contain at least 5 cases (groups ≤ 5 are depiced as ND). Results of single CNAs are depicted for all risk groups and for BCR-ABL1like/B-other cells, high hyperdiploid cells, and ETV6-RUNX1 cells. In addition, associations between the risk classifiers IKZF1 plus and integrated cytogenetic and CNA classification (poor risk) and cellular drug resistance are shown 18,20  www.nature.com/scientificreports www.nature.com/scientificreports/ frequency of CNAs in these B-cell development genes was found in the poor prognostic subtypes BCR-ABL1, BCR-ABL1-like, and B-other. Cooperative lesions may favor the aggressive phenotype of a leukemia, such as exemplified by the synergistic effect between loss of IKZF1 and the BCR-ABL1 fusion gene in leukemogenesis 30 , and the antagonizing effect of IKZF1 deletions in the response to imatinib 31 . In contrast, the prognosis of ETV6-RUNX1, DUX4-rearranged, and ERG-deleted BCP-ALL is probably not affected by IKZF1 deletions, but numbers with IKZF1 deletions in these subtypes are low 11,12,15,32,33 . These observations indicate that the genetic context influences the functional effect of CNAs in B-cell development genes. The importance of the genetic context is exemplified by the fact that isolated deletions of BTG1 do not affect cellular drug resistance or the prognosis of BCP-ALL cases, whereas all four patients with concomitant loss of BTG1 and IKZF1 experienced an event. Moreover, combined loss BTG1 and IKZF1 was shown to enhance glucocorticoid resistance 27 . In contrast to BTG1-IKZF1 synergy, we observed that CNAs in PAX5 may counteract the effect of an IKZF1 deletion on prednisolone resistance. Various combinations of cooperative lesions may therefore have different effects on the pathobiology of B-cell precursor ALL cells.
In the present study we observed an association between deletion of IKZF1 and prednisolone resistance, especially in high hyperdiploid cells. In correspondence, a direct association has been demonstrated between IKZF1 deletion and glucocorticoid-induced cell death 25,34 . IKZF1 functions as a metabolic gatekeeper and consequently loss of IKZF1 results in increased intracellular ATP and glucose levels 34 . Interestingly, we previously observed a direct relation between an increased glycolytic rate and prednisolone resistance in primary BCP-ALL cells 35,36 . In these leukemic cells, inhibition of glycolysis restored the efficacy of prednisolone 36 . Hence, inhibition of glycolysis might also be a potential treatment strategy to re-sensitize IKZF1-deleted cells to prednisolone and should be explored in more detail in future studies, also in the context of BTG1 and PAX5.
In contrast to high hyperdiploid cells, deletion of IKZF1 was not linked to prednisolone resistance in primary BCR-ABL1-like and B-other ALL cells, suggesting that additional factors (e.g. differentiation stage, other oncogenic drivers) are important for the functional effect of a deletion of the IKZF1 gene in these type of cells. Instead of prednisolone resistance, we observed thiopurines resistance in these BCR-ABL1-like and B-other ALL cells. Thiopurine resistance might be caused by deficiencies in the DNA mismatch repair system and indeed DNA repair genes were reported to be downregulated in IKZF1-deleted cells 37,38 . Interestingly, this characteristic might offer opportunities to target these leukemic cells via the DNA mismatch repair apparatus, e.g. by PARP inhibitors, and warrants further studies.
Besides IKZF1, deletion of RB1 was predictive for a poor outcome in the MR-risk group of the DCOG-ALL10 cohort. RB1 deletions are known to be enriched in poor prognostic iAMP21 and hypodiploid cases, which might explain the unfavorable outcome of RB1-deleted cases 13,39,40 . However, the unfavorable outcome could not be explained by or cellular resistance against induction therapy drugs.
Recently, two independent studies showed that integration of genetic aberrations improved the risk stratification of BCP-ALL in children 18,20 . Both IKZF1 plus and integrated cytogenetic and CNA classification predicted poor outcome in the DCOG-ALL10 cohort, and associated with drug resistance to thiopurines, or L-asparaginase and thiopurines, respectively. The cellular drug resistance could be attributed to overrepresentation of BCR-ABL1-like cases in these risk groups 4 . Taken together, our results suggest that the prognostic value of CNAs in B-cell development genes is linked to subtype-related drug resistance.
In the current study, we restricted our analyses to CNAs in eight genes that are recurrently deleted in pediatric BCP-ALL. However, additional genetic aberrations may be of importance for prognosis and cellular drug resistance and should be explored in future research. Moreover, as we performed an explorative study, it is of importance to confirm the associations that are proposed in the present paper in independent studies.
In conclusion, results obtained in the present study revealed that, with the exception of an IKZF1 deletion, none of the remaining CNAs as single marker associated both with an unfavorable clinical prognosis and cellular drug resistance. Our results indicate that the biological and clinical importance of CNAs in B-cell development genes (and presumably also other genetic aberrations) is highly context dependent and differs between the diverse oncogenic drivers of pediatric BCP-ALL. Functional studies that focus on potential causes of cellular drug resistance should therefore take the oncogenic driver and additional genetic aberrations into account.

Methods
processing of primary patient material. Bone marrow and/or peripheral blood samples were obtained from children (1-18 years) with newly diagnosed ALL. Written informed consent was obtained from parents or guardians to use excess of diagnostic material for research purposes, as approved by the Medical Ethics Committee of the Erasmus Medical Center, The Netherlands. These studies were conducted in accordance with the Declaration of Helsinki. Mononuclear cells were isolated using Lymphoprep gradient separation and the leukemic blast percentage was determined microscopically by May-Grünwald Giemsa stained cytospin preparations, as described previously 21 . Samples were enriched to over 90% purity of leukemic cells by depletion of non-leukemic cells using immunomagnetic beads. Primary leukemic cells were maintained in RPMI-1640 Dutch modification supplemented with 20% fetal calf serum (Integro), with 0.1% insulin-transferrin-sodium selenite (Sigma), 0.4 mM glutamine (Invitrogen), 0.25 μg/ml gentamycine (Gibco), 100 IU/ml penicillin (Gibco), 100 μg/ ml streptomycin (Gibco), 0.125 μg/ml fungizone (Gibco).
Clinical characteristics and statistics. To identify whether CNAs were underrepresented or enriched in a subtype, the Fisher's exact test was applied using R software (version 3.2.1). Obtained odds ratios (ORs), 95% confidence interval, and p-values are reported. The Fisher's exact test was also applied to compare minimal residual disease (MRD) levels after induction and first consolidation therapy between patients groups with CNAs and wildtype patients. Cumulative incidence of relapse (CIR) was estimated using a competing risk model and significance was determined using the Gray's test. Relapse and non-response (counted as event at day 79) were considered as event, with death as competing event. Event-free survival (EFS) probabilities were estimated using cox regression and compared using the Wald test. Relapse, non-response, secondary malignancies and death were counted as events. Outcome analyses were performed in R (version 3.2.1), using the packages cmprsk version 2.2-7 46 , mstate version 0.2.7 47 and survival version 2.38-4 48 . Five-year EFS and CIR are reported. The DCOG-ALL10 trial is the most recently completed nationwide trial in which patients were risk-stratified by minimal residual disease (MRD) levels and for whom sufficient long-term follow-up data were available. Therefore, we restricted the analysis of associations between CNAs and clinical response parameters (MRD, clinical outcome) to this cohort. In addition, the genetic subtypes are represented with a distribution that is comparable to the general pediatric BCP-ALL population (excluding BCR-ABL1-positive cases since these patients are eligible for the EsPhALL protocol), i.e.12.2% BCR-ABL1-like, 13.9% non-BCR-ABL1-like B-other, 33.5% ETV6-RUNX1, 32.7% high hyperdiploid, 2.0% KMT2A-rearranged, and 5.7% TCF3-PBX1 positive cases. The clinical characteristics of this cohort are displayed in Supplementary Table 2.

Data Availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.