Introduction

ETV6::RUNX1, arising from the chromosomal translocation t(12;21)(p13;q22), is the most common fusion gene in childhood acute lymphoblastic leukemia (ALL)1. Around 20%–30% of pediatric ALL harbor this rearrangement, making it one of the largest subtypes within the spectrum of this cancer2, with some variation by population ancestry3. In most frontline trials, ETV6::RUNX1-rearranged ALL has excellent outcomes with overall survival exceeding 90%4,5. However, due to the large number of patients in this subtype, it still accounts for a substantial proportion of the relapses of ALL6,7. Interestingly, there is growing evidence that a certain subset of ETV6::RUNX1 ALL can be cured by only one year of chemotherapy, instead of the usual 2-year regimens8, suggesting over-treatment of some patients with current risk stratification strategies9. Thus, further understanding of the heterogeneity within ETV6::RUNX1 ALL is necessary to accurately gauge treatment responses.

The leukemogenesis of ETV6::RUNX1 ALL exemplifies the classical two-hit paradigm of cancer pathogenesis. The fusion gene itself, which has a prenatal origin, promotes pre-cancerous proliferation10, but overt leukemic transformation requires the acquisition of secondary genomic lesions during early childhood11. Somatic deletions targeting genes related to hematopoietic differentiation such as PAX5, BTG1, or ETV6 are frequently observed in this ALL subtype2. It is plausible that the type of secondary alteration could influence divergent molecular processes of leukemogenesis and thereby affect leukemia biology. However, the link between the secondary alterations and the heterogeneity of ETV6::RUNX1 ALL as well as their therapeutic implications, remains incompletely understood.

In this study, we comprehensively characterize ETV6::RUNX1 ALL using whole transcriptome and genome sequencing, coupled with leukemia ex vivo sensitivity profiling for 20 targeted or cytotoxic chemotherapeutic agents. Our analysis reveals two distinct subtypes within ETV6::RUNX1 ALL marked by differential gene expression, genomic alterations, and drug sensitivity.

Results

Transcriptomic subtypes of ETV6::RUNX1 ALL

We first obtained RNA-Seq data of 194 ETV6::RUNX1 ALL patients (age 1–16 years) enrolled on the consecutive frontline ALL trials at St. Jude Children’s Research Hospital (SJCRH)12,13. Non-negative matrix factorization of the gene expression profiles revealed two components of molecular heterogeneity, after which a total of 252 genes that contributed most to these components were selected (see “Methods” section for details)14. Based on these genes, unsupervised clustering revealed that there are two sub-clusters with distinct gene expression profiles, as shown in the Uniform Manifold Approximation and Projection (UMAP) in Fig. 1a. The first sub-cluster (C1) comprised 61% (n = 118) of the patients while the second sub-cluster (C2) accounts for 39% (n = 76; Fig. 1a). The two sub-clusters displayed distinct clinical characteristics: C1 has a higher proportion of cases diagnosed at age 5 or older (36% vs 21% in C1 vs C2; P = 0.023; Fig. 1b), whereas C2 patients were more likely to present with high white blood cell count (WBC) with 27% of patients above 50,000/μL, compared to only 5% in C1 (P =  2.6× 10−4; Fig. 1c).

Fig. 1: Distinct subtypes within ETV6::RUNX1 ALL.
figure 1

a UMAP dimension reduction of the gene expression signature of the primary discovery cohort based on 252 genes identified using non-negative matrix factorization. b, c Comparison of b the age group and c the presenting white blood cell (WBC) count between the two subtypes in the primary discovery cohort. d UMAP of the validation cohort based on 122 genes identified independently on the validation cohort using non-negative matrix factorization. e, f Comparison of e the age group and f the presenting WBC between the two subtypes in the validation cohort. In b, c, e, f association between categorical variables are tested using Chi-squared test. Source data are provided as a Source Data file.

Full size image

For validation, we assembled an independent cohort of 95 patients of non-European descent: 74 Malaysian and Singaporean patients from the MaSpore ALL 2003/2010 studies (age 1–14 years)15,16 and 21 children from the Guatemala LLAG-0707 trial (age 2–13 years)17. Using prediction analysis of microarrays (PAM)18, we trained a model on the discovery cohort and predicted the sub-cluster identity of each patient in the validation cohort, which assigned 64% (n = 61) and 36% (n = 34) of patients to C1 and C2, respectively. In addition, we performed gene selection and UMAP analysis independently on the validation cohort, which also revealed two sub-clusters that are highly consistent with the PAM-predicted classes with 92% (n = 87 of 95; Fig. 1d) assigned to the same sub-clusters. Similar to the discovery cohort, compared to C2, C1 cases in the validation dataset had higher proportion of patients older than 5 at diagnosis (54% vs 29%; P = 0.020; Fig. 1e) and less likely to present with high WBC above 50,000/μL (12% vs 32%; P = 0.014; Fig. 1f). Taken together, these results point to the existence of two separate molecular subtypes within ETV6::RUNX1-rearranged ALL which are associated with presenting features regardless of ancestral background.

Differential maturation state of ETV6::RUNX1 subtypes

To further explore the biological features of each subtype within ETV6::RUNX1 ALL, we performed differential gene expression analysis comparing the two subtypes (Supplementary Fig. 1), followed by gene set enrichment analysis (Fig. 2a and Supplementary Table 1). Several pathways, such as cytokine-cytokine receptor interaction, graft versus host disease, and hematopoietic stem cell differentiation were found to be upregulated in C1 (Fig. 2a). Therefore, we hypothesized that the two subtypes were characterized by distinct B cell maturation arrest during leukemogenesis. To this end, we performed digital cytometry19 based on RNA-Seq gene expression and a B cell development signature gene matrix20, to reconstruct the maturation arrest of each ETV6::RUNX1 case in this cohort. This analysis revealed that leukemia blasts in this cohort mostly comprised populations resembling pro-B cells, followed by pre-pro-B-like cells, with a small but appreciable level of hematopoietic stem-like cells (HSCs) (Fig. 2b). Between the two subtypes, C1 demonstrated a higher proportion of HSC-like cells (P = 1.2 × 10−5; Fig. 2c), e.g., those with upregulated expression of GATA2 and CD34 (Supplementary Fig. 1). In C2, a higher proportion of pre-pro-B-like (P = 1.9 × 10−5; Fig. 2c) and pro-B-like (P = 1.9 × 10−5; Fig. 2c) cells were seen, which was in line with the CD38 upregulation (Supplementary Fig. 1). These observations suggest plausible differences in B cell developmental state at which leukemic transformation occurred.

Fig. 2: Signature of B cell differentiation of ETV6::RUNX1 ALL subtypes.
figure 2

a Gene set enrichment analysis comparing the C2 versus C1 using the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway and WikiPathways (WP) databases. Pathways with adjusted P < 0.001 are shown. The significance of a pathway is tested by permutation and corrected for multiple testing using the Benjamini–Hochberg procedure. b Percentages of cells in different B cell differentiation stages using gene expression-based deconvolution analysis. Only stages along the B cell differentiation are included. c Comparison of the deconvoluted percentages of HSC-like, pre-pro-B-like, and pro-B-like cells between the two ETV6::RUNX1 subtypes. Difference between the two subtypes is tested using the two-sided Mann–Whitney U test. Center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range; points, outliers. HSC-like, hematopoietic stem cell-like; CLP-like, common lymphoid progenitor-like. Source data are provided as a Source Data file.

Full size image

Treatment responses and drug sensitivity of ETV6::RUNX1 subtypes

We obtained early treatment response data, i.e. minimal residual response (MRD), of 133 patients in SJCRH TOTAL XV and XVI studies. Despite lower presenting WBC, patients in C1 are characterized with poorer early treatment response, in terms of MRD measured at both day 15 (P = 0.021; Fig. 3a) and end of induction (day 42; P = 0.0030; Fig. 3b). To understand the pharmacological basis of the differences in MRD, we tested ex vivo drug sensitivity of ETV6::RUNX1 ALL to 20 commonly used anti-leukemia agents, with a total of 1546 LC50 measurements21 (Fig. 3c). In general, samples in C1 were more drug-resistant, especially to mercaptopurine (P = 1.3 × 10−7; Fig. 3d), thioguanine (P = 4.8 × 10−4; Fig. 3d), and prednisolone (P = 0.0022; Fig. 3d). The statistical significance of the difference in mercaptopurine seems to be greater than thioguanine, likely because the former was tested in more patients. Several other drugs such as dexamethasone, cytarabine, venetoclax, etc., also demonstrated a trend toward resistance in C1 (Fig. 3c), although statistical significance was not reached due to smaller sample sizes.

Fig. 3: Treatment responses of ETV6::RUNX1 subtypes.
figure 3

a, b Comparison of the a day 15 and b end-of-induction (EOI) minimal residual disease (MRD) between the C1 and C2 cases. Horizontal dashed lines separate cases with positive and negative MRD. A small random value is added to negative MRD (<0.01%) values for visualization. c Comparison of the ex vivo drug response to 20 therapeutic agents. Red dashed line indicate Bonferroni corrected P = 0.05. d Mercaptopurine, thioguanine and prednisolone were more resistant in C1 compared to C2. Red horizontal bars indicate median values. In all panels, difference between the two subtypes is compared using two-sided Mann–Whitney U test. Source data are provided as a Source Data file.

Full size image

Focal deletion of PAX5 is associated with ETV6::RUNX1 molecular subtypes

We next assessed the somatic genomic alterations including single-nucleotide variants (SNVs), short indels and copy number alterations (CNAs), using whole genome sequencing, or whole exome sequencing in combination with single-nucleotide polymorphism (SNP) arrays for a total of 177 patients22,23. Overall, frequent deletions were observed at the VPREB1 (55%), ETV6 (44%), PAX5 (29%), CDKN2A/B (19%), CDKN1B (15%), BTG1 (15%) and 6p22.2 histone cluster (14%) loci, along with recurrent mutations in KRAS (10%) and NRAS (6%) genes (Fig. 4a), with prevalence consistent with previously reported data24. Interestingly, most of the PAX5 deletions were found in C2, accounting for 64% (43 of 67) of patients in this subtype, 8 times more frequent than in C1 (7%; 8 of 110; P = 5.1 × 10−16; Fig. 4a). All PAX5 deletions in this cohort were heterozygous, and most affected exons 2–5 (Fig. 4b), which spans the DNA binding domain, suggesting that these deletions are likely loss-of-function. In addition, C2 also demonstrated a higher frequency of RAS (KRAS or NRAS) mutations (30% vs 6% of C1; P = 2.8 × 10−5; Fig. 4a). Of note, we also observed heterozygous IKZF1 deletions in 3% of ETV6::RUNX1 ALL, but with no significant difference in frequency between the two subtypes (Supplementary Fig. 2).

Fig. 4: Genomic alterations of the two ETV6::RUNX1 ALL subtypes.
figure 4

a Somatic single-nucleotide variant (SNV), short indels, and focal copy number alterations. Only genes altered in at least 10 cases are shown. b Positions of the PAX5 deletions. c Recurrent aneuploidies identified by RNA-Seq, and compared between the two subtypes using Chi-squared test. d, e Distribution of the burden of SNVs separated by mutation signatures of d C1 and e C2 cases. f Violin plots comparing the burden of SBS2 and SBS13 between C1 and C2 subtypes. Red horizontal bars indicate median values. Difference between the two subtypes is tested using the two-sided Mann–Whitney U test. Source data are provided as a Source Data file.

Full size image

It has been previously reported that ETV6::RUNX1 ALL is associated with frequent RAG-mediated structural variations (SVs)25. We examined RAG-mediated SVs based on the presence of recombination signal sequence (RSS) motifs around the junctions of SVs, but did not observe a significant difference between the two subtypes (Supplementary Fig. 3). Using RNA-Seq data, we identified recurrent chromosomal copy number changes: gain of chromosome 10 (6%), 16 (5%), and 21 (15%), as well as loss of chromosome 15 (1%; Fig. 4c). Notably cases in C1 had more frequent trisomy 16 (8% vs 1%; P = 0.049) but less frequent trisomy or tetrasomy 21 compared to C2 (9% vs 24%; P = 0.0072; Fig. 4c). In 89 cases with WGS data, we examined the Catalogue of Somatic Mutations in Cancer (COSMIC) Single Base Substitution (SBS) mutation signatures. Consistent with previous findings2, the clock-like signatures SBS1 and SBS5 were universally present. The AID/APOBEC-related mutation signatures SBS2 and SBS13, which were previously shown to be enriched in ETV6::RUNX1 ALL2,26,27, were present in 53% of patients (Fig. 4d, e). Notably, SBS2 and SBS13 are more prevalent in C2, accounting for 60% of the SNVs in this subtype, compared to 26% in C1 (P = 0.0060 and 0.0098, respectively; Fig. 4f).

To validate these findings, we examined RNA-Seq and whole genome seq data of a large independent cohort of ETV6::RUNX1 ALL (n = 552) from the Children’s Oncology Group28. Using PAM, we identified 341 (62%) and 211 (38%) cases as C1 and C2 subtypes, consistent with the frequencies seen in the discovery cohort. Also relative to C1 subtype, C2 cases were more likely to have PAX5 deletions (63% vs 7%; P = 3.6 × 10−43), RAS mutations (37% vs 15%; P = 5.8 × 10−9; Supplementary Fig. 4), and SBS2 (P = 5.8 × 10−9) and SBS13 (P = 1.9 × 10−12) mutational signatures, again replicating the findings in the discovery cohort. Taken together, these data point to the distinct etiology and divergent leukemogenic processes of the two ETV6::RUNX1 ALL subtypes.

PAX5 expression conferred resistance to mercaptopurine in ETV6::RUNX1 C2 cells

Because PAX5 is the gene that demonstrated the most significant difference between the two subtypes, we elected to functionally explore the role of this gene in the context of ETV6::RUNX1 ALL. In particular, we are interested in the mechanism by which PAX5 regulates mercaptopurine sensitivity. First, we identified the REH cell line as ETV6::RUNX1 C2 subtype based on the global gene expression pattern (Fig. 5a) and the presence of heterozygous PAX5 frameshift mutation (p.A322Rfs*19)29. As shown in Fig. 5b and Supplementary Fig. 5, re-introduction of PAX5 by ectopic expression conferred a transcriptomic switch of the subtype identity from C2 to C1 in REH cells (Fig. 5a). Similar to patients of C1 subtype, REH-PAX5 cells were significantly more resistant to mercaptopurine compared to the control cells transduced with empty vector (REH-EV; Fig. 5c). Mercaptopurine, as a pro-drug, is extensively metabolized and its incorporation into double-stranded DNA during S phase of the cell cycle is required to induce apoptosis. We directly measured DNA-incorporated thioguanine (DNA-TG), the active drug metabolite required for mercaptopurine cytotoxicity, by tandem mass spectrometry, and observed a significantly lower DNA-TG level in REH-PAX5 cells (t = 4.881, degrees of freedom[df]=4, P = 0.0082; Fig. 5d), consistent with the drug resistance.

Fig. 5: Ectopic expression of PAX5 in the REH cell line of the C2 subtype conferred resistance to mercaptopurine.
figure 5

a PAM score to predict the C1/C2 subtype identity using the gene expression profiles showed the switch from C2 to C1 after overexpression of PAX5 in REH cells. b PAX5 protein is overexpressed in REH-PAX5 cells. Results are representative of three independent experiments. c REH-PAX5 cells demonstrated resistance to mercaptopurine when compared to REH-EV cells. Data are presented as mean ± SD for three biological replicates and results are representative of three independent experiments. d DNA-TG incorporation was significantly reduced in REH-PAX5 cells. Data are presented as mean ± SD for three biological replicates. Comparisons are performed using a two-sided t-test. e Venn diagram showing the overlap among the top 1000 genes that demonstrated resistance to mercaptopurine upon knockout and the top 1000 genes that are upregulated in C2 (compared to C1) and REH-PAX5 (compared to REH-EV). f CDK6 protein and pRB are reduced in REH-PAX5 cells. Results are representative of two independent experiments. g REH-PAX5 cells demonstrated a significantly slower growth rate compared to REH-EV. Data are presented as mean ± SD for three biological replicates and results are representative of three independent experiments. P-value is by linear regression adjusted for days of culturing. h Percentage of non-proliferative cells by CellTrace assay is significantly higher in REH-PAX5 cells. Data are presented as mean ± SD for three biological replicates and comparisons are performed using a two-sided t-test. i Percentage of BrdU+ cells is significantly lower in REH-PAX5 cells. Data are presented as mean ± SD for three biological replicates and comparisons are performed using a two-sided t-test. j REH cells pre-treated with palbociclib demonstrated mercaptopurine resistance. Data are presented as mean ± SD for three biological replicates. Error bars indicate standard deviations. EV, empty vector. Source data are provided as a Source Data file.

Full size image

PAX5 partially blocks the G1 to S transition by downregulating CDK6

To identify the functional targets downstream of PAX5, we compared the following three gene lists: (1) top 1000 mercaptopurine-resistance genes previously identified in a genome-wide CRISPR screen in the REH cell line30, (2) top 1000 upregulated genes in C2 patients compared to C1, and (3) top 1000 upregulated genes in REH-EV compared to REH-PAX5 cells. Eight genes overlapped across these three gene lists (Fig. 5e): CDK6, PXK, STK39, CDS2, NAE1, CELF2, SUPT4H1, and CHAF1B. Of them, CDK6 was known to be essential for G1 to S phase transition in the cell cycle, a process that may affect DNA-TG incorporation, providing a plausible explanation of the impact on mercaptopurine sensitivity. In fact, CDK6 was highly expressed in all C2 cases, as well as C1 cases with PAX5 deletions; Finally, CDK6 upregulation was also directly associated with mercaptopurine sensitivity within ETV6::RUNX1 ALL (P = 3.4 × 10−6), even after adjusting for C1/C2 status (P = 0.01; Supplementary Fig. 6).

As shown in Fig. 5f, REH cells completely lost CDK6 expression upon PAX5 re-introduction, which also led to a decreased level of phosphorylated RB (pRB), the substrate of CDK6 required for G1 to S transition (Fig. 5f). Meanwhile, we observed a significant decrease in the growth rate (P = 2.2 × 10−4; Fig. 5g) and a lower percentage of non-proliferative cells of REH-PAX5 cells, relative to REH-EV cells (5.89% vs 41.07%, t = 32.22, df=4, P = 5.5 × 10−6; Fig. 5h and Supplementary Fig. 7). Using BrdU incorporation assay, we found that the percentage of S Phase cells was significantly lower in REH-PAX5 cells (37.73% vs 18.50%, t = 132.4, df = 4, P = 2.0 × 10−8; Fig. 5i and Supplementary Fig. 7). These observations are consistent with the downregulation of DNA replication pathway in C1 subtype (Supplementary Table 1). Similar to REH-PAX5 cells, parental REH cells pre-treated with palbociclib, a CDK4/6 inhibitor, also exhibited marked resistance to mercaptopurine (Fig. 5j), decreased proliferation rate (P = 1.8 × 10−4; Supplementary Fig. 7), and reduced percentage of cells in the S phase (t = 80.61, df = 4, P = 1.4 × 10−7; Supplementary Fig. 7), further confirming the role of CDK6-mediated G1 to S phase transition in mercaptopurine sensitivity in ETV6::RUNX1 ALL. Therefore, these data clearly suggest that genomic loss of PAX5 enables CDK6 transcription, cell cycling, and leukemia cell sensitivity to mercaptopurine.

Discussion

PAX5 deletion is one of the most frequently observed secondary alterations in ETV6::RUNX1-rearranged ALL. Our analyses revealed that PAX5 deletion is related to a distinct subtype (C2) of disease within ETV6::RUNX1 ALL, indicating that the type of second hit determines the tumor cell identity in ETV6::RUNX1 ALL. In addition, the C2 subtype is also associated with increased frequencies of APOBEC/AID-related mutations, further confirming the distinct etiology of the two subtypes.

ETV6::RUNX1 ALL is characterized by a transcriptomic profile highly resembling pro-B-like cells31, consistent with the notion that this subtype has a more differentiated developmental state. A previous single-cell study of six cases showed that ETV6::RUNX1 ALL cells uniformly express DNTT and mostly resemble the pro-B state, despite the variance in cell surface marker expression32. Reanalyzing these single-cell sequencing data, we confirmed four and two cases of the C1 and C2 subtypes, respectively. Of interest, C1 cases in this series showed higher CD34 surface expression by flow cytometry, compared to two C2 cases (Supplementary Fig. 8), pointing to HSC-like features in C1. Therefore, we reason that the HSC-like gene expression profile in C1 arises from the retention of the HSC transcriptomic program across all cells in C1 blasts, rather than an increase of HSC-like cells in heterogeneous ALL blasts. However, larger-scale single-cell profiling studies are required to confirm these hypotheses.

PAX5 is an important transcription factor for B-lineage commitment and identity maintenance. The exact role of PAX5 deletion in B-ALL leukemogenesis is yet to be fully elucidated, but it seems to induce leukemic arrest at more downstream stages along B cell differentiation than those without this deletion, in the common ETV6::RUNX1 subtype as we described in this study and in the rare BCR::ABL1 subtype as reported previously33. The functional consequence of PAX5 deletions is likely to be subtype-dependent, due to its broad ability of transcriptional regulation. In C1 ETV6::RUNX1 B-ALL, PAX5 downregulates DNA replication by blocking the transition from G1 to the S phase, conferring resistance to mercaptopurine. This highlights the importance of the cell cycle in mercaptopurine sensitivity, especially the S phase when thioguanine is incorporated in the genome. While drug combination with CDK4/6 inhibitors is increasingly considered in ALL treatment34, our data suggest that simultaneous use with mercaptopurine should be avoided.

Though PAX5 deletion is over-represented in the C2 subtype, it was not obligatory, with one-third of C2 cases showing wildtype PAX5, raising the possibility that factors other than PAX5 also contribute to molecular and pharmacological phenotypes of C2. In fact, cases with PAX5 deletion (in both C1 and C2) almost always exhibited high CDK6 expression and drug sensitivity, although a fraction of PAX5 wildtype samples also show these phenotypes (plausibly driven by genomic abnormalities independent of PAX5, Supplementary Fig. 6). For example, NRAS or KRAS mutations were also more frequently seen in C2, which may influence cell cycle35 and drug response. In addition, C1 cases displayed a trend of having more IKZF1 deletions (Supplementary Fig. 2), which has already been associated with resistance to cytotoxic chemotherapeutics36,37. Taken together, we reason that C1/C2 identity (reflecting overall ALL transcriptomic state) is a stronger predictor of drug response than single genomic lesions such as PAX5 (Supplementary Fig. 9). Future studies are certainly required to elucidate the role of individual genomic factors on molecular and pharmacological heterogeneity of ETV6::RUNX1 ALL.

It is interesting that cases in the drug-sensitive subtype C2 are characterized by higher presenting WBC in peripheral blood, which often means that these patients receive more intensive chemotherapy (e.g., per the “NCI risk” stratification). Therefore, the clinical value of our finding may lie in the possibility of treatment de-intensification for ETV6::RUNX1 patients with the C2 subtype, given their sensitivity to thiopurines and superior early treatment response (low MRD at the end of induction). This is in line with the previous observation that a certain subset of ETV6::RUNX1 ALL can be cured with significantly shortened maintenance therapy8, although whether these patients were indeed of the C2 subtype remains unknown.

Clinical diagnosis of C1/C2 subtype within ETV6::RUNX1 ALL requires gene expression analysis, which can be done using RNA-Seq. By comparison, real-time quantitative polymerase chain reaction (qPCR) or low-density array based on selected genes can be cost-effective alternatives. To explore the feasibility, we selected different numbers of the top differentially expressed genes between C1 and C2 in the discovery cohort and developed a subtype score (see “Methods” section for details) to classify cases in the validation set. While there was a slight improvement when more genes were used (i.e., 10, 20, 30, or 50 genes), the performance was consistently high across gene panels (AUC ≥ 0.95; Supplementary Fig. 10).

In summary, the presence of two distinct subtypes within ETV6::RUNX1 ALL implies that there are divergent pathways involved in the leukemia development that lead to variations in clinical characteristics and treatment responses. Our data provided valuable insights into the pathogenesis of ETV6::RUNX1 ALL and could potentially guide more tailored therapeutic approaches.

Methods

Patients and samples

This study complies with all relevant ethical regulations and was approved by the St. Jude Children’s Research Hospital Institutional Review Board. A total of 194 patients with ETV6::RUNX1-rearranged ALL enrolled in the SJCRH Total XV13 (ClinicalTrials.gov NCT00137111), XVI12 (NCT00549848), and XVII (NCT03117751) studies were included in the primary analysis. MRD data were obtained from SJCRH Total Therapy XV and XVI clinical trials12,13 and no outcome-related data from SJCRH Total XVII study was included in this report. Genomic data availability of SJCRH patients is included in Supplementary Data 1. Another 74 patients treated in Malaysia and Singapore (MaSpore) ALL 2003 (NCT02894645) and 2010 studies15,16, and 21 patients treated in Guatemala LLAG-0707 trial17 were included as a validation set. Samples were collected with informed consent obtained from patients, parents, and/or guardians, in written and signed form. Sex or gender was not considered in the study design. However, the self-reported biological sex of the patients is analyzed in this study (Supplementary Fig. 1), because it has been recognized that the biological sex of patients affects the biology and outcome of acute lymphoblastic leukemia. Source data on sex is provided in the source data file.

RNA-Seq and analysis

For SJCRH and Guatemala samples, RNA-Seq libraries were generated using TruSeq stranded total RNA library prep kit (Illumina) on diagnostic samples and sequenced using Illumina HiSeq 2000/2500 or NovaSeq platform with 2 × 101 bp read length. RNA-Seq data of MaSpore patients were obtained from previous publication38. The presence of ETV6::RUNX1 rearrangement was confirmed using CICERO39 (v0.3.0) and/or fusioncatcher40 (v0.99.7d). Reads were mapped to hg19 genome reference by STAR41 (v2.3.0e_r291). Reads mapped to each gene were counted using featureCounts42 (v1.5.1) and expression levels of genes were obtained using variant stabling transformation43. Differential gene expression analyses were performed using DESeq243 (v1.40.2). Chromosomal gains were called using RNASeqCNV44 (v.1.2.1), a method that combines variant allele coverage of SNPs and gene expression of each chromosome for aneuploidy identification.

Unsupervised clustering was performed using non-negative matrix factorization45 followed by UMAP. Non-negative matrix factorization (NMF) was performed using the top 2000 (or top 1000 for the validation set due to the smaller number of samples) most variable genes with the highest median absolute deviations. NMF represents a high-dimensional gene expression matrix using a smaller number of metagenes. We varied the rank (number of metagenes) used in NMF between 2 and 8, and a rank of 2 was selected based on the cophenetic coefficient. The genes that contribute most to the metagenes were extracted from the NMF analysis based on gene score as previously described14. In brief, a higher gene score value in NMF indicates the gene expression is more specific to a metagene, and genes with a score higher than three times the median absolute deviation above the median are selected. Using the selected genes, the gene expression was reduced to two dimensions using UMAP, followed by clustering using DBSCAN. This process was performed independently on both the primary and validation data.

To investigate if the C1/C2 subtype within ETV6::RUNX1 ALL can be predicted using a small number of genes by qPCR or low-density arrays, we performed simulation analysis on the RNA-Seq based gene expression data by selecting the top differentially expressed genes between C1 and C2. We developed a simple score as the difference between the sum of top upregulated genes in C2 and the sum of top upregulated genes in C1. The number of top genes used varied between 10 and 50. A high score indicates the C2 subtype, and a lower score indicates the C1 subtype.

Whole genome sequencing, whole exome sequencing, and SNP array

Whole genome sequencing and whole exome sequencing libraries were constructed using the TruSeq DNA PCR-Free sample preparation kit (Illumina) and the TruSeq exome enrichment kit (Illumina), respectively. Libraries were sequenced using the Illumina HiSeq 2000/2500 or NovaSeq platform. Reads were aligned to hg19 genome reference using bwa46 (v 0.7.12-r1039), and single-nucleotide variants and small indels were called using Bambino47. Copy number variants were called from whole genome sequencing data using CONSERTING48, or from SNP 6.0 arrays (Affymetrix) using reference normalization49 and circular binary segmentation50. To perform mutation signature analysis, single-nucleotide variants were grouped into the 96 trinucleotide categories and assigned to the Catalogue Of Somatic Mutations In Cancer (COSMIC) signatures using SigProfiler51 (v2.5.1.7).

Ex vivo pharmacotyping

Ex vivo drug sensitivity was a measure for 20 drugs: mercaptopurine (n = 174 measurements), prednisolone (n = 163), vincristine (n = 158), asparaginase (n = 158), thioguanine (n = 115), dexamethasone (n = 111), cytarabine (n = 86), ibrutinib (n = 61), dasatinib (n = 61), trametinib (n = 59), bortezomib (n = 57), venetoclax (n = 56), daunorubicin (n = 56), vorinostat (n = 50), CHZ868 (n = 46), nelarabine (n = 44), inotuzumab (n = 33), ruxolitinib (n = 32), palbociclib (n = 24), gilteritinib (n = 24). Three different assays were used to measure the concentration of drug lethal to 50% of the cells (LC50): (1) a 4-day in vitro 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide (MTT) drug resistance assay52,53, (2) a mesenchymal stem cell (MSC) co-culture system assay with flow cytometry54 or, (3) an imaging-based assay55. LC50 values were log-transformed and scaled to the range between 0 and 1 as previously described21, and quantile normalized when multiple assays were used to test the same drug.

Cell culturing

All the cell lines were maintained in 37 °C humidified cell culture incubators with 5% CO2. REH cell line was purchased from ATCC (CRL-8286) and cultured in RPMI1640 medium (GIBCO, #11875-093) supplemented with 10% fetal bovine serum (FBS, Hyclone, # SH30071.03). Lenti-X 293T cells were purchased from Takara (#632180) and cultured in DMEM medium (Sigma-Aldrich, #D6429) supplemented with 10% FBS. Cell numbers and viability were analyzed using Countess 3 (Invitrogen).

Plasmid construction and lentivirus production

PAX5 fragment was amplified by polymerase chain reaction (PCR) and then cloned into cl20c-IRES-GFP lentiviral plasmid using NEBuilder HiFi DNA Assembly Master Mix (NEB, #E2621L). The sequences of the cloning primers are as follows: PAX5-F, 5′-TCCTTCTCTAGGCGCCGGAATTCATGGATTTAGAGAAAAATTATC-3′; PAX5-R, 5′-TCTGCATGGATCCCTAGGAATTCTCAGTGACGGTCATAGGCAG-3′ (synthesized by Integrated DNA Technologies). Purified cl20c-IRES-GFP empty vector and cl20c-PAX5-IRES-GFP plasmids were respectively transfected together with packaging vectors into Lenti-X 293T cells (Takara, #632180). After 48 h of transfection, the culture media containing lentivirus particles were collected to infect the REH cell line for another 48 h. The transduction efficiencies were determined by flow cytometry.

Western blotting

1 × 106 cells for each group were harvested and washed once with cold PBS buffer. The cells were lysed with 100 µl RIPA lysis buffer (Thermo Scientific, #89901). The protein concentrations of cell lysate were determined with a BCA assay kit (Thermo Scientific, #23228). 20–40 µg protein for each sample was boiled and subjected to SDS-PAGE running. After transferring and blocking steps, the PVDF membranes were blotted with anti-PAX5 antibody (Abcam, polyclonal antibody, #ab15164, Lot No.GR3217582-1, 1:1000 dilution in TBST buffer with 3% BSA) to detect the overexpressed PAX5 proteins. Anti-Vinculin antibody (CST, recombinant monoclonal antibody, #13901, Lot No.7, 1:1000 dilution in TBST buffer with 3% BSA) was used to detect endogenous vinculin levels.

Drug sensitivity assessment

To examine the sensitivity to mercaptopurine, REH cells were seeded into a 384-well microplate (PerkinElmer, #6007680) with 2000 cells per well. Gradient-diluted mercaptopurine (Millipore Sigma, #85267-5G-A) was added into the wells of 384-well microplates (Thermo Scientific, #242765) and incubated for 72 h. CellTiter-Glo (CTG, Promega, G7573) assay was used to test cell viability after drug treatment. The luminescence values were determined by Synergy H4 hybrid reader (Biotek). To evaluate the impact of palbociclib on mercaptopurine sensitivity, REH cells were pre-treated with 1 µM palbociclib for 24 h, and then treated with a gradient dilution of mercaptopurine while maintaining 1 µM palbociclib in the medium for 72 h, followed by CTG assay. As comparison, the cells treated with DMSO were also tested for the LC50 of mercaptopurine.

DNA-TG level examination

Genomic DNA was extracted and purified from REH cells using the DNeasy Blood & Tissue Kit (QIAGEN, #69506). Samples were diluted to 10 ng/µl into ultrapure water. All enzymes were from Sigma. DNA was digested into deoxynucleosides for 24 h at 37 °C with an enzyme cocktail of benzonase (250 units, U, E1014), phosphodiesterase I (300 mU, P3243), and alkaline phosphatase (200 U, P7923) diluted in Tris-HCl (20 mM, pH 7.9, Lonza, 51238), 100 mM NaCl (Thermo Scientific, J61890), 20 mM MgCl2 (Thermo Fisher Scientific, AM9530G) and 100 mM DL-dithiothreitol (DTT, Sigma, D0632). The reaction was stopped with ethylenediaminetetraacetic acid (EDTA, 5 mM, pH 8.0, Thermo Fisher Scientific, AM9260G). 6-thio-2′-deoxyguanosine (dTG) was quantified by high-performance liquid chromatography coupled with tandem mass spectrometry using in-house synthesized 2′-deoxy-6-thio-guanosine-13C2,15 N as the internal standard. DNA-TG concentration was adjusted to total DNA (fmol/ug DNA).

CellTrace proliferation assay

The CellTrace proliferation assay was performed following the instruction of the CellTrace™ Far Red Cell Proliferation Kit (Thermo Scientific, C34564). After centrifugation at 450 × g for 5 min, REH cells were washed with PBS once. The cell pellet was gently resuspended with CellTrace™ Far Red (1:1000 dilution) staining solution and incubated at 37 °C for 20 min. After that, a complete culture medium was added and incubated at 37 °C for 5 min. Next, the cells were centrifuged and resuspended in a fresh, pre-warmed complete culture medium. The cells were harvested at different time points (Day 0 to Day 5) and analyzed by flow cytometry (BD LSRFortessa™ Cell Analyzer).

BrdU incorporation assay

BrdU incorporation rate was analyzed using Apoptosis, DNA Damage, and Cell Proliferation Kit (BD, #562253). Briefly, 1 × 106 cells of each group were pulsed with 1 mM BrdU in a complete culture medium and incubated at 37 °C for 1 h. Then the centrifuged cells were fixed and permeated with BD Cytofix/Cytoperm Fixation/Permeabilization Solution and BD Cytofix/Cytoperm Plus Permeabilization Buffer. After that, the cells were pelleted and refixed in BD Cytofix/Cytoperm Fixation/Permeabilization Solution. After washing with 1× BD Perm/Wash Buffer, the cells were digested with DNase (30 μg for each sample) for 1 h at 37 °C. Next, the harvested cell pellet was resuspended in 1× BD Perm/Wash Buffer and stained with PerCP-Cy™5.5 Anti-BrdU antibody (5 μl/test) for 20 min at room temperature. After washing with 1× BD Perm/Wash Buffer, the cells were responded in Staining Buffer (PBS supplemented with 2% FBS) and analyzed by flow cytometry (BD LSRFortessa™ Cell Analyzer).

Statistical analysis

Unless otherwise specified, the association between two categorical variables was tested using the χ2 test, and the association between a categorical variable and a continuous variable was tested using the Mann–Whitney test. Comparison between groups of cells in functional experiments was performed using a t-test. All statistical tests are unpaired and two sided. A P-value below 0.05 is considered statistical significance. For the comparison of the 20 drugs between the two subtypes, multiple testing was controlled using Bonferroni correction. For GSEA analysis, a false discovery rate was used to correct for multiple testing. Statistical analysis was performed using R (v4.3.0) on the Linux operating system, except for analysis related to the REH cell line experiment, which was analyzed using GraphPad Prism (v9.5.0) on Windows.

Inclusion & ethics statement

This study complies with all relevant ethical regulations and was approved by the local Institutional Review Board or Ethnics Committee. All co-authors of this study fulfill the authorship criteria of Nature Portfolio journals. The collaborators participated in the study design, study implementation, data collection, data analysis, manuscript writing, and approval. The study is locally relevant and has this been determined in collaboration with local partners. The roles and responsibilities were agreed amongst collaborators ahead of the research. This research was not severely restricted or prohibited in the setting of the researchers, nor did it result in stigmatization, incrimination, discrimination, or personal risk to participants. Local and regional research relevant to our study was considered in citations.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.