Molecular classification of cancer: class discovery and class prediction by gene expression monitoring.

Although cancer classification has improved over the past 30 years, there has been no general approach for identifying new cancer classes (class discovery) or for assigning tumors to known classes (class prediction). Here, a generic approach to cancer classification based on gene expression monitoring by DNA microarrays is described and applied to human acute leukemias as a test case. A class discovery procedure automatically discovered the distinction between acute myeloid leukemia (AML) and acute lymphoblastic leukemia (ALL) without previous knowledge of these classes. An automatically derived class predictor was able to determine the class of new leukemia cases. The results demonstrate the feasibility of cancer classification based solely on gene expression monitoring and suggest a general strategy for discovering and predicting cancer classes for other types of cancer, independent of previous biological knowledge.

Genome-wide single-nucleotide polymorphism analysis in juvenile myelomonocytic leukemia identifies uniparental disomy surrounding the NF1 locus in cases associated with neurofibromatosis but not in cases with mutant RAS or PTPN11.

Juvenile myelomonocytic leukemia (JMML) is a malignant hematopoietic disorder whose proliferative component is a result of RAS pathway deregulation caused by somatic mutation in the RAS or PTPN11 oncogenes or in patients with underlying neurofibromatosis type 1 (NF-1), by loss of NF1 gene function. To search for potential collaborating genetic abnormalities, we used oligonucleotide arrays to analyse over 116 000 single-nucleotide polymorphisms across the genome in 16 JMML samples with normal karyotype. Evaluation of the SNP genotypes identified large regions of homozygosity on chromosome 17q, including the NF1 locus, in four of the five samples from patients with JMML and NF-1. The homozygous region was at least 55 million base pairs in each case. The genomic copy number was normal within the homozygous region, indicating uniparental disomy (UPD). In contrast, the array data provided no evidence for 17q UPD in any of the 11 JMML cases without NF-1. We used array-based comparative genomic hybridization to confirm 17q disomy, and microsatellite analysis was performed to verify homozygosity. Mutational analysis demonstrated that the inactivating NF1 lesion was present on both alleles in each case. In summary, our data indicate that a mitotic recombination event in a JMML-initiating cell led to 17q UPD with homozygous loss of normal NF1, provide confirmatory evidence that the NF1 gene is crucial for the increased incidence of JMML in NF-1 patients, and corroborate the concept that RAS pathway deregulation is central to JMML pathogenesis.

Pediatric acute myeloid leukemia with NPM1 mutations is characterized by a gene expression profile with dysregulated HOX gene expression distinct from MLL-rearranged leukemias.

Somatic mutations in nucleophosmin (NPM1) occur in approximately 35% of adult acute myeloid leukemia (AML). To assess the frequency of NPM1 mutations in pediatric AML, we sequenced NPM1 in the diagnostic blasts from 93 pediatric AML patients. Six cases harbored NPM1 mutations, with each case lacking common cytogenetic abnormalities. To explore the phenotype of the AMLs with NPM1 mutations, gene expression profiles were obtained using Affymetrix U133A microarrays. NPM1 mutations were associated with increased expression of multiple homeobox genes including HOXA9, A10, B2, B6 and MEIS1. As dysregulated homeobox gene expression is also a feature of MLL-rearranged leukemia, the gene expression signatures of NPM1-mutated and MLL-rearranged leukemias were compared. Significant differences were identified between these leukemia subtypes including the expression of different HOX genes, with NPM1-mutated AML showing higher levels of expression of HOXB2, B3, B6 and D4. These results confirm recent reports of perturbed HOX expression in NPM1-mutated adult AML, and provide the first evidence that the NPM1-mutated signature is distinct from MLL-rearranged AML. These findings suggest that mutated NPM1 leads to dysregulated HOX expression via a different mechanism than MLL rearrangement.

MN1 overexpression is an important step in the development of inv(16) AML.

The gene encoding the transcriptional co-activator MN1 is the target of the reciprocal chromosome translocation (12;22)(p13;q12) in some patients with acute myeloid leukemia (AML). In addition, expression array analysis showed that MN1 was overexpressed in AML specified by inv(16), in some AML overexpressing ecotropic viral integration 1 site (EVI1) and in some AML without karyotypic abnormalities. Here we describe that mice receiving transplants of bone marrow (BM) overexpressing MN1 rapidly developed myeloproliferative disease (MPD). This BM also generated myeloid cell lines in culture. By mimicking the situation in human inv(16) AML, forced coexpression of MN1 and Cbfbeta-SMMHC rapidly caused AML in mice. These findings identify MN1 as a highly effective hematopoietic oncogene and suggest that MN1 overexpression is an important cooperative event in human inv(16) AML.

A recessive cellular mutation in v-fes-transformed mink cells restores contact inhibition and anchorage-dependent growth.

A contact-inhibited revertant of mink cells transformed by the Gardner-Arnstein strain of feline sarcoma virus was isolated by fluorescence-activated sorting of cells stained with the mitochondria-specific dye rhodamine 123. The revertant cell line exhibited a decrease in its proliferative rate and saturation density and a complete loss of its capacity for anchorage-independent growth, but it remained tumorigenic when inoculated into nude mice. The revertant cells retained a rescuable Gardner-Arnstein feline sarcoma provirus, expressed high levels of the v-fes oncogene product and its associated tyrosine kinase activity, manifested elevated levels of phosphotyrosine-containing cellular proteins similar to those observed in v-fes-transformed cells, and were refractory to retransformation by retroviruses containing the v-fes, v-fms, and v-ras oncogenes. Fusion of the revertant and parental cells generated somatic cell hybrids which formed colonies in semisolid medium, indicating that the block in transformation was recessive. These data together with the observation that the revertant phenotype is unstable in continuous culture suggest that the loss of transformation is due to the presence of limiting quantities of a gene product which functions downstream of the v-fes-coded kinase in the mitogenic pathway.

Ligand-induced tyrosine kinase activity of the colony-stimulating factor 1 receptor in a murine macrophage cell line.

Metabolic labeling of simian virus 40-immortalized murine macrophages with 32Pi and immunoblotting with antibodies to phosphotyrosine demonstrated that the c-fms proto-oncogene product (colony-stimulating factor 1 [CSF-1] receptor) was phosphorylated on tyrosine in vivo and rapidly degraded in response to CSF-1. Stimulation of the CSF-1 receptor also induced immediate phosphorylation of several other cellular proteins on tyrosine. By contrast, the mature cell surface glycoprotein encoded by the v-fms oncogene was phosphorylated on tyrosine in the absence of CSF-1, suggesting that it functions as a ligand-independent kinase.

Ligand and protein kinase C downmodulate the colony-stimulating factor 1 receptor by independent mechanisms.

The turnover of the colony-stimulating factor 1 receptor (CSF-1R), the c-fms proto-oncogene product, is accelerated by ligand binding or by activators of protein kinase C (PKC), such as the phorbol ester 12-O-tetradecanoylphorbol-13-acetate (TPA). The mechanisms of ligand- and TPA-induced downmodulation were shown to differ by the following criteria. First, in cells in which PKC was downmodulated, CSF-1R reexpressed at the cell surface remained sensitive to ligand but was refractory to TPA-induced degradation. Second, a kinase-defective receptor containing a methionine-for-lysine substitution at amino acid 616 at its ATP-binding site failed to undergo ligand-induced downmodulation but remained responsive to TPA. Following CSF-1 stimulation, no intermediates of receptor degradation could be immunoprecipitated with polyvalent antisera to CSF-1R. In contrast, TPA induced specific proteolytic cleavage of the receptor near its transmembrane segment, resulting in the release of the extracellular ligand-binding domain from the cell and the generation of an intracellular fragment containing the kinase domain. Two-dimensional phosphopeptide mapping demonstrated no new sites of phosphorylation in response to TPA in either the residual intact receptor or the intracellular proteolytic fragment. Therefore, PKC appears not to trigger downmodulation by directly phosphorylating the receptor but, rather, activates a protease which recognizes CSF-1R as a substrate.

Identification of AML-1 and the (8;21) translocation protein (AML-1/ETO) as sequence-specific DNA-binding proteins: the runt homology domain is required for DNA binding and protein-protein interactions.

The AML1 gene on chromosome 21 is disrupted in the (8;21)(q22;q22) translocation associated with acute myelogenous leukemia and encodes a protein with a central 118-amino-acid domain with 69% homology to the Drosophila pair-rule gene, runt. We demonstrate that AML-1 is a DNA-binding protein which specifically interacts with a sequence belonging to the group of enhancer core motifs, TGT/cGGT. Electrophoretic mobility shift analysis of cell extracts identified two AML-1-containing protein-DNA complexes whose electrophoretic mobilities were slower than those of complexes formed with AML-1 produced in vitro. Mixing of in vitro-produced AML-1 with cell extracts prior to gel mobility shift analysis resulted in the formation of higher-order complexes. Deletion mutagenesis of AML-1 revealed that the runt homology domain mediates both sequence-specific DNA binding and protein-protein interactions. The hybrid product, AML-1/ETO, which results from the (8;21) translocation and retains the runt homology domain, both recognizes the AML-1 consensus sequence and interacts with other cellular proteins.

Peptide antisera to human colony-stimulating factor 1 receptor detect ligand-induced conformational changes and a binding site for phosphatidylinositol 3-kinase.

A peptide antiserum (anti-A) directed to the intracellular, juxtamembrane region (residues 552 to 574) of the human colony-stimulating factor 1 receptor (CSF-1R) precipitated only ligand-activated, native receptors from solution but bound to unstimulated forms after their denaturation. Two peptide antisera (anti-KI1 and -KI2), directed to residues 679 to 700 and 701 to 721, respectively, in the CSF-1R kinase insert (KI) domain and including mapped sites of ligand-induced phosphorylation at Tyr-699 and Tyr-708, bound at least 80% of the receptor molecules expressed in either CSF-1-stimulated or unstimulated cells. Immune complexes formed with anti-KI1, anti-A, or a peptide antiserum to the CSF-1R carboxyl terminus (anti-C-ter) coprecipitated CSF-1R complexed to a phosphatidylinositol 3-kinase (PtdIns 3-K) from CSF-1-stimulated cells, whereas anti-KI2 serum did not. In an in vitro assay, binding of CSF-1R to PtdIns 3-K required receptor tyrosine phosphorylation but not CSF-1R-mediated phosphorylation of the lipid kinase, and the association was specifically blocked by anti-KI2 or antibodies to phosphotyrosine. Neither anti-KI1, anti-A, nor anti-C-ter serum inhibited binding. We conclude that (i) only a minority of ligand-activated receptors form a stable complex with PtdIns 3-K in vivo, (ii) efficient binding of the lipid kinase requires receptor tyrosine phosphorylation within the CSF-1R KI domain, and (iii) a region within the KI domain defined by residues 701 to 721 at least partially overlaps the PtdIns 3-K binding site.

Ligand-induced phosphorylation of the colony-stimulating factor 1 receptor can occur through an intermolecular reaction that triggers receptor down modulation.

Ligand-induced tyrosine phosphorylation of the human colony-stimulating factor 1 receptor (CSF-1R) could involve either an intra- or intermolecular mechanism. We therefore examined the ability of a CSF-1R carboxy-terminal truncation mutant to phosphorylate a kinase-defective receptor, CSF-1R[met 616], that contains a methionine-for-lysine substitution at its ATP-binding site. By using an antipeptide serum that specifically reacts with epitopes deleted from the enzymatically competent truncation mutant, cross-phosphorylation of CSF-1R[met 616] on tyrosine was demonstrated, both in immune-complex kinase reactions and in intact cells stimulated with CSF-1. Both in vitro and in vivo, CSF-1R[met 616] was phosphorylated on tryptic peptides identical to those derived from wild-type CSF-1R, suggesting that receptor phosphorylation on tyrosine can proceed via an intermolecular interaction between receptor monomers. When expressed alone, CSF-1R[met 616] did not undergo ligand-induced down modulation, but its phosphorylation in cells coexpressing the kinase-active truncation mutant accelerated its degradation.

Both TEL and AML-1 contribute repression domains to the t(12;21) fusion protein.

t(12;21) is the most frequent translocation found in pediatric B-cell acute lymphoblastic leukemias. This translocation fuses a putative repressor domain from the TEL DNA-binding protein to nearly all of the AML-1B transcription factor. Here, we demonstrate that fusion of the TEL pointed domain to the GAL4 DNA-binding domain resulted in sequence-specific transcriptional repression, indicating that the pointed domain is a portable repression motif. The TEL pointed domain functioned equally well when the GAL4 DNA-binding sites were moved 600 bp from the promoter, suggesting an active mechanism of repression. This lead us to demonstrate that wild-type TEL and the t(12;21) fusion protein bind the mSin3A corepressor. In the fusion protein, both TEL and AML-1B contribute mSin3 interaction domains. Deletion mutagenesis indicated that both the TEL and AML-1B mSin3-binding domains contribute to repression by the fusion protein. While both TEL and AML-1B associate with mSin3A, TEL/AML-1B appears to bind this corepressor much more stably than either wild-type protein, suggesting a mode of action for the t(12;21) fusion protein.

HERF1, a novel hematopoiesis-specific RING finger protein, is required for terminal differentiation of erythroid cells.

The AML1/core binding factor beta (CBFbeta) transcription factor is essential for definitive hematopoiesis; however, the downstream pathways through which it functions remain incompletely defined. Using a differential cloning approach to define components of this pathway, we have identified a novel gene designated HERF1 (for hematopoietic RING finger 1), whose expression during development is dependent on the presence of functional AML1/CBFbeta. HERF1 contains a tripartite RING finger-B box-alpha-helical coiled-coil domain and a C-terminal region homologous to the ret proto-oncogene-encoded finger protein. Expression of HERF1 during embryogenesis coincides with the appearance of definitive erythropoiesis and in adult mice is restricted to erythroid cells, increasing 30-fold during terminal differentiation. Importantly, inhibition of HERF1 expression blocked terminal erythroid differentiation of the murine erythroleukemia cell line MEL, whereas its overexpression induced erythroid maturation. These results suggest an important role for this protein in erythropoiesis.

The t(8;21) fusion product, AML-1-ETO, associates with C/EBP-alpha, inhibits C/EBP-alpha-dependent transcription, and blocks granulocytic differentiation.

AML-1B is a hematopoietic transcription factor that is functionally inactivated by multiple chromosomal translocations in human acute myeloblastic and B-cell lymphocytic leukemias. The t(8;21)(q22;q22) translocation replaces the C terminus, including the transactivation domain of AML-1B, with ETO, a nuclear protein of unknown function. We previously showed that AML-1-ETO is a dominant inhibitor of AML-1B-dependent transcriptional activation. Here we demonstrate that AML-1-ETO also inhibits C/EBP-alpha-dependent activation of the myeloid cell-specific, rat defensin NP-3 promoter. AML-1B bound the core enhancer motifs present in the NP-3 promoter and activated transcription approximately sixfold. Similarly, C/EBP-alpha bound NP-3 promoter sequences and activated transcription approximately sixfold. Coexpression of C/EBP-alpha with AML-1B or its family members, AML-2 and murine AML-3, synergistically activated the NP-3 promoter up to 60-fold. The t(8;21) product, AML-1-ETO, repressed AML-1B-dependent activation of NP-3 and completely inhibited C/EBP-alpha-dependent activity as well as the synergistic activation. In contrast, the inv(16) product, which indirectly targets AML family members by fusing their heterodimeric DNA binding partner, CBF-beta, to the myosin heavy chain, inhibited AML-1B but not C/EBP-alpha activation or the synergistic activation. AML-1-ETO and C/EBP-alpha were coimmunoprecipitated and thus physically interact in vivo. Deletion mutants demonstrated that the C terminus of ETO was required for AML-1-ETO-mediated repression of the synergistic activation but not for association with C/EBP-alpha. Finally, overexpression of AML-1-ETO in myeloid progenitor cells prevented granulocyte colony-stimulating factor-induced differentiation. Thus, AML-1-ETO may contribute to leukemogenesis by specifically inhibiting C/EBP-alpha- and AML-1B-dependent activation of myeloid promoters and blocking differentiation.

Biological characteristics of the leukemia-associated transcriptional factor AML1 disclosed by hematopoietic rescue of AML1-deficient embryonic stem cells by using a knock-in strategy.

AML1 is one of the most frequently mutated genes associated with human acute leukemia and encodes the DNA-binding subunit of the heterodimering transcriptional factor complex, core-binding factor (CBF) (or polyoma enhancer binding protein 2 [PEBP2]). A null mutation in either AML1 or its dimerizing partner, CBFbeta, results in embryonic lethality secondary to a complete block in fetal liver hematopoiesis, indicating an essential role of this transcription complex in the development of definitive hematopoiesis. The hematopoietic phenotype that results from the loss of AML1 can be replicated in vitro with a two-step culture system of murine embryonic stem (ES) cells. Using this experimental system, we now demonstrate that this hematopoietic defect can be rescued by expressing the PEBP2alphaB1 (AML1b) isoform under the endogenous AML1-regulatory sequences through a knock-in (targeted insertion) approach. Moreover, we demonstrate that the rescued AML1(-/-) ES cell clones contribute to lymphohematopoiesis within the context of chimeric animals. Rescue requires the transcription activation domain of AML1 but does not require the C-terminal VWRPY motif, which is conserved in all AML1 family members and has been shown to interact with the transcriptional corepressor, Groucho/transducin-like Enhancer of split. Taken together, these data provide compelling evidence that the phenotype seen in AML1-deficient mice is due solely to the loss of transcriptionally active AML1.

The t(12;21) translocation converts AML-1B from an activator to a repressor of transcription.

The t(12;21) translocation is present in up to 30% of childhood B-cell acute lymphoblastic and fuses a potential dimerization motif from the ets-related factor TEL to the N terminus of AML1. The t(12;21) translocation encodes a 93-kDa fusion protein that localizes to a high-salt- and detergent-resistant nuclear compartment. This protein binds the enhancer core motif, TGTGGT, and interacts with the AML-1-binding protein, core-binding factor beta. Although TEL/AML-1B retains the C-terminal domain of AML-1B that is required for transactivation of the T-cell receptor beta enhancer, it fails to activate transcription but rather inhibits the basal activity of this enhancer. TEL/AML-1B efficiently interferes with AML-1B dependent transactivation of the T-cell receptor beta enhancer, and coexpression of wild-type TEL does not reverse this inhibition. The N-terminal TEL helix-loop-helix domain is essential for TEL/AML-1B-mediated repression. Thus, the t(12;21) fusion protein dominantly interferes with AML-1B-dependent transcription, suggesting that the inhibition of expression of AML-1 genes is critical for B-cell leukemogenesis.

ETO, a target of t(8;21) in acute leukemia, makes distinct contacts with multiple histone deacetylases and binds mSin3A through its oligomerization domain.

t(8;21) and t(16;21) create two fusion proteins, AML-1-ETO and AML-1-MTG16, respectively, which fuse the AML-1 DNA binding domain to putative transcriptional corepressors, ETO and MTG16. Here, we show that distinct domains of ETO contact the mSin3A and N-CoR corepressors and define two binding sites within ETO for each of these corepressors. In addition, of eight histone deacetylases (HDACs) tested, only the class I HDACs HDAC-1, HDAC-2, and HDAC-3 bind ETO. However, these HDACs bind ETO through different domains. We also show that the murine homologue of MTG16, ETO-2, is also a transcriptional corepressor that works through a similar but distinct mechanism. Like ETO, ETO-2 interacts with N-CoR, but ETO-2 fails to bind mSin3A. Furthermore, ETO-2 binds HDAC-1, HDAC-2, and HDAC-3 but also interacts with HDAC-6 and HDAC-8. In addition, we show that expression of AML-1-ETO causes disruption of the cell cycle in the G(1) phase. Disruption of the cell cycle required the ability of AML-1-ETO to repress transcription because a mutant of AML-1-ETO, Delta469, which removes the majority of the corepressor binding sites, had no phenotype. Moreover, treatment of AML-1-ETO-expressing cells with trichostatin A, an HDAC inhibitor, restored cell cycle control. Thus, AML-1-ETO makes distinct contacts with multiple HDACs and an HDAC inhibitor biologically inactivates this fusion protein.

The E2A-HLF oncoprotein activates Groucho-related genes and suppresses Runx1.

The E2A-HLF fusion gene, formed by the t(17;19)(q22;p13) chromosomal translocation in leukemic pro-B cells, encodes a chimeric transcription factor consisting of the transactivation domain of E2A linked to the bZIP DNA-binding and protein dimerization domain of hepatic leukemia factor (HLF). This oncoprotein blocks apoptosis induced by growth factor deprivation or irradiation, but the mechanism for this effect remains unclear. We therefore performed representational difference analysis (RDA) to identify downstream genetic targets of E2A-HLF, using a murine FL5.12 pro-B cell line that had been stably transfected with E2A-HLF cDNA under the control of a zinc-regulated metallothionein promoter. Two RDA clones, designated RDA1 and RDA3, were differentially upregulated in E2A-HLF-positive cells after zinc induction. The corresponding cDNAs encoded two WD40 repeat-containing proteins, Grg2 and Grg6. Both are related to the Drosophila protein Groucho, a transcriptional corepressor that lacks DNA-binding activity on its own but can act in concert with other proteins to regulate embryologic development of the fly. Expression of both Grg2 and Grg6 was upregulated 10- to 50-fold by E2A-HLF. Immunoblot analysis detected increased amounts of two additional Groucho-related proteins, Grg1 and Grg4, in cells expressing E2A-HLF. A mutant E2A-HLF protein with a disabled DNA-binding region also mediated pro-B cell survival and activated Groucho-related genes. Among the transcription factors known to interact with Groucho-related protein, only RUNX1 was appreciably downregulated by E2A-HLF. Our results identify a highly conserved family of transcriptional corepressors that are activated by E2A-HLF, and they suggest that downregulation of RUNX1 may contribute to E2A-HLF-mediated leukemogenesis.

Genome-wide approach to identify risk factors for therapy-related myeloid leukemia.

Using a target gene approach, only a few host genetic risk factors for treatment-related myeloid leukemia (t-ML) have been defined. Gene expression microarrays allow for a more genome-wide approach to assess possible genetic risk factors for t-ML. We assessed gene expression profiles (n=12 625 probe sets) in diagnostic acute lymphoblastic leukemic cells from 228 children treated on protocols that included leukemogenic agents such as etoposide, 13 of whom developed t-ML. Expression of 68 probes, corresponding to 63 genes, was significantly related to risk of t-ML. Hierarchical clustering of these probe sets clustered patients into three groups with 94, 122 and 12 patients, respectively; 12 of the 13 patients who went on to develop t-ML were overrepresented in the latter group (P<0.0001). A permutation test indicated a low likelihood that these probe sets and clusters were obtained by chance (P<0.001). Distinguishing genes included transcription-related oncogenes (v-Myb, Pax-5), cyclins (CCNG1, CCNG2 and CCND1) and histone HIST1H4C. Common transcription factor recognition elements among similarly up- or downregulated genes included several involved in hematopoietic differentiation or leukemogenesis (Maz, PU.1, ARNT). This approach has identified several genes whose expression distinguishes patients at risk of t-ML, and suggests targets for assessing germline predisposition to leukemogenesis.

AMKL chimeric transcription factors are potent inducers of leukemia.

Acute megakaryoblastic leukemia in patients without Down syndrome is a rare malignancy with a poor prognosis. RNA sequencing of fourteen pediatric cases previously identified novel fusion transcripts that are predicted to be pathological including CBFA2T3-GLIS2, GATA2-HOXA9, MN1-FLI and NIPBL-HOXB9. In contrast to CBFA2T3-GLIS2, which is insufficient to induce leukemia, we demonstrate that the introduction of GATA2-HOXA9, MN1-FLI1 or NIPBL-HOXB9 into murine bone marrow induces overt disease in syngeneic transplant models. With the exception of MN1, full penetrance was not achieved through the introduction of fusion partner genes alone, suggesting that the chimeric transcripts possess a unique gain-of-function phenotype. Leukemias were found to exhibit elements of the megakaryocyte erythroid progenitor gene expression program, as well as unique leukemia-specific signatures that contribute to transformation. Comprehensive genomic analyses of resultant murine tumors revealed few cooperating mutations confirming the strength of the fusion genes and their role as pathological drivers. These models are critical for both the understanding of the biology of disease as well as providing a tool for the identification of effective therapeutic agents in preclinical studies.

Clinical impact of minimal residual disease in children with different subtypes of acute lymphoblastic leukemia treated with Response-Adapted therapy.

To determine the clinical significance of minimal residual disease (MRD) in patients with prognostically relevant subtypes of childhood acute lymphoblastic leukemia (ALL), we analyzed data from 488 patients treated in St Jude Total Therapy Study XV with treatment intensity based mainly on MRD levels measured during remission induction. MRD levels on day 19 predicted treatment outcome for patients with hyperdiploid >50 ALL, National Cancer Institute (NCI) standard-risk B-ALL or T-cell ALL, while MRD levels on day 46 were prognostic for patients with NCI standard-risk or high-risk B-ALL. Patients with t(12;21)/(ETV6-RUNX1) or hyperdiploidy >50 ALL had the best prognosis; those with a negative MRD on day 19 had a particularly low risk of relapse: 1.9% and 3.8%, respectively. Patients with NCI high-risk B-ALL or T-cell ALL had an inferior outcome; even with undetectable MRD on day 46, cumulative risk of relapse was 12.7% and 15.5%, respectively. Among patients with NCI standard-risk B-ALL, the outcome was intermediate overall but was poor if MRD was ⩾1% on day 19 or MRD was detectable at any level on day 46. Our results indicate that the clinical impact of MRD on treatment outcome in childhood ALL varies considerably according to leukemia subtype and time of measurement.