Using a set of genes shown to be differentially expressed in ETPs compared with more mature thymic subpopulations (Supplementary Table 1
we searched for leukemias with an ETP-related gene profile among 55 newly diagnosed cases of T-ALL. By unsupervised clustering analysis, we identified a cluster of 13 cases with gene expression profiles that strongly resembled those described for ETPs (). Overexpressed genes in this group included CD44
, while CD1
were underexpressed. We then examined the cell marker profile of the 13 clustered cases and found 9 with a strikingly distinct immature immunophenotype characterized by lack of CD1a and CD8 expression, weak CD5 expression with less than 75% positive blasts, and expression of one or more of the following myeloid or stem cell markers on at least 25% of lymphoblasts: CD117, CD34, HLA-DR, CD13, CD33, CD11b, and/or CD65. This cell marker profile clearly differed from that of most normal thymocytes (). The four other cases that clustered with the ETP cases in the unsupervised analysis of ETP gene expression had somewhat different immunophenotypic features: 3 had expression of CD5 (with absence of myeloid or stem cell markers in one case); the fourth case had very high expression of surface CD3 and TCR gamma/delta.
Figure 2 Gene expression and immunophenotypic features of ETP-ALL. (a) Unsupervised clustering analysis of diagnostic bone marrow samples from 55 pediatric patients with T-ALL, using a set of genes differentially expressed in ETP (Supplementary Table 1).11;14 (more ...)
Among 139 patients with T-ALL enrolled in three consecutive trials at St Jude Children’s Research Hospital (including the 55 studied with gene expression arrays), 17 (12.2%) had leukemic lymphoblasts with the ETP immunophenotype defined above (; Supplementary Table 2a
). CD1a and CD8 were absent in all 17 patients; this occurred in only 15 of the remaining 122 patients (p <0.0001). Second, expression of CD5 was low (10- to >200-fold lower than that of normal peripheral blood T-lymphocytes; ) resulting in a percentage of positive leukemic cells consistently below 75% (median, 45%); median CD5 expression among the 122 remaining cases was >99%, with more than 75% positive blasts in all cases, except for the one case with high TCR gamma/delta expression (Supplementary Table 2a
). Third, all 17 cases expressed CD117, CD34, HLA-DR, CD13, CD33, CD11b, and/or CD65 on at least 25% of lymphoblasts, a feature found in 60 of the remaining 122 cases (p <0.0001). TCR gene rearrangement analysis was performed in 58 of the 139 cases. It identified at least one rearranged TCR gene in 8 of the 9 ETP-ALL cases studied (), further corroborating the overall diagnosis of T-ALL. The same analysis detected at least one rearranged TCR
gene in 47 of the 49 other T-ALL cases studied (36 TCRB
, 27 TCRG
and 20 TCRD
Selected clinical and biologic presenting features of St. Jude patients with ETP-ALL
We re-examined our database of gene expression profiles from 55 cases of T-ALL, using a supervised analysis, and identified 1082 probe sets that were differentially expressed in the 9 ETP-ALL cases (FDR of 0.10; top 150 probe sets shown in and Supplementary Table 3
To determine whether the gene profile associated with ETP-ALL in the St. Jude cohort could be validated in an independent cohort, we studied the gene expression data for 34 T-ALL patients enrolled in the AIEOP ALL-2000 study, including three whose lymphoblast immunophenotypic features resembled those of the St Jude ETP cases. Class prediction analysis by PAM, with the St. Jude cohort used as a training set and the AIEOP cohort as a test set, identified all three ETP-ALL-like cases in the AIEOP cohort (Supplementary Fig. 1
). The predictive model also recognized two additional cases as ETP-ALL: both lacked CD1a and CD8, and expressed CD34 and CD117, but their expression of CD5 was not low enough to meet our stringent phenotypic criteria for ETP-ALL.
To further confirm the cell of origin of the ETP-ALL cases, we tested the gene expression signature for murine ETP, as described by Rothenberg et al.,11
together with 1687 other gene sets applying GSEA to our entire set of microarray data. As shown in Supplementary Table 4
, the set of genes upregulated in ETPs was highly enriched in ETP-ALL (p = 0.0021; FDR = 0.18), while the set of downregulated genes was highly enriched in typical T-ALL (p = 0.0062; FDR = 0.05); other highly enriched gene sets included genes also found in the ETP signature. The close correspondence between the gene expression signature of ETPs and the transformed lymphoblasts of ETP-ALL is illustrated in Supplementary Fig. 2
, which shows the combined analysis of up- and downregulated genes in ETPs (p = 0.0059; FDR = 0.13). These findings support the ETP as the likely cellular target for transformation giving rise to the cases identified in this study.
Others have used levels of expression of certain oncogenic transcription factors to identify discrete subtypes of T-ALL.7;29
Our cases of ETP-ALL had higher expression of LMO1
(p = 0.0119), LYL1
(p = 0.0002) and ERG
(p = 0.0008) () but no clear distinction between ETP- and typical T-ALL cases could be made on the basis of the expression of these transcription factors. Expression of TAL1
was not significantly different. Expression of Hairy-enhancer-of-split 1 (HES1
), a target gene in the NOTCH signaling pathway30
, did not differ significantly between the ETP- and typical T-ALL cases, suggesting that the apparently aborted T-cell differentiation of these leukemias was not associated with their inability to transduce NOTCH signals. Likewise, expression of FBXW7
, genes often mutated or deleted in T-ALL,31
did not differ between the two subgroups.
Figure 3 Genetic abnormalities in ETP-ALL. (a) Expression of transcription factors previously implicated in the pathogenesis of T-ALL7;29 in cases with an ETP origin (n = 9) or typical T-ALL (n = 46) immunophenotypes, as measured by the Affymetrix 133A GeneChip. (more ...)
The 17 cases of ETP-ALL had highly variable karyotypes (). Notably, four cases had the 13q- abnormality compared with only four of the 116 typical cases with available chromosomal findings (p = 0.0095). No other significant differences were found. We used SNP arrays to screen for genetic lesions in 11 of the 17 ETP-ALL cases and in 43 of the 122 typical cases. Together, ETP-ALLs had significantly more DNA copy number abnormalities (mean 14.1 vs. 6.3 lesions, p = 0.0033), including both genomic gains (3.4 vs. 0.7 lesions, p = 0.0005) and losses (10.7 vs. 5.6 lesions, p = 0.0134; , Supplementary Fig. 3
and Supplementary Table 5
). The overall sizes of the genomic regions spanned by these gains and deletions were significantly larger in the ETP-ALL subgroup (mean size of gains per case, 97.5 Mb vs. 28.6 Mb, p = 0.0027; losses per case, 119.0 vs. 42.4 Mb, p = 0.0068; , Supplementary Fig. 3
and Supplementary Table 5
We found no significant associations between a diagnosis of ETP-ALL and clinical presenting features, except that 13 of the 17 patients were 10 years or older, compared to 49 of the 122 patients with typical T-ALL (p = 0.0187) (Supplementary Table 6
The clearance of leukemic cells after the first phase of remission induction therapy was markedly inferior in patients with ETP-ALL: all 13 patients studied after 15 to 19 days of treatment had detectable MRD (≥0.01% leukemic cells in bone marrow), compared with 55 of the 91 having typical T-ALL (p = 0.0037; ). Levels of MRD were also significantly higher in patients with ETP-ALL: 10 of 13 had MRD ≥5% versus 12 of 91 with typical T-ALL (p <0.0001). Results of MRD measurements at the end of induction therapy (day 43) were consistent with these findings: of 14 patients with ETP-ALL who were studied at this interval, 10 had MRD compared with 28 of 116 with typical T-ALL (p = 0.0007; ). Six of the 14 patients had MRD ≥1%, indicating an extremely poor prognosis,22
in contrast to only 6 of 116 in the comparison group (p = 0.0003).
Figure 4 Prevalence of MRD during the early phases of therapy for patients with ETP or typical T-ALL. MRD levels were measured by flow cytometry (a) or by polymerase chain reaction amplification of antigen-receptor genes (b). Horizontal bars indicate median values, (more ...)
The diagnosis of ETP-ALL was associated with a significantly worse outcome (p <0.0001 by log rank tests for OS and EFS; ), with 10-year OS for ETP-ALL patients of 19% (95% CI, 0% to 92%) versus 84% (95% CI, 72% to 96%) for all remaining patients; 10-year EFS, 22% (95% CI, 5% to 49%) versus 69% (95% CI, 53% to 84%). All nine remission failures or relapses recorded in the ETP-ALL subgroup occurred in bone marrow, whereas 11 of the 22 relapses in the typical subgroup were confined to extramedullary sites. The cumulative incidence of remission failure or hematologic relapse was significantly higher in patients with ETP-ALL (p <0.0001 by Gray’s test; ). The 10-year cumulative incidence of remission failure or hematologic relapse was 72% (95% CI, 40% to 100%) for patients with ETP-ALL versus 10% (95% CI, 4% to 16%) for those with typical T-ALL. For patients who relapsed, the median time to relapse was 1.22 years for ETP-ALL and 1.74 years for typical T-ALL (p = 0.14). In univariate and multivariate analyses (), the diagnosis of ETP-ALL was by far the cofactor exerting the strongest negative impact on EFS (hazard ratio, 12.0; 95% CI, 4.6 to 31.3; p <0.0001). Five of the 17 patients received allogeneic hematopoietic stem cell transplant because of >1% MRD on day 43 (n = 4) or persistent MRD during continuation therapy: 2 patients are in complete remission 3 and 6 years post-transplant, 1 died in remission 7 months post-transplant, and 2 relapsed 0.5 and 1.5 years post-transplant.
Figure 5 Kaplan-Meier plots of (a, d) overall survival, (b, e) event-free survival, and (c, f) the cumulative incidence of remission failure or hematologic relapse in patients with typical T-ALL (gray line) versus ETP-ALL (black line) treated on either St. Jude (more ...)
Univariate and multivariate analysis of event-free survival according to the diagnosis of ETP-ALL and selected variables in St. Jude patientsa
To validate our results, we studied response to therapy in a cohort of 100 T-ALL patients (96 white, 4 of other race; 79 male, 21 female; 62 aged 1–9 years, 38 ≥10 years), enrolled in the AIEOP ALL-2000 study (including the 34 with gene expression array data). Thirteen patients (13%) had immunophenotypic features characteristic of ETP-ALL (Supplementary Table 2b
). Early development of drug resistance, a key feature of ETP-ALL, was also apparent among the AIEOP patients. Response to 1 week of prednisone alone (a strong prognostic factor in AIEOP studies32
but not evaluated in St Jude trials) was markedly inferior in patients with ETP-ALL (Supplementary Fig. 4
). The prevalence and levels of MRD in the AIEOP cohort were remarkably similar to findings in the St. Jude patients (compare ). After 78 days, 9 of the 10 patients with ETP-ALL and MRD measurements had detectable MRD, compared with only 34 of the 78 with typical T-ALL (p = 0.0067) (). As in the St. Jude cohort, the diagnosis of ETP-ALL conferred a dismal clinical outcome (; p <0.0001 by log rank tests for OS and EFS). The 2-year OS of patients with ETP ALL was 45% (95% CI, 0% to 90%) versus 90% (82% to 99%) for those with typical T-ALL; 2-year EFS was 22% (0% to 59%) versus 71% (59% to 84%). The cumulative incidence of remission failure or hematologic relapse was much higher in patients with ETP ALL (p <0.0001 by Gray’s test; ). At 2 years, it was 57% (25% to 89%) for patients with ETP ALL versus 14% (6% to 22%) for those with typical TALL. For patients who relapsed, the median time to relapse was 1.21 years for ETP-ALL and 1.37 years for typical T-ALL (p = 0.25).
The ETP phenotype also appeared to have a prognostic impact in analyses limited to MRD-positive patients: 5-year cumulative incidence of relapse for the 10 patients with ETP ALL and MRD ≥ 0.01% on day 43 in the St Jude cohort was 50% ± 17% versus 25% ± 9% for the 28 MRD-positive patients with typical T-ALL (p = 0.0173). The 2-year estimates for the AIEOP patients who had MRD ≥ 0.01% on day 78 were 69% ± 21% for the 8 with ETP ALL cases versus 32% ± 10% for the 27 with typical T-ALL (p = 0.0257).