We investigated 230 adult T-ALL patients enrolled in GMALL registries and trials with a median age of 32 years (range 17-83 years); of these 11 patients (5%) were older than 55 years. 166 (72%) patients were male: 107 (46.5%) patients showed an immunophenotype of a thymic T-ALL, 43 (18.7%) of a mature T-ALL, and 60 (26.1%) of an early T-ALL, for 20 patients immunophenotype was missing (Table 1).

Table 1 Cohort characterization.Transcriptome sequencing reveals nine molecular subgroups in adult T-ALL

For the molecular subgroup assignment, we built a class prediction model based on two cohorts. One part of the investigated cohort served as an internal reference cohort, consisting of 169 of the 230 adult T-ALL GMALL samples. This cohort was previously assigned based on oncogene expression and served as a training cohort. In addition, an external cohort of pediatric T-ALL cases (n = 241) published by Liu and colleagues was used as a second reference cohort [5]. We were able to assigned the predicted molecular subgroup in 175 samples (76%) based on overlapping predictions using the internal reference cohort and the external reference cohort. Of the remaining 55 samples, we assigned additional 11 samples based on their specific fusion genes and 23 samples manually classified by a clear oncogene overexpression and similarity of gene expression profiles to those of already defined samples (Supplementary Fig. S1). Twenty-one of 230 T-ALL samples remained unclassified, representing 9.1% of the total cohort (Fig. 1).

Fig. 1: Molecular subgroups in adult T-ALL.

Classification of 230 adult T-ALL patients into molecular subgroups based on RNAseq. Clinical data, fusions genes, and oncogene expression are shown.

The largest molecular subgroup in adult T-ALL was the HOXA subgroup defined by overexpression of HOXA genes in 57 samples (24.7%) (Fig. 1, Supplementary Table S4). In the majority of these cases early stem cell markers like MEF2C and BAALC were co-overexpressed compared to the expression in subgroups TLX1, TAL1/LMO, TLX3 and LMO1 (p < 0.001). The expression of stem cell genes is also a hallmark of the immature subgroup LYL1/LMO2, covering 32 samples (13.9%) and lacking expression of HOXA cluster genes. Samples with TLX1 overexpression comprised 44 samples (19.1%) harboring the defining TLX1 overexpression. Eleven of these TLX1 samples expressed NKX2, similar to the NKX2 subgroup comprising 4 samples (1.7%) lacking TLX1 overexpression. Of the TAL1/LMO subgroup (in total 45 samples), 15 samples (6.5%) had an LMO1 overexpression, while the remaining 30 samples (13%) were lacking LMO1 expression. Consecutively, we defined these two groups LMO1 and TAL1/LMO as separate entities. While samples of the TLX1 and TLX3 subgroup showed some HOXA expression, TAL1/LMO and LMO1 cases lacked overexpression of HOXA genes. TLX3 overexpression was found in 26 samples (11.3%) defining the TLX3 subgroup with absence of expression of TLX3 in all other subgroups. We identified a single sample with TAL2 overexpression. Notably, no fusion or overexpression of SPI1 was detected in our cohort.

The average age ranged from 24 years to 38 years across the molecular cohort and displayed highest in LYL1/LMO2 with a median of 39 years and lowest in NKX2 with a median of 21.5 years (Fig. 2A). In our cohort, age distribution revealed younger patients with TAL1/LMO overexpression (16-25 years: 35% versus >35 years: 3%; p = 0.001) and more LYL1/LMO2 and HOXA overexpression among older patients (16–25 years: 23% vs. >35 years: 40%; n.s.).

Fig. 2: Clinical features of molecular subgroups in adult T-ALL.

A Age distribution according to the molecular subtypes. B Immunophenotype according to molecular subgroups. Number of samples varies from (A) and (B) due to missing immunophenotypes for 15 samples. C Gender distribution in the molecular subgroups.

The subgroups indicated a clear correlation with immunophenotype reflecting the T-cell development stage (Fig. 2B). All patients in the TLX1 and NKX2 subgroup revealed a thymic immunophenotype (p < 0.001). LYL1/LMO2 (p = 0.017) and HOXA (p = 0.05) were characterized by a more immature phenotype, showing surface expression of antigens typical for an early T-ALL in 50% of samples. TLX3 was not associated with a specific immunophenotype in this cohort with 23 samples (early: n = 7, thymic: n = 10, mature: n = 6, n.s.). On the other hand, TAL1/LMO was associated with more mature T-cell development stages, reflected by a thymic and mature T-ALL immunophenotype (p = 0.002). Only a single TAL1/LMO cases had an early T-ALL immunophenotype. Twenty-three (34%) of the 67 early T-ALL samples showed an ETP-ALL immunophenotype. By note, of 23 samples with an ETP-ALL immunophenotype 11 belong to the HOXA molecular subgroups, six were in LYL1/LMO2, five in TLX3, and one could not be classified. Regarding sex distribution, no significant differences were observed between molecular subgroups (Fig. 2C).

HOXA13 as distinct T-ALL subgroup

Seven patients (3% of the total cohort) showed a distinct overexpression of HOXA13 with associated overexpression of the lncRNA HOTTIP and EVX1 (Figs. 1 and 3A). Notably, in these samples the anterior HOXA genes were not expressed on transcript level. In five out of the seven samples, fusion genes targeting the posterior HOXA complex were identified. Three cases with recurrent MIR181A1HG::HOTTIP were caused by a chromosomal t(1;7) translocation with subsequent overexpression of lncRNA HOTTIP (Fig. 3B). In two distinct detected fusion genes in HOXA13 cluster, HOXA13 and EVX1 were involved (Supplementary Table S5). All 7 samples depicted an early immunophenotype reflected by missing sCD3 expression. Four additional samples showed an exclusive HOXA13 overexpression among the HOXA genes, but were classified in the LYL1/LMO2 subgroup. All four samples were lacking EVX1 overexpression.

Fig. 3: HOXA13 cluster with recurrent mir181A1HG_HOTTIP fusion.

A Expression of posterior HOXA genes in HOXA13 cluster. B Fusion gene mir181A1HG_HOTTIP with the 5′ fusion partner breakpoint after exon partner and full usage of 3′ fusion partner HOTTIP. C Differentially expressed genes between HOXA13 and HOXA cluster. GSEA enrichment analysis for JAK/STAT and TP53 pathway compared between HOXA13 and HOXA.

HOXA13 cases showed a distinct gene expression profile compared to the HOXA samples with the majority of differentially expressed genes being downregulated in HOXA13 cluster, including expected downregulation of anterior HOXA genes (Supplementary Table S6). In gene set enrichment analyses of KEGG and Hallmark of cancer gene lists, we identified 88 differentially regulated gene sets (FDR < 0.1, Supplementary Table S7). Among those, JAK-STAT-signaling and TP53 pathway were altered in HOXA13 compared to HOXA (Fig. 3C), which have been associated with poor prognosis (TP53) and association to the unfavorable immunophenotype of early T-ALL (JAK/STAT mutations).

Underlying fusion genes drive T-ALL leukemogenesis

In 106 of the 230 T-ALL samples we detected underlying fusion genes in T-ALL (Fig. 1, Supplementary Table S5). The most frequent rearrangements in our cohort were STIL::TAL1 (n = 14) and TLX1-rearrangements (n = 15). SET::NUP214 occurred in 12 samples and was associated with the HOXA subgroup. Fusions of NKX2 were a rare events in adult T-ALL patients with a rate of 1% in the overall cohort. Further recurrent gene fusions affected among others KMT2A (n = 6), MLLT10 (n = 5) or LMO2 (n = 2) (Supplementary Table S5). In LYL1/LMO2 only few rearrangements were found, underlying the stem cell character of LYL1/LMO2 with similarities to acute myeloid leukemia (AML). Notably, all detected fusions confirmed subgroup assignment based on gene expression.

Mutational spectrum of T-ALL according to molecular subgroups

The different composition of molecular subgroups was reflected by the mutational spectrum of adult T-ALL. We investigated full protein coding sequence of 206 leukemia-associated genes (Supplementary Table S3). Twenty-two of these genes were affected at least in 3% of all samples. The most frequently altered gene was NOTCH1 with 42 out of 83 (51%) mutated samples followed by PHF6 (31/83 samples; 37%) and PTEN (15/83 samples; 18%). The small sample sizes in the molecular subgroups made firm comparisons between them difficult (see Supplementary Tables S8 and S9). However, a trend indicated that the TLX1 subgroup has a higher rate of NOTCH1 mutations, with 15 out of 21 samples (71%) being mutated, compared to the more immature molecular subgroups such as LYL1/LMO2 (5/12; 42% NOTCH1 mutated) or HOXA (4/12; 33% NOTCH1 mutated; not significant). On the other hand, genes affecting the JAK/STAT pathway (JAK1, JAK3, STAT5B, SH2B3, IL7R), were more frequently affected in the HOXA (5/12, 42%) and LYL1/LMO (7/12, 58%) subgroups compared to the TLX1 subgroup (6/21, 29%). However, these differences remained descriptive and did not reach statistical significance.

In addition, for 78 samples (all TAL1 and LMO1 samples and unclassified samples), we investigated the mutational status of the enhancer region of TAL1 with a previously reported mutation in the non-protein-coding region [31]. We only found two mutations (2/78, 3%) in the investigated cohort, both in samples of the molecular subgroup TAL1. As it is unlikely to find enhancer mutations in non-TAL1 overexpressing samples, the rate of 1% referring to the total adult T-ALL cohort seems lower than reported for pediatric patients.

Methylation patterns revealed hypomethylation in STIL::TAL1 subgroup

In 84 investigated samples, four clusters were identified based on unsupervised clustering, reflecting molecular subgroups on DNA methylation level (Fig. 4A, Supplementary Fig. S2). Cluster M1 consisted of 25 samples, comprising 8 out of 10 TLX3 samples, 8 LYL1/LMO2 samples and 8 HOXA samples. Only one NKX-1 sample was included. In terms of immunophenotype, cluster M1 reflected a more heterogeneous picture comprising high-risk features according to GMALL protocols (10 mature T-ALL samples and 8 out of 12 samples with an early immunophenotype; Fig. 4B). Cluster M2 included 20 samples, nearly exclusively TAL1/LMO samples (n = 18). Vice versa 18 of the 20 TAL1/LMO2 samples were assigned into the M2 cluster, including all samples with a TAL1 fusion (STIL::TAL1 fusion n = 5; TCF7::TAL1 fusion n = 1) showing a significant global hypomethylation in CpG islands (Fig. 4C). The largest cluster, M3, included 30 samples. Remarkably, 29 samples in this cluster showed a thymic immunophenotype and all TLX1 samples (n = 20) demonstrated a DNA methylation pattern representative of cluster M3. In the remaining samples of cluster M3 we found 6 HOXA samples, and one TLX3, NKX2-1, and TAL1/LMO sample each. The fourth cluster M4 revealed to be the smallest with only 7 samples (4 LYL1/LMO2, 2 HOXA and 1 TLX3 samples). Taken together, DNA methylation profiling enabled a robust subclassification of adult T-ALL samples. TLX1 and TAL1/LMO subgroups were characterized by a homogenous common DNA methylation pattern reflected by the assignment in the same methylation cluster (TLX1: 20/20 in M3, TAL1/LMO: 18/20 in M2), separating both subgroups from other molecular subgroups. On the other hand, TLX3, HOXA and LYL1/LMO2 had common DNA methylation patterns in cluster M1 and M4, associated with an early immunophenotype (early T-ALL: 17/19 in M1 and M4 vs. 2/19 in M2 and M3, p < 0.0001). Among the differentially methylated regions (DMRs) for each of the four clusters, defining oncogenes of the corresponding molecular subgroups were found (Fig. 4C, Supplementary Tables S10 and S11). The association of mutational events with the methylation clusters revealed a high rate of NOTCH1 mutations in cluster M3 and an increased rate of mutations in the JAK/STAT pathway and epigenetic regulators in M1 and M4 clusters (data not shown).

Fig. 4: Methylation profile of adult T-ALL samples.

A Principal component analysis of 84 T-ALL samples based on their methylation status using the 2000 most variant CpGs resulting in four stable clusters (M1–M4). B Methylation clusters showed a high concordance with molecular subgroups and immunophenotype. Methylation level of the 2000 most variant CpGs are presented with a clear hypomethylation for cluster M2. Below the graph most prominent molecular subgroups in the methylation clusters are depicted. Samples with an underlying TAL1 fusion are marked. C Number of differentially methylated regions and their methylation status according to the four methylation subgroups. Genes in the differentially methylated regions are reflecting driving oncogenes of molecular T-ALL subgroups.

MRD response according to molecular subgroup in adult T-ALL

With respect to response, 95 (68%) patients with a molecular marker achieved a molecular CR (MolCR) after consolidation I, 25 (18%) patients showed an intermediate molecular response (IntMR) and 19 (14%) failed molecular remission (MolFail) (Table 2). In the subgroup of thymic T-ALL, 65 patients (79%) achieved MolCR, 9 (11%) patients IntMR and 8 (10%) patients failed. Between the molecular subgroups, MRD after consolidation I differed significantly. Noteworthy, 37/39 (95%) of TLX1 patients had a MolCR with only two patients in TLX1 subgroup with a MolFail. On the other side, in LYL1/LMO2 only 3/9 (33%) got into MolCR. Similarly, TLX3 and TAL1/LMO subgroup showed a significantly worse response with a rate of only 50% of MolCR (Table 1). HOXA subgroups showed an intermediate response with a rate of 65% (20/37 patients) of MolCR. These response rates of the molecular subgroups are similar in patients with a thymic immunophenotype despite their favorable CD1a expression: MolCR rate in HOXA 73%, in TLX3 60% and in TAL1/LMO 40% with only limited number of patients. Thymic T-ALL patients with a TLX1 or NKX2 profile showed excellent MRD response (95% resp. 100% of MolCR; Table 2).

Table 2 Minimal residual disease in molecular subgroups.Favorable outcome in adult T-ALL is associated with TLX1, NKX2, and LMO1 subgroups

MRD response translated in OS with differences among the molecular subgroups. OS after three years differed significantly across the molecular subgroups in adult T-ALL (p = 0.0016, Fig. 5A). Among the larger T-ALL molecular subgroups, TLX1 patients (n = 44) showed an exceptional favorable 3y-OS of 92%. Although the sample size is too small to draw firm conclusions, in the small groups of NKX2-1 (n = 4) and TAL1 (n = 1), all patients were alive after three years. In addition, patients of the LMO1 subgroup (n = 14) had a very favorable 3y-OS of 92%. Patients of the HOXA (n = 47) and of the TLX3 (n = 25) subgroups with a 62% 3y-OS showed a slightly inferior prognosis (70% 3y-OS and 62% 3y-OS, respectively). In contrast, the subgroups LYL1/LMO2 (55% 3y-OS, n = 29), HOXA13 (33% 2y-OS, n = 5) and TAL1 (without LMO1 overexpression, 56% 3y-OS, n = 27) had a poorer prognosis.

Fig. 5: Overall survival for molecular subgroups in T-ALL.

(A) Overall survival for the molecular subgroups in 196 adult T-ALL patients independent of their immunophenotype. Unassigned samples are not displayed. Colors for the different subgroups are shown below the figure. B-D Overall survival of adult T-ALL patients according to their molecular risk classification. Good risk group comprises 62 patients: TLX1 (n = 44), NKX2-1 (n = 4) and LMO1 (n = 14); intermediate risk group 47 patients: HOXA (n = 47); poor risk subgroup 76 patients: HOXA13 (n = 5), TLX3 (n = 25), LYL1/LMO2 (n = 29), TAL1/LMO2 (n = 27). E Overall survival for the molecular subgroups in 99 adult T-ALL patients with thymic immunophenotype. Unassigned samples are not displayed. Colors for the different subgroups are shown below the figure.

Based on our cohort, we were able to classify T-ALL patients based on their molecular subgroups into good risk (TLX1, NKX2-1, LMO1), intermediate risk (HOXA), and poor risk (LYL1/LMO2, HOXA13, TAL1/LMO, TLX3). (Fig. 5 B-D). The favorable impact of TLX1 (92% 3y-OS, n = 44), LMO1 (89% 3y-OS, n = 8) and NKX2-1 (100% 3y-OS, n = 4) was not only found in the overall cohort but also within the already good risk subgroup of thymic T-ALL patients (Fig. 5E). In thymic T-ALL, fewer patients were assigned to the unfavorable subgroups of LYL1/LMO2 (0% 3y-OS, n = 2), TAL1 (55% 3y-OS, n = 12), TLX3 (64% 3y-OS, n = 11) and HOXA (79% 3y-OS, n = 13) compared to mature and early T-ALL. Although the relative incidences of these subgroups were lower, 45% of thymic T-ALL belonged to the HOXA, TLX3, TAL1 or LYL1/LMO2 subgroup. Therefore, a relevant percentage lacked the good prognosis. The above established molecular risk classification (Fig. 5 B–D) also allowed to stratify until now standard risk thymic T-ALL patients according to their molecular subgroup into distinct risk groups (good risk: 92% 3y-OS, intermediate risk 79% 3y-OS, poor risk 58% 3y-OS, p = 0.0022, Supplementary Fig. S3).