Comprehensive integrative genomic characterization has advanced the classification of B-cell acute lymphoblastic leukemia (B-ALL) [1]. DUX4 leukemia is generally characterized by IGH::DUX4 fusions resulting in the expression of truncated DUX4 isoforms and a distinctive gene expression signature [2,3,4]. Due to the cryptic nature of the rearrangement, next-generation sequencing (NGS) approaches are optimal for accurately detecting this subtype [2,3,4]. Other characteristic features of DUX4 leukemia include a high incidence of ERG deletion, CD371 cell surface expression [2,3,4,5], and a propensity to undergo a transient switch toward the monocytic lineage (swALL) early during induction therapy [6, 7].

Several studies have shown that DUX4 confers a good prognosis, however, most patients have been treated with non-standard or high-risk protocols [8,9,10,11]. In this study, we aimed to investigate the yet unknown prognostic value of DUX4 in childhood and adolescent B-ALL patients treated with AIEOP-BFM ALL regimens and to determine potential risk factors for therapy optimization.

We retrospectively investigated 1237 B-ALL patients enrolled in consecutive clinical trials conducted in Austria over more than two decades. To identify patients belonging to the DUX4 subtype, we employed an iterative combination of genetic analyses and immunophenotyping (Supplementary Fig. S1, Supplementary Methods and Notes). Sixty-six DUX4 cases were identified by RNA-seq, three additional cases by the presence of an ERG deletion and positive IGH::DUX4-specific genomic-PCR, and one genetically undefined case showed strong CD371 and CD2 expression. As this marker combination is typical of DUX4 B-ALL [5, 12] (Supplementary Table S4), we assigned the latter patient to the DUX4 group exclusively based on immunophenotyping (Supplementary Notes).

Notably, in our B-other cohort, 9.4% (3/32) of the ERG-deleted cases belong to other genetic groups, suggesting some caution in using ERG status alone to assign patients to this subtype (Supplementary Table S4). On the other hand, we confirmed the high positive/negative predictive values of CD371 expression for DUX4 B-ALL [5] (Supplementary Table S4), making immunophenotyping a fast and reliable alternative for identifying this subtype when NGS technologies are unavailable (Supplementary Notes).

In total, we classified 70 patients as DUX4 (Table 1), accounting for roughly 19% of the B-other and 6% of the entire B-ALL cohort, which is consistent with previous population-based studies [2, 4, 10, 13].

Most of the 70 DUX4 patients were treated according to the medium risk (MRG) or high-risk (HRG) protocols of the respective clinical trials (Table 1, Supplementary Table S2). Only 7.1% (5/70) of patients relapsed, of which 80% (4/5) died, 60% (3/5) of progressive disease, indicating a need for improved salvage therapy. One additional patient died early while still on therapy, bringing the total number of deaths to 7.1% (5/70). At a median follow-up of 7.4 years (range 0.7–17.4) the estimated 5-year and 10-year event-free survival (EFS) of all patients was 92.0 ± 3.5% and 89.6 ± 4.1%, while the 5-year and 10-year overall survival (OS) was 95.1 ± 2.8% and 88.1 ± 5.7%, and the 5-year and 10-year cumulative incidence of leukemia-related events (CIL) was 8.0 ± 3.5% and 10.4 ± 4.2%, respectively (Fig. 1A). Consistent with previously reported survival rates [8,9,10,11], patients with DUX4 leukemia treated according to AIEOP-BFM ALL protocols also have a very good overall outcome.

A Kaplan–Meier survival curves of event-free (EFS) and overall survival (OS) as well as cumulative incidence of leukemia-related events (CIL) of all DUX4 cases (n = 70). B Summary of secondary genetic alterations found in DUX4 cases. mut mutation, del deletion, alt mutation and/or deletion, IKZF1plus deletion profile, swALL DUX4 ALL with monocytic lineage switch determined by flow cytometry on day 15 of treatment; depicted numbers correspond to patient numbers in Supplementary Table S2. C B-ALL multipotency score of DUX4 samples with different alterations. mut mutation, del deletion, alt mutation, and/or deletion; numbers of cases per subgroup are depicted at the bottom of the boxplots; Wilcoxon rank sum p-value. D–F Kaplan–Meier survival curves of EFS based on (D) IKZF1plus deletion profile; red, positive, blue, negative. E TP53 mutation status; red, TP53-mutated, blue, TP53 wild-type. F slow early response (SER); red, yes, blue, no. Log-ranks test p-values for EFS. G Violin plot showing the distribution of FCM-MRD blasts (combining lymphoblasts [lym] and switch blasts [sw]) in DUX4 swALL and non-swALL. H Violin plots showing the distribution of lym, sw-blasts, and monocytes (monos) in DUX4 swALL and non-swALL. I Kaplan–Meier survival curves of EFS based on DUX4 ALL with monocytic lineage switch (swALL); red, yes, blue, no. Log-ranks test p-values for EFS. J Boxplots showing B-ALL multipotency score for swALL vs. non-swALL DUX4 cases; Wilcoxon rank sum p-value.

Recently, it has been demonstrated that in B-ALL, distinct developmental states are associated with genomic alterations and clinical characteristics [14]. Therefore, we analyzed the frequency of recurrent secondary genetic alterations and their association with developmental stage and impact on outcome (Fig. 1B, Supplementary Fig. S2A, Supplementary Table S5). While, as previously shown [14], samples with RAS-MAPK-pathway mutations (37.9%, 25/66) were associated with the early lymphoid state and higher B-ALL multipotency scores, those with ERG deletion (43%, 29/68) and/or TBL1XR1 alterations (23.4%, 15/64), had significantly lower B-ALL multipotency scores (Fig. 1C), however, we did not observe any effect on outcome (Supplementary Fig. S2C–E). Neither KMT2D (12.1%, 8/66) nor ZEB2 (6.1%, 4/66) mutation was associated with any developmental state or had any impact on survival (Supplementary Fig. S2B, F, G).

Samples with IKZF1 deletion and the IKZF1plus deletion profile [15] showed higher B-ALL multipotency scores (Supplementary Fig. S2B). As previously reported by others [4, 9, 12], IKZF1 deletion alone (27.1%, 19/70) had no impact on outcome (Supplementary Fig. S2H), while IKZF1plus patients (7.2%, 5/69) had a significantly worse 5-year EFS of 60 ± 21.9% vs. 94.5 ± 3.1% (p = 0.0033) (Fig. 1D).

Notably, the IKZF1plus deletion profile coincided with TP53 mutation (60%, 3/5; Fig. 1B, Supplementary Table S6). At diagnosis, we detected TP53 mutations in 5.7% (4/70) of patients co-occurring with deletions of the second allele, and in one patient, who died shortly after relapse, a TP53 mutation was present only at this time-point. Although the number of TP53-mutated DUX4 patients in our cohort is low, they had a significantly worse outcome (5-year EFS 50 ± 25% vs. 94.5 ± 3.1%, p = <0.001; Fig. 1E), confirming their reported dismal survival [11].

The risk stratification parameters changed from the AIEOP-BFM ALL 2000 to the 2009/2017 trials. Therefore, in addition to using the actual risk groups (RGs) for outcome analysis, we performed a virtual RG classification applying the AIEOP-BFM ALL 2017 risk factors (Supplementary Fig. S3A, Supplementary Table S7).

We found no significant difference in the survival probabilities of DUX4 patients stratified into the actual RGs according to the respective protocol parameters (Supplementary Fig. S3B, Supplementary Table S5). Remarkably, while the actual PCR-based measurable residual disease (PCR-MRD) RGs showed significantly different outcomes (5-year EFS LR 100 ± 0%, IR 97.4 ± 2.5%, HR 69.6 ± 12.9%, p = 0.0061), virtual re-stratification diminished the predictive value of PCR-MRD (5-year EFS LR 100 ± 0%, IR 96.7 ± 3.3%, HR 81.8 ± 8.3%, p = 0.2; Supplementary Fig. S3C, D, Supplementary Table S5).

The changes in MRD-RG classification are based on a slow early response (SER) to therapy, defined as PCR-MRD of ≥5 × 10−4 after induction therapy and any PCR-MRD positivity <5 × 10−4 after consolidation therapy, and used as high-risk stratification factor only in the AIEOP-BFM ALL 2009/2017 clinical trials [15]. Remarkably, in our DUX4 cohort, SER (28.4%, 19/67 with available data) to treatment had no significant impact on survival (5-year EFS 87.7 ± 8.2% vs. 93.1 ± 3.8%, p = 0.76; Fig. 1F). This is of particular interest, because 33.3% (8/24) of the DUX4 MRG patients treated according to the AIEOP-BFM ALL 2000 regimen, would nowadays be assigned to the HRG (n = 6 SER, n = 2 SER plus FCM-MRD ≥ 10% at day 15), however, none of these patients relapsed or died despite non-HR treatment (median follow-up 9.6 years, range 2.1–15.4 years; Supplementary Fig. S3A). This finding suggests that in the absence of other high-risk parameters, DUX4 SER patients may not require HR treatment.

Due to the occurrence of the swALL phenomenon (changes in lymphoid antigen expression and scatter levels representing gradual lympho-monocytoid transdifferentiation, accumulation of mature-appearing monocytes) [7], flow cytometry (FCM) based MRD monitoring may be challenging. Applying the current recommendation to include lymphoblasts and switch blasts for FCM-MRD assessment [6], on day 15 of treatment patients with swALL showed a significantly higher MRD burden than those with non-swALL (median 3.9%, range 0.1–65% vs. median 0.21%, range 0.0–21%; Fisher p-value < 0.001; Fig. 1G) and those with swALL also had higher lymphoblast counts than non-swALL cases (Fig. 1H).

All patients without detectable swALL by FCM on day 15 (38.5%, 25/65 with available data) remained in long-term remission, while those with swALL (61.5%, 40/65) had a worse outcome (5-year EFS 100% vs. 86.1 ± 5.8%, p = 0.04; Fig. 1I). Unexpectedly, DUX4 swALL did not show higher abundances of earlier developmental stages or a higher B-ALL multipotency score than non-swALL (Fig. 1J). We have no reasonable explanation for this finding, however, developmental and phenotypic plasticity assessed by transcriptome profiling and immunophenotyping, respectively, may reflect different characteristics of the blast cells.

Although not reaching statistical significance, DUX4 patients with <0.1% FCM-MRD lymphoblasts at day 15 (27.7%, 18/65 with available data) did not experience an event (Supplementary Fig. S4A) (83.3%, 15/18 non-HR treatment; Supplementary Table S8). A similar trend was also observed within the group of DUX4 patients with swALL (Supplementary Fig. S4B). Applying the currently used ≥10% cutoff for FCM-MRD assessment of high-risk disease, including only lymphoblasts better discriminated patients with a worse outcome than including both lymphoblasts and switch blasts (Supplementary Fig. S4C, D) or adding the monocyte-like cells (Supplementary Fig. S4E).

Notably, the best discrimination of outcome was achieved using FCM-based MRD measurement at day 15, including both lymphoblasts and switch blasts and a cutoff of 1%, as used in US-based protocols [8] (5-year EFS 100% vs. 83.4 ± 6.8%, p = 0.015; Supplementary Fig. S5A). Accordingly, though also not a risk stratification criterion in AIEOP-BFM ALL studies, PCR-MRD < 10−2 vs. ≥10−2 measured on day 15 also clearly distinguished patients with long-term EFS from those who experienced an event (Supplementary Fig. S5B).

Collectively, we show that patients with DUX4 B-ALL generally have a favorable outcome when treated according to AIEOP-BFM ALL protocols. However, recurrent secondary genetic alterations, such as the IKZF1plus deletion profile and TP53 mutation or a switch to the monocytic lineage during treatment, are indicative of a worse prognosis. The relevance of MRD response kinetics, particularly slow early response to therapy, the significance thresholds, and cell populations considered for FCM-MRD assessment may differ between DUX4 B-ALL and other subtypes. Therefore, we advocate a roadmap to revisit the prognostic factors in DUX4 patients across clinical trials to ultimately refine risk stratification and optimize therapy for this subgroup of leukemia.