Prognostic impact of the PredicT-ALL classifier in the AALL0434 trial: a model combining NGS, MRD, and WBC at diagnosis Open Access

TO THE EDITOR:

T-cell acute lymphoblastic leukemia (T-ALL) is a rare and aggressive neoplasm arising from the differentiation blockage and proliferation of T-cell precursors, which accounts for ∼15% of pediatric and 25% of adult ALL cases.^1,2 

The identification of robust biomarkers to guide clinicians in risk stratification remains an urgent need, both for early identification of patients who are viewed as high risk and prone to relapse, and for reducing the chemotherapy burden in patients viewed as very low risk. This need is especially critical given the increasing intensity of backbone chemotherapy regimens for T-ALL, where treatment-related toxicity has emerged as a significant cause of adverse events in some clinical trials.³ 

We previously reported a better outcome in adult and pediatric patients with T-ALL harboring NOTCH1 and/or FBXW7 mutations without alterations of K-N-RAS and PTEN genes, establishing the first classifier in T-ALL.^4,5 More recently, we report the impact of a next-generation sequencing (NGS)-based stratification classifier with improved accuracy through the incorporation of a larger panel of alterations in genes associated with favorable outcomes (NOTCH1/FBXW7, EP300, and PHF6), or poor outcomes (phosphatidylinositol 3-kinase [PI3K] pathway alteration including PTEN, PIK3CA, PI3K3R1, and IKZF1, TP53, DNMT3A, and N-KRAS alterations).⁶ Furthermore, replicating our strategy to integrate the white blood cell (WBC) count at diagnosis and minimal residual disease (MRD) levels at the end of induction (MRD at EOI),⁵ we generated a new classifier, now called PredicT-ALL, enabling refined risk stratification in T-ALL.⁵ 

In their landmark study, Pölönen et al⁷ recently analyzed an unprecedented cohort of nearly 1300 pediatric patients with T-ALL from the AALL0434 trial (NCT00408005),⁸ comprehensively characterizing their molecular landscape. Through multivariable outcome models, they demonstrated how genetic subtypes and alterations predict treatment failure and survival, providing valuable insights into disease classification and risk stratification.

Building on this exceptionally well-characterized molecular resource, we first validated our NGS-based classifier within this cohort, then confirmed the prognostic impact of the combined PredicT-ALL approach. Our analysis establishes a precise risk stratification system based on targeted NGS analysis, treatment response markers, and WBC count at diagnosis that could be readily incorporated into future therapeutic trials.

To ensure a harmonized analysis and accurately represent the pediatric T-ALL population, we focused on patients aged ≤18 years from 2 cohorts (FRALLE2000T and AALL0434 trials). The mutational landscape of patients from the AALL0434 cohort was defined based on the study by Pölönen et al, using comprehensive molecular analysis to identify gene alterations included in our NGS-based classifier.⁷ The FRALLE2000T data were from our previous study.⁶ The clinical and biological characteristics of both cohorts are presented in supplemental Table 1.

We performed a comparative analysis between 1249 pediatric patients (aged ≤18 years) enrolled in the AALL0434 trial (2007-2014), and 257 patients from the French FRALLE2000T cohort (2000-2010). The 2 cohorts were clinically comparable (Table 1). A male predominance was observed in both groups (74% in AALL0434 vs 77% in FRALLE2000T, P = .31). The median age was similar (8 vs 9 years, P = .13). The WBC count at diagnosis showed no significant difference (median: 76 10⁹/L [Q1, Q3] [21, 233] vs 98 10⁹/L [26, 220], P = .42), and the rate of central nervous system (CNS) involvement (CNS3; ≥5 WBCs per mL, with blasts or clinical signs of CNS involvement in both trials) was comparable (8% vs 11%, P = .13). MRD positivity at EOI (evaluated by flow cytometry in AALL0434, and based on detection of rearranged immunoglobulin and T-cell receptor genes in real-time quantitative polymerase chain reaction in FRALLE2000T), using a 10^–4 cut-off, was also similar (40% vs 39%, P = .76). Regarding survival outcomes, AALL0434 was associated with significantly better outcomes than FRALLE2000T. The 5-year overall survival (OS) was 90% (95% confidence interval [CI], 88-92) vs 76% (95% CI, 70-81), P < .001, and the 5-year disease-free survival (DFS) was 87% (95% CI, 85-89) vs 71% (95% CI, 65-77), P < .001, respectively. Differences in chemotherapy regimens may explain this disparity—for example, the use of continuous chemotherapy including nelarabine and the Capizzi methotrexate regimens in AALL0434, vs block-based consolidation chemotherapy for high-risk T-ALL in the FRALLE2000T protocol.

Using the NGS-based classifier previously validated in both pediatric (FRALLE2000T) and adult (GRAALL-2003/05) T-ALL French trials,⁶ we stratified patients from the AALL0434 trial based on key prognostic genetic alterations.

NOTCH1 intragenic deletions and intronic variants identified by Pölönen et al were excluded, as they were not captured by our NGS approach. The incidence of genetic alterations was similar in the 2 cohorts (Table 1). NOTCH1 and/or FBXW7 alterations occurred at identical frequencies (71% in both cohorts). EP300 and PHF6 alterations also showed similar distributions between AALL0434 and FRALLE2000T (2% vs 3%, P = .12, and 26% vs 22%, P = .27, respectively). Regarding high-risk alterations, while no IDH1/2 alterations were reported in the AALL0434 trial compared with a very low frequency in FRALLE2000T (2%), other alterations showed similar distribution between the trials: PI3K alterations (combining PTEN alterations, PIK3CA and PIK3R1 mutations) were found in 24% of patients with T-ALL from the AALL0434 trial vs 19% of patients from the FRALLE2000T trial (P = .08); DNMT3A mutations were found in 1% of patients in both trials; TP53 alterations (combining 17p deletion and/or TP53 mutation) occurred at similar frequencies (3% vs 4%, P = .43); IKZF1 alterations (deletion and/or mutation) were detected in 4% of patients from the AALL0434 trial and 5% in patients from the FRALLE2000T trial (P = .28); and N-KRAS mutations were also comparable (14% vs 11%, P = .36). Overall, this similar distribution of genetic alterations strengthens the validity of comparisons between the 2 populations.

Low-risk NGS (LR-NGS) patients harbored NOTCH1/FBXW7 and/or PHF6 and/or EP300 mutations without any high-risk alterations (supplemental Figure 1A). This classification identified 52% (658/1249) of patients from the AALL0434 trial as high-risk NGS (HR-NGS), a proportion comparable to that found in the FRALLE2000T trial (48%; 124/257, P = .2; Table 1). In line with our results in the FRALLE2000T trial, HR-NGS patients from the AALL0434 trial showed significantly inferior outcomes, with lower 5-year DFS; 83% (95% CI, 79-86) vs 92% (95% CI, 89-95), P < .001, and 5-year OS; 87% (95% CI, 84-90) vs 93% (95% CI, 91-95), P < .001 (Figure 1A-B; supplemental Table 2).

We have shown previously^5,6 that integrating the NGS-based classification with initial WBC count and MRD at EOI resulted in enhanced risk stratification accuracy. This integrated approach to predict relapse in T-ALL (PredicT-ALL) identified a very low-risk group (PredicT-ALL_Low), comprising LR-NGS patients with WBC <200 10⁹/L and EOI MRD <10^–4, representing 24% (292/1242) of patients from the AALL0434 trial, a proportion comparable to the 28% (60/215) found in the FRALLE2000T trial; P = .2. Conversely, the high-risk group (PredicT-ALL_High), consisting of HR-NGS patients with either WBC ≥200 10⁹/L and/or EOI MRD ≥10^–4, comprised 33% (411/1242) of patients from the AALL0434 trial, a proportion comparable to the 28% (60/215) found in the FRALLE2000T trial, P = .2. The remaining patients were classified as intermediate risk (PredicT-ALL_Int; supplemental Figure 1B).

The AALL0434 PredicT-ALL_Low group showed excellent outcomes, with 5-year DFS of 96% (95% CI, 93-99) and OS of 98% (95% CI, 96-99). The PredicT-ALL_Int group also maintained relatively good outcomes, with 5-year DFS of 87% (95% CI, 84-90) and OS of 90% (95% CI, 88-93). In contrast, the PredicT-ALL_High group showed significantly poorer outcomes, with 5-year DFS of 80% (95% CI, 76-85) and OS of 84% (95% CI, 80-88), P < .001 (Figure 1C-D; supplemental Table 2).

Overall, we identify a subgroup with exceptional outcomes who could benefit from treatment de-escalation without compromising survival. Conversely, our analysis revealed a high-risk group with poor outcomes under current treatment protocols, representing prime candidates for novel therapeutic approaches including targeted drugs and immunotherapy.

These findings strongly support the integration of the PredicT-ALL classifier, using an NGS strategy compatible with clinical practice, into future pediatric T-ALL trials.

Contribution: M.S., V.A., and A.B. conceived the study; M.S. performed the statistical analysis; and all authors provided data and wrote the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Vahid Asnafi, Laboratory of Onco-Hematology, Necker Enfants Malades Hospital, Assistance Publique - Hôpitaux de Paris, Université Paris Cité, 149 Rue de Sèvres149 Rue de Sèvres, 75015 Paris, France; email: vahid.asnafi@aphp.fr; and André Baruchel, Department of Pediatric Hematology and Immunology, Robert Debré Hospital, Assitance Publique - Hôpitaux de Paris, Université Paris Cité, Paris, France; email: andre.baruchel@aphp.fr.