Highly sensitive quantitative real-time PCR (qPCR) measurements of leukemia-specific alterations, such as reciprocal translocations (e.g. RUNX1-RUNX1T1) or mutations (e.g. NPM1), enables monitoring of treatment response and measurable residual disease (MRD) levels in acute myeloid leukemia (AML) . Because of its high prevalence (>30% of all AML patients), mutated NPM1 (NPM1mut) is of particular importance. The NPM1mut cell burden was found to correlate with relapse risk and survival in AML [2, 3]. However, such approaches are widely limited to static analyses at pre-defined time points, not explicitly analyzing the dynamics of treatment response, remission, and relapse. Based on time course data from 340 NPM1 mutant-AML patients we investigated, whether response dynamics yield additional prognostic information, specifically whether slope and depth of initial MRD decrease correlates with long-term outcome. Herein, our focus was the identification of prognostic factors to improve relapse risk assessment of individual patients.
We selected all patients with available time courses of at least three subsequent NPM1 measurements (n = 340) from the AML2003 (NCT00180102) and AML60+ (NCT00180167) trials and the AML registry. The total number of samples was 1936. Bone marrow aspirates were taken at different time points during intensive treatment and follow-up. NPM1 transcripts with mutations of type A, B, and D were quantified using a previously described qPCR method  and are provided as NPM1 transcripts per 100 ABL transcripts. The complete anonymized dataset is made available as Supplementary Data. To facilitate a quantitative comparison, we defined three characteristic phases for each time course (Fig. 1A): (1) the elimination phase, during which induction treatment progressively reduces the NPM1 burden; (2) the remission phase, where the NPM1 burden remains below 1% (following recommendation in ); and (3) the relapse phase, in which the NPM1 burden increases after first remission, leading to molecular relapse (NPM1 > 1%). Representative time courses for each phase can be found in Supplementary Figure S1. Classification of data points for each time course was performed automatically by a corresponding algorithm (see Supplementary Materials & Methods).
First, we analyzed the pairwise correlations (Supplementary Table S1) of typical treatment response dynamic characteristics, namely the elimination slope (α), the minimal NPM1 level after primary treatment (induction + consolidation) within 6 months after treatment start (n), the maximum slope during relapse phase (β), the time until molecular relapse (d), and the overall survival time (s). The frequency distributions of these five characteristics within the patient population are visualized in Supplementary Figure S2. We demonstrate that time until relapse (d) and survival time (s) are positively correlated with correlation coefficient ρ = 0.73 (95% confidence interval: [0.51; 0.86]) and that these two parameters themselves anti-correlate with the relapse slope (β) with correlation coefficients ρ = −0.64 [−0.76; −0.50] and ρ = −0.56 [−0.76; −0.27], respectively (Supplementary Figure S3A and B). This indicates that a rapidly evolving relapse usually occurs earlier after therapy start than a slower one, and that this leads to shorter survival. Furthermore, we identified a moderate negative correlation ρ = −0.48 [−0.65; −0.26] between the NPM1 level after primary treatment (n) and the time until relapse (d), pointing toward the already known prognostic value of NPM1 levels for relapse incidence [3,4,5,6] (Supplementary Figure S3C).
Furthermore, we estimated cumulative incidence functions to investigate the prognostic potential of the dynamic response parameters with respect to the incidence of death, to adequately consider the competing risk between stem cell transplantations (SCT) and the target event death due to AML. This competing risk analysis is required as the allogeneic SCT itself cannot be considered to be a non-informative censored event, as patients with a poor prognosis are more likely to get transplanted . To do so, we stratified the patients according to their elimination slope (α), i.e., fast (α < −0.034) and slow (α > −0.034) therapy response. The threshold was chosen to maximize the difference between the groups (see Supplementary Materials & Methods). We observed a tendency for a decreased cumulative incidence of death in case of a rapid initial response (Fig. 1B).
Our analysis also suggests a link between the residual NPM1 level after induction treatment (n) and survival time (s). Specifically, our data provide evidence that a none-detectable residual NPM1 level (i.e., deep responders) might be associated with a lower incidence of death (Fig. 1C). To evaluate the prognostic value of the residual NPM1 level, we analyzed the minimal NPM1 level within the first six months after treatment start. However, at this point the NPM1 level is not always fully bottomed out. That is, in some patients the NPM1 level further decreased even beyond month 6. To investigate at which time point the NPM1 level is most informative, we run our analysis for different time points. That means, we evaluated the minimal NPM1 level, which had been achieved within 3, 6, 9, and 12 months after therapy start and calculated to what extent a separation with respect to 5-year cumulative incidence of death can be achieved. To avoid bias, we only included patients who did not experience relapse, died, or were treated with SCT until the time point analyzed. We could show that the threshold of 0.01% NPM1/ABL yielded a difference in cumulative incidence of death between deep and poor responders only for 9 and 12 months. However, the prediction quality does not substantially increase from 9 to 12 months (Fig. 1D). Using these findings and conducting a logistic regression (Fig. 1E), an odds ratio of 1.7 (P < 0.001) for the 5-year survival was estimated, indicating that a one log10-scale lower minimal NPM1 value within 9 months post treatment start nearly doubles the chance of surviving 5 years. Repeating these analyses including only patients who reached hematologic complete remission provides the same qualitative results. Due to the reduced sample size, however, these results are not statistically significant (Supplementary Figure S4).
Using the lowest NPM1 level within 9 months after treatment start as a landmark (i.e., deep vs. poor responders), we found a significant difference in the median elimination slope (α) between these two groups (Fig. 2A), confirming the weak positive correlation found earlier (Table S1). As visible in Fig. 2E, the reduction of the leukemic burden is not as efficient in the poor responder group compared to the deep responder group, whereas the average initial NPM1 value does not differ. This implies that a good initial response is essential to reach lower NPM1 levels. The significantly longer median time to relapse for deep responders (Fig. 2B) confirms the higher efficiency of primary therapy in these patients. However, the proportion of relapsing patients who reach a molecular remission (63.2% of poor responders, all deep responders) does not significantly differ between both groups (poor: 31.9% [24.2; 39.6]; deep: 35.0% [25.8; 44.2]). The same is true for patients with fast and slow response: of the 46.7% of patients that reached remission in the slow initial response group, 35.7% [21.2; 50.2] relapsed, while a similar proportion of patients (30.7% [23.9; 37.5]) with fast therapy response that reached remission (88.6%) also relapsed within the observed period. This suggests that a good initial NPM1mut clearing cannot prevent, but postpones relapse. However, this postponement positively effects overall survival (Fig. 1B), potentially because the patients can be salvaged more efficiently. On the other hand, when comparing the relapsed patients with the patients who stayed in remission at least for 2 years, no differences with respect to the analyzed parameters could be found. These findings suggest that relapse per se is not directly linked to initial treatment response.
The evolution of the leukemic clones might serve as a potential explanation for differences in relapse incidence. Co-occurrence of mutations beyond NPM1, such as the FLT3 mutation , has already been identified as an important risk factor in AML. Therefore, we investigated the influence of the FLT3 allelic ratio at diagnosis on the NPM1 level after primary treatment (n). Patients with a high allelic ratio (≥0.5 ) do not reach an NPM1 value as low as the patients without an FLT3 mutation (P < 0.001, global test across all three FLT3 levels: P = 0.001, Fig. 2C). The reason for this might be a reduced average number of treatment cycles for patients with high FLT3 ratio (3.2 ± 1.4 vs. 3.8 ± 1.6 for patients without FLT3), as there is no significant difference (P = 0.23) in the median therapy response (measured by elimination slope (α)) between the groups. The increased median FLT3 ratio for patients who died within 2 years after diagnosis compared with patients who stayed in remission for 2 years (P = 0.002), supports the worse prognosis for patients with high FLT3 allelic ratio (Fig. 2D).
In summary, analyzing the dynamic treatment response of a large patient cohort with NPM1mut-AML, we provide evidence that an efficient clearance of the leukemic burden provides prognostic information. We show that the elimination slope during primary therapy can be used as a predictor of treatment success and hence, provides additional information about the course of disease. Also, we demonstrate the most informative time point for the determination of the minimal NPM1 measurement as a predictor for survival and relapse risk is 9 months after therapy start. This may be of relevance for further monitoring of patients and decisions on potential therapeutic interventions including allogeneic hematopoietic cell transplantation. Our study emphasizes the significance of close NPM1 monitoring in the given mutational background for AML risk assessment and relapse prediction. It provides new information, which will impact the design of further studies and to guide novel MRD-guided therapeutic strategies .
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work has been supported by the German Federal Ministry of Education and Research (www.bmbf.de/en/), Grant number 031A424 “HaematoOpt”.