TL;DR
This paper develops a mathematical model to understand how two key mutations (JAK2<sup>V617F</sup> and CALR<sup>m</sup>) initiate and drive the development of myeloproliferative neoplasms (MPNs). By combining mathematical modeling with statistical inference on patient data, they reveal differences in the timing of mutation acquisition and the proliferative advantage of clones carrying these mutations, suggesting distinct disease progression pathways. This work also proposes optimized early screening strategies tailored to each mutation type.
What Problem Does This Paper Address?
Myeloproliferative neoplasms (MPNs) are blood cancers characterized by the abnormal proliferation of myeloid cells. These diseases arise from specific genetic mutations in hematopoietic stem cells (HSCs), the body's master blood cell producers. The two most common "driver" mutations are JAK2<sup>V617F</sup> and CALR<sup>m</sup>. However, directly observing the initiation and early development of these mutations in human HSCs is extremely challenging due to their rarity and inaccessibility. This makes it difficult to understand how these diseases start, progress, and to develop effective early detection methods. The paper aims to bridge this knowledge gap by using mathematical modeling and statistical inference to infer the dynamics of MPN development from mutation acquisition to disease onset.
Methods
The core of this work is a mathematical model that simulates the clonal expansion of mutated HSCs. Key aspects include:
- Stochastic Modeling: The initial stages of mutation acquisition and early clonal expansion are modeled using a Continuous Time Markov Chain (CTMC) process, reflecting the inherent randomness at this stage.
- Deterministic Approximation: As the number of mutated HSCs grows, the model transitions to a deterministic approximation, which is computationally more tractable.
- Bayesian Inference (ABC-SMC): To estimate model parameters (like mutation acquisition time and proliferative advantage) from patient data, the researchers employ Approximate Bayesian Computation via Sequential Monte Carlo (ABC-SMC). This is a likelihood-free method suitable for complex models where direct likelihood calculation is difficult.
- Optimal Assignment Algorithm: To speed up the ABC-SMC process, they utilize the Hungarian algorithm for optimal assignment, efficiently comparing model simulations to patient observations.
- Hypothesis Testing: The model explicitly considers two hypotheses for mutation acquisition: occurring during fetal life (T<sub>0</sub>=0) or after birth (T<sub>0</sub>>0), allowing for statistical comparison.
Datasets
The study utilizes data from previously published studies:
- CALR<sup>m</sup> MPN patients: Data from 15 patients (Mosca et al., 2021 and El-Khoury et al., 2019).
- JAK2<sup>V617F</sup> MPN patients: Data from 11 patients (Mosca et al., 2021 and El-Khoury et al., 2019).
The data consists of Clonal Fractions (CF) of CD34+ progenitor cells at specific time points for each patient. The researchers focus on patients with heterozygous clones.
Key Insights and Ideas
- Distinct Timelines: CALR<sup>m</sup> mutations appear to be acquired significantly later in life (average ~25 years) compared to JAK2<sup>V617F</sup> mutations (average ~15 years), with a non-negligible probability of JAK2<sup>V617F</sup> occurring even during fetal life.
- CALR<sup>m</sup>'s Proliferative Edge: Despite later acquisition, CALR<sup>m</sup>-mutated clones exhibit a higher proliferative advantage at the stem cell level than JAK2<sup>V617F</sup> clones. This means they expand more aggressively once initiated.
- Early Screening Potential: The model suggests that early screening for JAK2<sup>V617F</sup> mutations could be a viable clinical strategy, with an optimal screening age around 30 years, offering a high probability of early detection.
- CALR<sup>m</sup> Screening Challenges: Early screening for CALR<sup>m</sup> is less promising. Even at the optimal screening age (~35 years), a substantial portion of patients might be missed, and the detection window is narrower.
- Sensitivity Matters: The effectiveness of early screening is highly dependent on the sensitivity of the detection techniques (i.e., the minimum Variant Allele Frequency (VAF) that can be reliably detected). Higher sensitivity generally improves the probability of early detection.
Key Results
- Mutation Acquisition Time: Estimated mean acquisition time for JAK2<sup>V617F</sup> is ~15 years, with a 24% probability of fetal acquisition. For CALR<sup>m</sup>, the mean acquisition time is ~25 years, with a near-zero probability of fetal acquisition.
- Proliferative Advantage (Δ): The estimated proliferative advantage parameter (Δ) is higher for CALR<sup>m</sup> (0.026) than for JAK2<sup>V617F</sup> (0.017).
- Optimal Screening Age: The model identifies an optimal screening age of ~30 years for JAK2<sup>V617F</sup> and ~35 years for CALR<sup>m</sup>.
- Early Detection Rates: At their respective optimal ages, early detection is estimated to be ~79% for JAK2<sup>V617F</sup> but only ~42% for CALR<sup>m</sup>.
- Impact of Sensitivity: Increasing detection sensitivity (lower VAF threshold) improves early detection probabilities for both mutations, but the gap between JAK2<sup>V617F</sup> and CALR<sup>m</sup> remains. For example, even with high sensitivity (0.01% VAF), CALR<sup>m</sup> early detection probability is below 65%.
Limitations and Future Work
- Model Simplifications: The model assumes parameters are constant over time and do not depend on individual patient heterogeneity beyond the main driver mutation. It also assumes a constant pool of wild-type HSCs.
- Homozygous Subclones: The current model does not account for homozygous subclones, which might have an even greater proliferative advantage.
- Additional Mutations: The model focuses on single driver mutations. Future work could extend the model to include dynamics of subclones with additional mutations (e.g., TET2, DNMT3A).
- Threshold Effects: More complex models could incorporate threshold effects in clonal expansion, potentially reflecting regulatory mechanisms.
- WT HSC Dynamics: The number and behavior of wild-type HSCs might change with age, which could be incorporated into future models.
- CALR<sup>m</sup> Subtypes: The study did not differentiate between CALR<sup>m</sup> Type 1 and Type 2 mutations, which are clinically distinct. Future work could explore these differences.
- Large-Scale Screening: The authors envision extending their framework to simultaneously test for multiple mutations using next-generation sequencing (NGS), which would present a more complex optimization problem but could yield pragmatic prevention strategies.
Summary updated