A causal Bayesian network model of disease progression mechanisms in chronic myeloid leukemia
Graphical abstract
Introduction
Chronic myeloid leukemia (CML) is a type of cancer which progressively disturbs the balance of the hematopoietic system by dysregulated growth of myeloic cells in the bone marrow. In the vast majority of cases, the disease is caused by a mutation resulting from a reciprocal translocation between chromosomes 9 and 22 (Hehlmann et al., 2007, Quintás-Cardama et al., 2010). The translocation gives rise to the so called BCR-ABL fusion gene, triggering the characteristic pathophysiological processes of CML. BCR-ABL encodes for an overactive Bcr-Abl tyrosine kinase and disturbs many cellular processes such as differentiation, proliferation and apoptosis (Ren, 2005, Quintás-Cardama and Cortes, 2009). BCR-ABL positive cells therefore gain a growth advantage leading to an expanding population of cancer cells which impedes physiological hematopoiesis in the bone marrow.
CML can be divided into three phases: the chronic phase (CP), accelerated phase (AP) and blast crisis (BC), the final stage which resembles an acute leukemia. In the chronic phase, the disease is usually asymptomatic or symptoms are unspecific. Hematologically, CP-CML is characterized by an increased proliferation of myeloic precursor cells such as blasts, increased blood levels of mature granulocytes or platelets and often splenomegaly (Apperley, 2015). After 7–10 years without treatment the disease progresses to AP-CML, characterized by increased cell numbers in the peripheral blood, accumulation of additional mutations, lower treatment response and a more severe symptomatic burden. Although an established clinical category, the concept of AP as a biological entity sui generis has been challenged by the finding of its high genetic similarity to BC, to which the disease ultimately progresses (Radich et al., 2006). Depending on the definition, the final phase is characterized by an increase in blasts to >20–30%, although blast numbers up to 90% and more can be found in some patients (Palandri et al., 2008). While the abundance of blasts in the peripheral blood frequently causes thrombotic events, the excessive growth of cancer cells in the bone marrow successively displaces physiological hematopoiesis leading to anemia, immunodeficiency and susceptibility for infections (Chereda and Melo, 2015). The joint of effect of these processes dramatically limits the survival prospect of patients in BC (Kantarjian et al., 2001, Kantarjian et al., 2012).
First-line therapy with tyrosine kinase inhibitors (TKIs) such as imatinib re-establishes a normal level of proliferation and apoptosis by effectively inhibiting the activity of the Bcr-Abl tyrosine kinase. Imatinib treatment during CP-CML diminishes disease activity usually below the limit of detection, prevents disease progression and thereby has led to a remarkably increased overall survival (Druker et al., 2006). Yet, due to residual cancer cells in dormancy, lifelong TKI therapy is mandatory for most patients, otherwise disease relapse is likely (Apperley, 2015). Once BC has occurred, the response to TKIs is dramatically worse (Jabbour et al., 2014).
Although recent decades have shed light on many details of the disease's pathophysiology, the mechanisms of disease progression to BC are still poorly understood. Several cellular and molecular changes are believed to contribute to this process, e.g. increase of Bcr-Abl expression, altered adhesion behavior, additional mutations, differential gene expression or altered mRNA metabolism (Shet et al., 2002, Silver, 2009, Perrotti, 2010, Chereda and Melo, 2015). However, neither of these changes have been proven to be necessary or sufficient for disease progression. Moreover, the mentioned alterations are only an excerpt of frequently observed changes and do not exclude each other (Calabretta and Perrotti, 2004, Chereda and Melo, 2015).
The majority of the hitherto developed mathematical models of CML have focused on aspects like cell population dynamics, mechanisms for residual disease and drug resistance based on mathematical frameworks like differential equations or rule based stochastic processes (Michor et al., 2005, Roeder et al., 2006, Foo et al., 2009). Yet, to our knowledge, few mathematical models so far have dealt with disease progression (e.g. Michor, 2007 and Sachs et al., 2011). Further theoretical investigations can therefore help to better evaluate the proposed mechanisms of disease progression (Michor, 2008). In this study we present a mathematical model of CML based on a causal Bayesian network in order to study such mechanisms. The paper is organized as follows: Section 2 gives a short introduction to the employed formalism and describes the model and the validation procedure. Section 3 presents the simulation results. Starting with a very simple model, different mechanisms will be added to the model in order to investigate their possible influence on disease progression. Disease progression is assessed by taking into account hematological and clinical variables that are essential for the phenotype of CML as well as the response to therapy. In Section 4 the model and its results are discussed with respect to biological plausibility, other mathematical models, limitations of the current approach and future research directions.
Section snippets
Causal Bayesian networks
The causal interpretation of Bayesian networks (BNs) (Pearl, 1998, Neapolitan, 1990) was mainly developed by Spirtes, Glymour, and Scheines (Spirtes et al., 1993). A causal Bayesian network (CBN) consists of a directed acyclic graph G = (V,E) (DAG) and a probability distribution P over a set V = {X1,…,Xn} of random variables (RVs) of interest. Directed paths in such a graph represent causal relations between the RVs. Xi→…→Xj in a graph G, for example, stands for the hypothesis that Xi is a
Core model: increased Bcr-Abl levels do not suffice to explain BC-CML
As a first test of the core version of our model, we wanted to know whether it can reproduce the basic characteristics of CP-CML at the point of diagnosis, i.e. without imatinib therapy. As Fig. 3(A) shows, the core version of our model can successfully reproduce the phenotype of CP-CML with an acceptable or good fit to the validation data for all hematological output variables. Next, we investigated whether it can also capture the phenotype of untreated BC-CML. As Fig. 3(B) shows, the core
Discussion and conclusion
We have presented a causal Bayesian network model for CML and explored possible mechanisms of disease progression from chronic phase to blast crisis based on the model's predictions. The results of our investigations suggest in summary that:
- •
Increased expression of Bcr-Abl alone is not sufficient to explain the phenotype of treated and untreated BC-CML.
- •
Increased migration of cancer cells from the bone marrow to the peripheral blood could play a contributory role (e.g. by mediating part of the
Competing interests
We have no competing interests to declare.
Acknowledgment
We thank Stefan Göbbels, Matthias Müller and an anonymous referee for helpful comments on earlier drafts of this paper.
References (57)
Chronic myeloid leukaemia
Lancet
(2015)- et al.
Inhibition of BCR-ABL expression with antisense oligonucleotides restores ß1 integrin-mediated adhesion and proliferation inhibition in chronic myelogenous leukemia hematopoetic progenitors
Blood
(1998) - et al.
The biology of CML blast crisis
Blood
(2004) - et al.
Clonal evolution in chronic myelogenous leukemia
Hematol Oncol Clin. North Am.
(2004) Improved survival in chronic myeloid leukemia since the introduction of imatinib therapy: a single-institution historical experience
Blood
(2012)Role of BCR/ABL gene-expression levels in determining the phenotype and imatinib sensitivity of transformed human hematopoetic cells
Blood
(2007)Lack of adhesion molecules P-selectin and intercellular adhesion molecule-1 accelerate the development of BCR/ABL-induced chronic myeloid leukemia-like myeloproliferative disease in mice
Blood
(2004)- et al.
Molecular biology of bcr-abl1–positive chronic myeloid leukemia
Blood
(2009) Molecular cloning of human paxillin, a focal adhesion protein phosphorylated by P210BCR/ABL
J. Biol. Chem.
(1995)Imatinib induces hematologic and cytogenetic responses in patients with chronic myelogenous leukemia in myeloid blast crisis: results of a phase II study
Blood
(2002)
The blast phase of chronic myeloid leukaemia
Best Prac. Res. Clin. Hematol.
Pathophysiology of CML: do defects in integrin function contribute to the premature circulation and massive expansion of the BCR/ABL positive clone?
J. Lab. Clin. Med.
Bcr-Abl expression levels determine the rate of development of resistance to imatinib mesylate in chronic myeloid leukemia
Cancer Res.
Maintaining Low BCR-ABL signaling output to restrict CML progression and enable persistence
Curr. Hematol Malig. Rep.
Model Selection and Multimodal Inference. A Practical Information-Theoretic Approach
Natural course and biology of CML
Ann. Hematol.
Modeling mechanisms with causal cycles
Synthese
Maximum likelihood from incomplete data via the EM algorithm
J. R. Stat. Soc. Ser. B.
Efficacy and safety of a specific inhibitor of the bcr-abl tyrosine kinase in chronic myeloid leukemia
N. Engl. J. Med.
Activity of a specific inhibitor of the bcr-abl tyrosine kinase in the blast crisis of chronic myeloid leukemia and acute lymphoblastic leukemia with the Philadelphia chromosome
N. Engl. J. Med.
Five-year follow-up of patients receiving imatinib for chronic myeloid leukemia
N. Engl. J. Med.
Eradication of chronic myeloid leukemia stem cells: a novel mathematical model predicts no therapeutic benefit of adding G-CSF to imatinib
PLoS Comput. Biol.
Increase of bcr-abl chimeric mRNA expression in tumor cells of patients with chronic myeloid leukemia precedes disease progression
Blood
NasoNet, modeling the spread of nasopharyngeal cancer with networks of probabilistic events in discrete time
Artif. Intell. Med.
A modeling approach for mechanisms featuring causal cycles
Philos Sci.
Tyrosine phosphorylation and activation of focal adhesion kinase (p125FAK) by BCR-ABL oncoprotein
Exp. Hematol.
Detection of BCR-ABL proteins in blood cells of benign phase chronic myelogenous leukemia patients
Cancer Res.
Chronic myeloid leukaemia
Lancet
Cited by (25)
Recovery preparedness of global air transport influenced by COVID-19 pandemic: Policy intervention analysis
2021, Transport PolicyCitation Excerpt :Essentially, it is assumed that CBN satisfies the causal Markov condition, which means a node is conditionally independent of its non-descendent nodes. CBN models can build and test causal hypotheses (Pearl, 2009) as well as predict the effects of possible interventions with uncertainties, even if only non-experimental data is available (Koch et al., 2017). In this paper, the post-recovery of global air transport is conditioned on public policy and flight willingness, which is solving by a Leaky Noisy-OR gate and will be described in detail later, in section 4.
Untangling the complexity of multimorbidity with machine learning
2020, Mechanisms of Ageing and DevelopmentCitation Excerpt :As step further from interpretability is to investigate causal relationships (or mechanisms); this is an important goal for ML models’ success in medicine, where clinical evidence has been informed by well-conducted studies, such as randomised clinical trials. Causal inference from observational data is often a controversial issue (Collins et al., 2020) but many developments in the field have shown to be successful in tackling some of its limitations, such as the causal Bayesian networks which have been used to model disease progression mechanisms (Koch et al., 2017) or predict disease complications (Yousefi et al., 2020). Causal inference tools are also used to analyse the relationships between the learned deep learning phenotypes and patient outcomes and diseases of interest (Kale et al., 2015).
A quantitative study on the role of TKI combined with Wnt/β-catenin signaling and IFN-α in the treatment of CML through deterministic and stochastic approaches
2020, Chaos, Solitons and FractalsCitation Excerpt :The BCR-ABL oncogene encodes for tyrosine kinase (a protein) that triggers the white blood cells to proliferate at an abnormal rate as compared to normal white blood cells [2,4,6]. The progression of CML can typically be observed in three phases, namely, the chronic phase, the accelerated phase and finally the blast crisis, which is akin to acute leukemia [7]. Traditionally, the standard therapy for CML includes interferon alfa-2a (IFN-α) whose curative effectiveness was highly limited [8,9].
Progression, detection and remission: evolution of chronic myeloid leukemia using a three-stage probabilistic model
2023, Journal of Applied Mathematics and ComputingApplication of continuous bayesian networks in the association study between uric acid and chronic metabolic diseases
2023, Chinese Journal of Disease Control and Prevention