Mathematical Models for Erythropoiesis

Transcription

1 Mathematical Models for Erythropoiesis Joseph M. Mahaffy Department of Mathematical and Statistical Sciences San Diego State University January 2003 p.1/37

2 Collaborators Jacques Bélair - Université de Montréal Mike Mackey - McGill University NSF-REU Students Sam Polk Roland Roeder p.2/37

3 Outline Discuss the Physiology Develop the Age-Structured Model Reduce the Model to Delay Equations Bifurcation Analysis Compare to Examples Rabbit with Induced Auto-Immune Hemolytic Anemia Human Subject following a Phlebotomy Summary p.3/37

4 Hematopoiesis p.4/37

5 Erythropoiesis p.5/37

6 Features of Erythropoiesis BFU-E and CFU-E differentiate and proliferate in response to EPO Maturation requires about 6 days EPO accelerates maturation Lack of EPO causes apoptosis Cell divisions every 8 hours for about 4 days Reticulocytes do not divide - increase hemoglobin Erythrocytes lose nucleus - live 120 days Macrophages degrade RBCs EPO released near kidneys with half-life of 6 hours p.6/37

7 Age-Structured Model p.7/37

8 Detailed Model p.8/37

9 Active Degradation of RBCs RBCs age - Cell membrane breaks down Membrane marked with antibodies Macrophages destroy least pliable cells Model assumes constant supply macrophages Saturated consumption of Erythrocytes - Satiated predator Constant flux of RBCs being destroyed p.9/37

10 Constant Flux Boundary Condition Let Q be rate of removal of erythrocytes Erythrocytes lost are Q t Mean Value Theorem - average number RBCs m(ξ, ν F (ξ)) for ξ (t, t + t) Balance law Q t = W t m(ξ, ν F (ξ)) [ν F (t + t) ν F (t)]m(ξ, ν F (ξ)) As t 0, Q = [W ν F (t)]m(t, ν F (t)) p.10/37

11 Constant Flux - Diagram p.11/37

12 Simplifying Assumptions Several simplifying assumptions allow reduction of the age-structured model to delay differential equations Assume that V (E) = W = 1. Assume the birth rate β satisfies: β(µ, E) = { β, µ < µ1, 0, µ µ 1, Assume that γ is constant. p.12/37

13 Reduced PDEs The model satisfies the partial differential equations: p t + p µ = β(µ)p m t + m ν = γm p.13/37

14 Reduced PDEs The model satisfies the partial differential equations: p t + p µ = β(µ)p m t + m ν = γm with the boundary conditions: p(t, 0) = S 0 (E) and p(t, µ F ) = m(t, 0) (1 ν F (t))m(t, ν F (t)) = Q p.13/37

15 Negative Control The negative feedback by EPO satisfies the equation: E = a 1 + KM r ke where the total mature erythrocyte population is M(t) = νf (t) 0 m(t, ν)dν p.14/37

16 Method of Characteristics Applied to a Simplified Model The method of characteristics can be used to simplify the partial differential equations given above. Let P (s) = p(t(s), µ(s)), dp ds = p t dt ds + p µ dµ ds = β(µ(s), E(t(s)))P (s), which has the solution, P (s) = p(t, µ) = P (0) exp provided [ s 0 ] β(µ(r), E(t(r)))dr, dt ds = 1 and dµ ds = V (E(t(s))) = 1 p.15/37

17 Characteristics Diagram p.16/37

18 Evaluating p(t, µ) and m(t, ν) To find p at µ F, p(t, µ F ) = p(t 0, 0)exp [ µf 0 ] β(r)dr = p(t µ F, 0)e βµ 1 = e βµ 1 S 0 (E(t µ F )) Similar use of the characteristics gives m(t, ν) = m(t ν, 0)e γν, p.17/37

19 Finding M(t) M(t) = = = νf (t) 0 νf (t) 0 νf (t) 0 m(t ν, 0)e γν dν p(t µ F ν, 0)e γν dν e βµ 1 S 0 (E(t µ F ν))e γν dν, = e γ(t µ F ) e βµ 1 t µf t µ F ν F (t) S 0 (E(w))e γw dw, p.18/37

20 Leibnitz s Rule We apply Leibnitz s rule for differentiating an integral: M(t) = γe γ(t µ F ) e βµ 1 t µf t µ F ν F (t) S 0 (E(w))e γw dw +e βµ 1 [S 0 (E(t µ F )) ] S 0 (E(t µ F ν F (t)))e γν F (t) (1 ν F (t)) = γm(t) + e βµ 1 S 0 (E(t µ F )) Q, p.19/37

21 Model with Delays After reduction of PDEs, the state variables become total mature erythrocytes, M, EPO, E, and age of RBCs, ν F. dm(t) dt de(t) dt dν F (t) dt = e βµ 1 S 0 (E(t T 1 )) γm(t) Q = f(m(t)) ke(t) = 1 Qe βµ 1 eγν F (t) S 0 (E(t T 1 ν F (t)) where T 1 = µ F. This is a state-dependent delay differential equation. p.20/37

22 Properties of the Model State-dependent delay model has a unique positive equilibrium p.21/37

23 Properties of the Model State-dependent delay model has a unique positive equilibrium Delay T 1 accounts for maturing time p.21/37

24 Properties of the Model State-dependent delay model has a unique positive equilibrium Delay T 1 accounts for maturing time State-dependent delay in equation for ν F, the varying age of maturation p.21/37

25 Properties of the Model State-dependent delay model has a unique positive equilibrium Delay T 1 accounts for maturing time State-dependent delay in equation for ν F, the varying age of maturation The ν F differential equation is uncoupled from the differential equations for M and E p.21/37

26 Properties of the Model State-dependent delay model has a unique positive equilibrium Delay T 1 accounts for maturing time State-dependent delay in equation for ν F, the varying age of maturation The ν F differential equation is uncoupled from the differential equations for M and E Stability determined by equations for M and E p.21/37

27 Linear Analysis of the Model Linearizing about the unique equilibrium ( M, Ē, ν F ), M(t) = e βµ 1 S 0(Ē)E(t T 1) γm(t) E(t) = f ( M)M(t) ke(t) ν F (t) = 1 Ē E(t T 1 ν F ) γν F (t) p.22/37

28 Linear Analysis of the Model Linearizing about the unique equilibrium ( M, Ē, ν F ), M(t) = e βµ 1 S 0(Ē)E(t T 1) γm(t) E(t) = f ( M)M(t) ke(t) ν F (t) = 1 Ē E(t T 1 ν F ) γν F (t) The characteristic equation is given by (λ + γ) [ (λ + γ)(λ + k) + Āe λt 1] = 0, where A e βµ 1 S 0(Ē)f ( M) > 0. p.22/37

29 Linear Analysis of the Model Linearizing about the unique equilibrium ( M, Ē, ν F ), M(t) = e βµ 1 S 0(Ē)E(t T 1) γm(t) E(t) = f ( M)M(t) ke(t) ν F (t) = 1 Ē E(t T 1 ν F ) γν F (t) The characteristic equation is given by (λ + γ) [ (λ + γ)(λ + k) + 1] Āe λt = 0, where A e βµ 1 S 0(Ē)f ( M) > 0. One solution is λ = γ, which shows the stability of the ν F equation. p.22/37

30 Analysis of Characteristic Equation Remains to analyze (λ + γ)(λ + k) = Ae λt 1 p.23/37

31 Analysis of Characteristic Equation Remains to analyze (λ + γ)(λ + k) = Ae λt 1 A Hopf bifurcation occurs when λ = iω solves the characteristic equation p.23/37

32 Analysis of Characteristic Equation Remains to analyze (λ + γ)(λ + k) = Ae λt 1 A Hopf bifurcation occurs when λ = iω solves the characteristic equation From complex variables, we match the magnitudes and arguments: (iω + γ)(iω + k) = A ( ) ω ( ω ) Θ(ω) arctan + arctan γ k = π ωt 1 p.23/37

33 Analysis of Characteristic Equation Remains to analyze (λ + γ)(λ + k) = Ae λt 1 A Hopf bifurcation occurs when λ = iω solves the characteristic equation From complex variables, we match the magnitudes and arguments: (iω + γ)(iω + k) = A ( ) ω ( ω ) Θ(ω) arctan + arctan γ k = π ωt 1 Solve for ω by varying parameters such as γ p.23/37

34 Hopf - Argument Principle Link to Var. Vel. Anal. p.24/37

35 Variable Velocity of Maturing Physiologically, erythrocytes mature more rapidly with increasing EPO p.25/37

36 Variable Velocity of Maturing Physiologically, erythrocytes mature more rapidly with increasing EPO Velocity of maturation, V (E), is a non-decreasing function, e.g. V (E) = κ 1E κ 2 + E p.25/37

37 Variable Velocity of Maturing Physiologically, erythrocytes mature more rapidly with increasing EPO Velocity of maturation, V (E), is a non-decreasing function, e.g. V (E) = κ 1E κ 2 + E Method of characteristics leaves a threshold-type functional equation rather than simpler delay equation p.25/37

38 Variable Velocity of Maturing Physiologically, erythrocytes mature more rapidly with increasing EPO Velocity of maturation, V (E), is a non-decreasing function, e.g. V (E) = κ 1E κ 2 + E Method of characteristics leaves a threshold-type functional equation rather than simpler delay equation Linear analysis of the age-structured model relatively simple p.25/37

39 Linear Analysis with Variable Velocity The modified characteristic equation for the threshold-type functional equation becomes (λ + γ)(λ + k) e βµ 1 f ( M)(V 1 + V 2 e λµ F ) = 0 where V 2 V 1 V (Ē)S 0(Ē) V (Ē) V (Ē)S 0(Ē) V (Ē)S 0(Ē) V (Ē) p.26/37

40 Linear Analysis (continued) This has the form (λ + γ)(λ + k) + α 1 + (α 2 α 1 )e λµ F = 0 with α 1 = V 1 and α 2 positive. p.27/37

41 Linear Analysis (continued) This has the form (λ + γ)(λ + k) + α 1 + (α 2 α 1 )e λµ F = 0 with α 1 = V 1 and α 2 positive. Since α 2 = A from previous characteristic equation, the α 1 shifts our geometric diagram above to the right with a smaller radius circle. p.27/37

42 Linear Analysis (continued) This has the form (λ + γ)(λ + k) + α 1 + (α 2 α 1 )e λµ F = 0 with α 1 = V 1 and α 2 positive. Since α 2 = A from previous characteristic equation, the α 1 shifts our geometric diagram above to the right with a smaller radius circle. This can be readily seen to stabilize the model. p.27/37

43 Hopf - Variable Velocity Link to Anal. Link to Hopf p.28/37

44 Auto-Immune Induced Anemia Rabbits were injected with antibodies to their Red Blood Cells p.29/37

45 Identify Parameters Physiological parameters found for rabbit Extensive literature search Some gaps using other species studied Difficult, but essential process Increasing the destruction of Erythrocytes, γ, causes a Hopf Bifurcation Model most sensitive to parameters γ and µ F Variable velocity of maturation stabilizes the model p.30/37

46 Simulation p.31/37

47 Phlebotomy (Blood Donation) Normal blood donation is about 8% of blood O 2 sensors near kidneys probably sense concentration, not M(t) Blood donation loses erythrocytes and plasma, no concentration change Plasma recovers quickly p.32/37

48 Model for Phlebotomy Define the hemoglobin concentration h(t) = H M(t) M(t) + ρ(t) The plasma function chosen to fit data ρ(t) = α [ 1 + (β 1 t 0.08)e β 2t ] The age-structured model remains the same except E(t) = f(h(t)) ke(t) Fit model to data of Maeda et al and Wadsworth p.33/37

49 Model a Phlebotomy p.34/37

50 Model Compared to Data p.35/37

51 Summary The mathematical model provides a good example of how age-structured models are related to models with delays. The modeling of satiation for destruction of erythrocytes could prove valuable in other population models. The model for erythropoiesis can be fit to existing data and can hopefully provide insight into the study of some hematopoietic diseases. It is unlikely to aid in the study of normal individuals for improved blood donation schemes. p.36/37

52 Summary (cont) Our numerical simulations and analytical study show how a variable velocity of maturation stabilizes the model. This implies that plasticity in the precursor compartment may be an important evolutionary adaptation. Current studies have identified the most significant parameters in the model, which could give insight to the likely causes of the disease states and possible therapeutic approaches. New studies examine thrombopoietic systems using a multi-compartment model to account for size structures of megakaryocytes. p.37/37