Fig 1.
Illustration of the difference between the IC50 value assessed directly from experimental data and model-based deconvolution of mechanisms explaining downstream inhibition.
The cartoon shows the percent inhibition of granulocyte-macrophage progenitors (GMP) cells at increasing doses of drug. The IC50 value derived from this experimental data represents the cumulative loss along the pathway leading to the formation of GMP, including drug impacts on the early progenitors, hematopoietic stem cells (HSC) and multi-potent progenitors (MPP). On the right, model-derived Emax and EC50 values represent the drug’s direct effect on cell-killing or anti-proliferation on individual cell types, including HSC, MPP and GMP. In the proposed modeling framework, it is possible to separate effects on a specific cell type, from propagated effects, such as loss of upstream progenitors. The color gradients above all curves represent the percent inhibition (red/yellow color) or percent decrease (blue/yellow color) and these colors are used in later figures.
Fig 2.
A QSP model for describing in vitro hematopoiesis, cell kinetics, and drug effects.
The model contains 15 species representing 13 different cell types: hematopoietic stem cells (HSC), multi-potent progenitors (MPP), granulocyte-macrophage progenitors (GMP), monocyte progenitors (monocyte prog), monocytes (monocyte-lin), granulocyte progenitors (gran-lin prog), granulocytes (gran-lin), neutrophils, lymphoid progenitors, B-cells (B-lin), megakaryocyte cells (MK-lin), early erythroid cells (erythroid I), and late erythroid cells (erythroid II). The model also tracks total viable cells (this is the sum of the 13 hematopoietic lineages and only includes living cells), and total dead cells (we cannot distinguish between cell types in total dead cells). In addition, drug is also represented as a species in the model. The model includes a quiescent neutrophil population to adjust for the large number of neutrophils observed. This is similar to other published reports of semi-mechanistic modeling [14]. The solid arrowed lines in the diagram denote reactions resulting in the formation of products pointed to at the end of the arrow, whereas dashed lines denote reactions whereby the species in connection is both the substrate and product. Under no drug treatment, it is assumed that only the most mature cell types die, as indicated by the red arrows. For the simplicity of illustration, the anti-proliferative and killing effects of drugs on each cell type are not explicitly shown on the diagram, except for the drug-induced killing contributing towards the total number of dead cells. Shapes and colors do not encode any model information but are just meant to distinguish between cell types (A). The rate of increase in the number of a cell type (e.g. MPP) depends on the input flux from its predecessor cell type (e.g. HSC), proliferation flux (κ_MPP × [MPP]), as well as the fraction renewed (ρ_MPP) versus differentiated (1 - ρ_MPP). Note that drug effects are not represented in this diagram (B). Branching parameters are applied for progenitors that yield multiple cell types. All branching parameters sum to one for a given progenitor. Example shown for MPP (C). The model captures drug effects at the cell death reaction (“Drug cell-killing effects”) and at the proliferation reaction (“Drug anti-proliferation effects”). In this abbreviated schematic we show the drug effects on the GMP cell type and show the example equations for this species, but they are adapted for all other cell types in the model. Anti-proliferation drug effects alter the basal proliferation rate for the cell type, and this affects both proliferation and differentiation (D).
Fig 3.
In vitro QSP model captures proliferation kinetics across cell types in the absence of drug treatment.
Cells/mL are plotted against time (days) for 13 live cell types and the total dead cells. Error bars represent standard error of the mean of six unique donor samples. In all plots, experimental data are shown in blue dashed lines with “+” markers, and model data are shown in solid pink lines.
Fig 4.
Concentration-response data captures drug effects across multiple lineages.
We tested the concentration-response for ribociclib (A), abemaciclib (B), dinaciclib (C), palbociclib (D), paclitaxel (E), docetaxel (F), thalidomide (G), and pictilisib (H). Extent of shading represents relative cell counts where black or white correspond to fewer or equivalent cell counts relative to vehicle control wells respectively. In all cases, the x-axis represents doses in nM concentration (0.2, 1, 5, 25, 100, 500, 2500), and each row represents a different cell species (top: Hematopoietic Stem Cells, MPP, GMP, Gran-lin prog, Gran-lin, Monocyte prog, Monocyte-lin, Neutrophil-lin, Erythroid-lin I, Erythroid-lin II, MK-lin, B-lin, totalViableCells, and bottom: totalDeadCells).
Fig 5.
Percent inhibition plotted using IC50 values.
Percent inhibition is plotted for abemaciclib (A), dinaciclib (B), ribociclib (C), palbociclib (D), paclitaxel (E), docetaxel (F), thalidomide (G), and pictilisib (H). Percent Inhibition was calculated using IC50 values fitted to the concentration-response data. Black to yellow shading represents increasing inhibition of each cell type: note that IC50s for MK- and B-lin could not be determined from the treatment data, hence these cell types are shown as having 0% suppression. The results indicate that generally a broad set of cell types are inhibited under drug treatment. In all cases, the x-axis represents doses in nM concentration (0.2, 1, 5, 25, 100, 500, 2500), and each row represents a different cell species (top: HSCs, MPPs, GMPs, Mon, Gran, Neut, ErythI, ErythII, MK, and bottom: B lin).
Fig 6.
PCA of compounds using the log of experimentally derived IC50 values.
A set of 46 marketed compounds, including 15 negative controls were grouped using principal component analysis. Marker color indicates drug class. The result shows that while almost all the compounds ranged between classic chemotherapies and negative controls on the first principal component, the mechanisms differentiating the compound classes cannot be well identified using the IC50 values alone.
Table 1.
Examples of EC50 and EmaxT parameter values for compounds with known hematopoietic toxicity effects.
Thalidomide is included as a negative control and EC50 values above the max tested dose of 2500 nM are extrapolated from model fitting.
Fig 7.
EmaxT expressions show relative drug effects at increasing concentrations.
EmaxT effects are plotted against concentration for abemaciclib (A), dinaciclib (B), ribociclib (C), palbociclib (D), paclitaxel (E), docetaxel (F), thalidomide (G), and pictilisib (H). Color bars represent magnitude of EmaxT effect and concentrations are in nM. The results show that for select targeted therapies, the treatment data can be explained by drug effects on a parsimonious set of cell types. In all cases, the x-axis represents doses in nM concentration (0.2, 1, 5, 25, 100, 500, 2500), and each row represents a different cell species (top: HSC, MPP, GMP, GranP, MonoP, Mon, Neut, ErythroidI, ErythroidII, MK, LymP, and bottom: B lin).
Fig 8.
PCA separates drugs based on EmaxT effects.
The 51 compounds are plotted in PCA space (A). Marker color corresponds to drug class. Drugs and variables contributing to the top two components are plotted in PCA space (B). Note: in figure A there is only one drug in class 5 and it is marked with an * to distinguish this compound from the remaining class 2 drugs.
Fig 9.
Mechanistic effects of developmental compounds per cell type compared to the reference set.
EmaxT parameters per cell type are plotted against the EC50 values for each cell type. Marker shape represents cell type and marker shading represents either the reference set (black gradient, all figures), PI3K inhibitors (multi-color, A), or class 6 (purple, B). The same data are broken into cell type plots for PI3K inhibitors: HSCs, GMP, granulocytes, neutrophils, monocytes, B-cells (C,D,G,H,K,L); or for class 6 drugs: HSCs, GMP, granulocytes, neutrophils, monocytes, B-cells (E,F,I,J,M,N), respectively. In all plots, the x-axis represents the EC50 concentration in nM (0.01, 0.1, 1.0, 10, 100, 1000, 10000) and the y-axis represents EmaxT values from 0-4.The dashed line represents where EmaxT = 1.0. Only a subset of cell type plots is shown. Note: only compounds with EmaxT effects > 0.1 are plotted; compounds such as the negative control, thalidomide, had no effects above 0.1 and are not shown.