Summary of the experimental strategy.

A) Manual curation of FLT3-ITD-specific Prior Knowledge Network (PKN). B) Multiparametric analysis of signaling perturbations in TKI sensitive and TKI resistant cells. C) Model generation through the CellNOptR tool. D) Prediction of combinatorial treatments restoring TKI sensitivity. E) In vitro validation of novel combinatorial treatments. F) In silico prediction of co-treatment outcome in AML patients.

Generation of the training dataset.

A) Schematic representation of the FLT3-ITD PKN manual curation, integration of data-driven edges and manual integration of RTKs pathways involved in AML.

B) Schematic representation of the experimental design: FLT3ITD-JMD and FLT3ITD-TKD cells were cultured in starvation medium (w/o FBS) overnight and treated with PI3Ki, MEKi, mTORi and GSK3i, JNKi and p38i, in presence or absence of the FLT3i Midostaurin for 90 minutes. Then, the cells were stimulated either with IGF1 or TNFα for 10 minutes. Control cells were starved and treated with Midostaurin for 90 minutes. After treatment, samples were collected, and cell lysates were analyzed with a xMAP-based assay through the MagPix instrument. Per each experimental condition we measured the phosphorylation levels of 14 sentinels.

C) Network representation of a compressed PKN that shows the essential pathways monitored through the perturbation experiment. The perturbed nodes are tagged with a drug icon, and the measured nodes are colored green.

D) Principal Component Analysis (PCA) of FLT3ITD-JMD and FLT3ITD-TKD cells in the multiparametric analysis. Each point represents a different experimental condition.

Optimized Boolean models recapitulate the different sensitivity of FLT3ITD-JMD and FLT3ITD-TKD cells to TKI.

A) Color-coded representations of the experimental activity modulation (T90 – T0) of sentinel proteins used to train the two Boolean models (upper panel) and the average prediction of protein activities in the family of 100 best models (central panel). The protein activity modulation ranges from −1 to 1 and is represented with a gradient from blue (inhibited) to red (activated). The absolute value of the difference between experimental and simulated protein activity modulation (lower panel) is reported as a gradient from light yellow (error < 0.5) to red (1.85).

B-C) FLT3ITD-JMD (B) and FLT3ITD-TKD (C) high-confidence Boolean models. Perturbed proteins in the experimental setup are marked in red or green if inhibited or stimulated, respectively. Sentinel proteins are reported in blue. The edges’ weight represents their frequency in the family of 100 models and only the high-confidence ones (frequency > 0.4) are reported. Orange edges are cell-specific links.

D) Cartoon of the in silico conditions simulated to analyze the different TKI sensitivity of the FLT3ITD-JMD and FLT3ITD-TKD Boolean models. Untreated condition: TNFα+IGF1; FLT3i: TNFα+IGF1+FLT3 inhibition.

E) Heatmaps (left) report the activation level of positive and negative phenotype regulators present in the two Boolean models. Bar plots (right) showing the proliferation activation and apoptosis inhibition levels in untreated and FLT3i conditions in the steady states of FLT3ITD- JMD (blue) and FLT3ITD-TKD (yellow) Boolean models.

In silico simulation of the FLT3ITD-TKD logic model allows the prediction of novel combinatorial treatment reverting TKI resistance.

A) Cartoon of the in silico simulation conditions.

B-C) Bar plot showing the in silico simulation of proliferation activation (B) and apoptosis inhibition (C) levels in untreated and FLT3i conditions in combination with knock-out of each of 10 crucial kinases in FLT3ITD-JMD (blue) and -TKD (yellow) cells.

D-E) In FLT3ITD-JMD (blue) and -TKD (yellow) cells treated with 100nM Midostaurin and/or 10uM SP600125 (JNK inhibitor) for 24h, the percentage of Annexin V positive cells (D) and the absorbance values at 595nm (E), normalized on control condition, are reported in bar plots.

F) Primary samples from AML patients with the FLT3ITD-TKD mutation (n=2, yellow bars) or the FLT3ITD-JMD/TKD mutation (n=3, blue bars) were exposed to Midostaurin (100nM, PKC412), and JNK inhibitor (10µM, SP600125) for 48 hours, or combinations thereof. The specific cell death of gated AML blasts was calculated to account for treatment-unrelated spontaneous cell death. The bars on the graph represent the mean values with standard errors.

G) In FLT3ITD-JMD (blue) and FLT3ITD-TKD (yellow) cells treated with 100nM Midostaurin and/or 10uM SP600125, followed by 10’ of TNFα 10ng/ml, the protein levels of phospho-JNK (T183/Y185), JNK, phospho-CDK1 (Y15), phospho-CDK1 (T161), CDK1, phospho-CDK2 (T160), CDK2, CyclinB1, CycinE2, and Vinculin were evaluated by western blot analysis.

H) Cytofluorimetric cell cycle analysis of DAPI stained FLT3ITD-JMD (blue) and FLT3ITD-TKD (yellow) cells treated with 100nM Midostaurin and/or 10uM SP600125 (JNK inhibitor) for 24h.

I) Cartoon of the molecular mechanism proposed for FLT3ITD-JMD and FLT3ITD-TKD cells.

FLT3-ITD patient-specific Boolean models

A) Schematic representation of our computational approach to obtain personalized logic models.

B) Hierarchical clustering of patients according to their clinical characteristics (response to chemotherapy, vital status, and AML recurrence). Resistant, alive or deceased responders, and deceased with general or FLT3-ITD AML recurrence patients are reported in purple, light or dark green, and light or dark brown, respectively.

C-D) Hierarchical clustering of patients according to their mutational profile (C) and their expression profile of 262 genes (D).

E) Heatmap representing patient-specific in silico apoptosis inhibition (left panel) and proliferation levels (right panel) upon each simulation condition.

F) Patient-specific (JMD1 and TKD2) Boolean models. In the JMD1 model (left panel), nodes’ activity has been simulated in control (bottom-left part) and FLT3 inhibition conditions (upper-right part). In the TKD2 model (right panel), nodes’ activity has been simulated in FLT3 inhibition (bottom-left part) and FLT3 and JNK co-inhibition conditions (upper-right part).

Small molecule inhibitors and stimuli for the multiparametric analysis

xMAP analytes