RAS activation induces synthetic lethality of MEK inhibition with mitochondrial oxidative metabolism in acute myeloid leukemia

Despite recent advances in acute myeloid leukemia (AML) molecular characterization and targeted therapies, a majority of AML cases still lack therapeutically actionable targets. In 127 AML cases with unmet therapeutic needs, as defined by the exclusion of ELN favorable cases and of FLT3-ITD mutations, we identified 51 (40%) cases with alterations in RAS pathway genes (RAS+, mostly NF1, NRAS, KRAS, and PTPN11 genes). In 79 homogeneously treated AML patients from this cohort, RAS+ status were associated with higher white blood cell count, higher LDH, and reduced survival. In AML models of oncogenic addiction to RAS-MEK signaling, the MEK inhibitor trametinib demonstrated antileukemic activity in vitro and in vivo. However, the efficacy of trametinib was heterogeneous in ex vivo cultures of primary RAS+ AML patient specimens. From repurposing drug screens in RAS-activated AML cells, we identified pyrvinium pamoate, an anti-helminthic agent efficiently inhibiting the growth of RAS+ primary AML cells ex vivo, preferentially in trametinib-resistant PTPN11- or KRAS-mutated samples. Metabolic and genetic complementarity between trametinib and pyrvinium pamoate translated into anti-AML synergy in vitro. Moreover, this combination inhibited the propagation of RA+ AML cells in vivo in mice, indicating a potential for future clinical development of this strategy in AML.


Supplemental Figure 3. Heterogeneous activity of MEK inhibitors against RAS+ AML. A.
We applied the target selective inhibitors library (592 unique compounds) to CTR (cultured with GM-CSF as indicated by *) and NF1 KO_2 TF-1 cells for 48h, before cell viability quantification using the uptiblue fluorescent reagent. Each dot on the graph represents the mean of three independent experiments. B. Dose-response curves of the MEK inhibitor trametinib (from 10 -6 to 13.7x10 -9 M) on CTR (with IL-3, as indicated by *), NF1 KO_1 , NF1 KO_2 and NRAS G12D Ba/F3 cells. The uptiblue luminescent reagent measured cell viability. C. Experiments in CTR, NF1 KO_1 and NF1 KO_2 TF-1 and Ba/F3 cells cultured with GM-CSF (TF-1) or IL-3 (Ba/F3) as indicated and dose-range concentrations of trametinib. Left panel: cell viability measured by the uptiblue reagent in TF-1 cells cultured with GM-CSF. Middle panel: annexin V flow cytometry assay in TF-1 cells cultured without or with GM-CSF as indicated. Right panel: cell viability measured by the uptiblue reagent in Ba/F3 cells cultured with IL-3. D. CTR (with GM-CSF, as indicated by *) and NF1 KO_1 TF-1 cells were cultured with or without 5nM trametinib for 24h. Protein extracts were immunoblotted using anti-phospho-ERK, anti-PARP, anti-caspase-3 and anti-β-actin antibodies. E. Nine human AML cell lines with different RAS mutational status were assayed for in vitro sensitivity to trametinib. The OCI-AML2 cell line was recently shown to bear a RAF1 activating fusion 2 . We detected a KRAS-mutated subclone at a 7% VAF in Kasumi-1 cell line, which we did not led us to consider Kasumi-1 among RAS-mutated AML cell lines due to the low VAF and the absence of other reports of this mutation. F. Immunoblots in a panel of AML cell lines using anti-phospho-ERK,phospho-STAT5, -phospho-P70S6K, -ERK, STAT5, P70S6K and HSC70 antibodies. G. Number of L-CFU colonies in RAS+ (n=15) and RAS-(n=24) untreated primary AML samples (Log10 scale). H. Protein extracts from three and five RAS+ and RAS-primary AML samples, respectively, incubated with vehicle or 25nM trametinib were investigated for ERK phosphorylation and expression by Western blot. Underlined samples have results from L-CFU assays as presented in Figure 3G when the other were not considered as having a significant ex vivo L-CFU growth. I. In the CeGAL study, 114 AML samples (RAS+, N=34, RAS-, N=75) were incubated with dose-range Trametinib in liquid culture. Cell viability was determined after 48h using a luminescent reagent and half-maximal effective concentrations (EC50) values of Trametinib were calculated for each sample.*:p<0.05. J. BEAT AML data on inhibitory concentration 50 (IC50) values of primary AML cells exposed to the MEK inhibitors trametinib or selumetinib dependent on their RAS+ status as indicated. Heatmap representation of calculated basal, maximal and ATP-linked OCR. E. Glucose consumption and lactate production in CTR*, NF1 KO and NRAS G12D cells incubated with vehicle or 250nM pyrvinium for 24h. F. CTR*, NF1 KO_1 or NRAS G12D TF-1 cells were incubated during 6h with vehicle (CTR) or 25nM trametinib in bioenergetic analysis assays measuring OCR. Results are presented as a heatmap of calculated basal, maximal and ATP-linked OCR. G. Glucose consumption and lactate production in CTR*, NF1 KO and NRAS G12D TF-1 cells incubated with vehicle or 25nM trametinib for 24h. *: indicate that Ba/F3 and TF-1 CTR cells are incubated with IL-3 or GM-CSF, respectively. Vertical bars indicate standard deviations. *:p<0.05, **:p<0.01, ***:p<0.001.

Supplemental Figure 6. Synergy between the MEK inhibitor trametinib and pyrvinium pamoate in RAS activated cells. A.
Combination drugs dose-range assays in CTR (with IL3), NF1 KO and NRASG12D Ba/F3 cells incubated with pyrvinium and/or trametinib for 48h. Heat maps provided the combined results of three independent experiments. Synergy scores calculated by the SynergyFinder software (24) from these matrixes are provided for each condition at the top of the heatmap. B. L-CFU assays in RAS+ primary AML samples incubated with vehicle, 50nM trametinib, 250nM pyrvinium pamoate or trametinib/pyrvinium combination (combo) during 7 days. Two-by-two comparisons between pyrvinium/trametinib, trametinib/combo and pyrvinium/combo represented with a connecting line between each condition for each sample. Statistical analysis was performed using a Wilcoxon matchedpairs signed rank test. C. HL-60 cell line transduced with a vector expressing luciferase (HL-60 Luc+) was injected to immunodeficient NSG recipient mice. Treatment with vehicle, 0.25mg/kg/d trametinib (oral gavage), 0.5mg/kg pyrvinium (intraperitoneal injection) or combination of trametinib and pyrvinium started the day of transplantation (N=5 mice per treatment group). D. Evolution of mice weight during the HL60 luc+ CLDX experiment. A dashed line indicates the 10% up and down margins surrounding the initial weight (100%). The number of mice at the beginning and at the end of the experiment are indicated following the color code for each experimental group. E. Quantification of murine hematopoietic cells using flow cytometry mCD45 staining at day 22 of the HL60 luc+ CLDX experiment. F. Evolution of mice weight during the PDX#1 experiment. A dashed line indicates the 10% up and down margins surrounding the initial weight (100%). The number of mice at the beginning and at the end of the experiment are indicated following the color code for each experimental group. As three mice died at day two of the treatment in the combo group, pyrvinium dose was reduced from 0.5 to 0.25mg/kg/d, as indicated, without further toxicity problem during the further three weeks of treatment. G. Quantification of murine hematopoietic cells using flow cytometry mCD45 staining at different time points of the PDX#1 experiment. *: indicate that Ba/F3 CTR cells are incubated with IL-3. Vertical bars indicate standard deviations. ns: not significant. ***:p<0.001. Table 1. RAS mutations found in AML cell lines. Targeted DNA sequencing was performed on a panel of human AML cell lines. NRAS or KRAS mutations may be absent or present, with details of nucleotide substitution and amino acid modification.  To identify differentially expressed genes, we applied a classical analysis of variance (ANOVA) with an FDR permutation-base for each gene. We created a new matrix with only the significant ANOVA site and performed Z-scoring of rows. Hierarchical clustering by Pearson's dissimilarity and average linkage and principal components analysis (PCA) were conducted in an unsupervised fashion to control for experimental bias or outlier samples. We set a filter for those genes that displayed at least a ≥1,5 or ≤-1,5-fold difference in expression between groups and achieved an FDR of <0.05. Data were then interrogated for evidence of biologic pathway dysregulation using Gene set enrichment analysis (GSEA, Broad Institute). Enrichment rates were considered significant when the P-value <0.05 and the FDR ≤0.1.

RNA-seq quantification
Paired-end reads were aligned to the human reference transcriptome "Homo_sapiens.GRCh38.cdna.all.fa.gz" using Salmon 1.3.0. For mapping, we used salmon quant command with parameters like -gcBias, --useVBOpt, --seqBias, and -validateMappings, correcting specific biases in the sequenced reads and increasing the sensitivity and specificity of the mapping.
After mapping, R (R version R version 4.0.2) packages tximport (version 1.14.0) and DESeq2 (version 1.26.0) were used to summarize transcript-level estimate to the gene-level estimates. Further, we used the raw gene-level abundance estimates from salmon to perform to identify differentially expressed genes.

Differential gene expression analysis
Differential gene expression analysis was performed using an R Bioconductor software package edgeR 4 . First, we filtered the genes with a meager count using an in-built function in edgeR filterByExpr(), and the library size was recomputed. Following this, the normalization factors were estimated using the calcNormFactors() function, and the per gene dispersion was assessed using the estimateDisp() function in edgeR. Here, a design formula (linear model, ~Rasopathy) was provided for dispersion calculation. Finally, the negative binomial generalized linear model was fit using glmFit() function, and likelihood ratio test glmLRT() was used to test for differential expression. We assessed the differentially expressed genes between the Rasopathy-Yes vs. Rasopathy-No conditions. The topTags() function was used to extract the top-ranked genes, and the results were presented as a volcano plot using an R Bioconductor package EnhancedVolcano. We also performed gene set enrichment analysis 5 using the oncogenic signature gene set in which we manually added several gene sets from recent literature [6][7][8][9] .

Lentivirus production and cell line infections
Lentivirus production and cell line infections were done as previously described 10 . Briefly, we used 293-T packaging cells to produce all of the constructed recombinant lentivirus through co-transfection of these cells with the packaging plasmids pMD2.G and psPAX2 encoding lentiviral proteins (Gag, Pol, and Env) using Lipofectamine 2000 Transfection Reagen (Thermo Fischer Scientific, Waltham, Massachusetts, US). Supernatants were collected and ultracentrifuged for 48 h after transfection over two consecutive days, and then stored at -80°C. AML cell lines were seeded at 2x10 6 /ml and 10μl of lentiviral supernatants were added for 24h. Cells were further selected with puromycin, or cell sorted with an ARIA 3 cytometer in case of GFP or mCherry expression as selection marker.

TF-1 differentiation
Cells were washed 3 times in PBS to remove GM-CSF, and then cultured 7 days with 2 IU/mL EPO. Cells were spin down to collect pellets in which color change from white to purple reflected hemoglobinization.

Trypan Blue dye exclusion assay
The Trypan Blue dye (Sigma Aldrich) exclusion assay was used to determine the number of viable cells present in the cell suspension. A Malassez counting chamber was filled with the cell suspension mixed with the dye. Cells were then visually examined and counted under a microscope: cells taking up the dye were considered dead and cells excluding the dye were considered alive.

Immunohistochemistry
Femurs, tibias and spleens of mice were fixed for 24h in 4% paraformaldehyde. Decalcification was carried out using 15% formic acid at 4°C for 4h, followed by a second fixation in 4% paraformaldehyde during 24h. Samples were paraffin embedded and then sliced using a microtome.

Prestwick chemical library® (PCL) screen
Primary screening PCL compounds were added 24h after cell seeding: 2µL/well of each compound (from 10 mM stock plates) were diluted in 8µL of DMSO in order to obtain 2mM/well stock plates. Next, 2µl/well at 2mM were mixed in a predilution plate containing 78µL/well of cell media, and 10µL of this solution were dispensed into each 384-cell plate well using the MultiChannel Arm™ 384 (TECAN, Männedorf, Switzerland), to a final concentration of 10µM and 0.5% of DMSO. The plates were incubated for 72h at 37°C in 5% CO2 before submitted to CellTiter-Glo® 2.0 cell viability assay. Biological duplicates were performed.

Secondary screening
We selected the top 60 hits from our primary screen to perform a set of dose-response experiments using the same workflow. We performed three independent experiments using 8 consecutive threefold dilutions from 10 -6 M to 4.57 x 10 -9 M. Both primary and secondary screens were performed on the same batches of viably frozen cells and at the same passage stage (four passages from thawing).

Data processing
Values of all plates were visually inspected for systematic bias (i.e., edge effects). Raw luminescence intensity values were log2 transformed, then median polishing was applied to progressively corrects columns, rows, and entire plates by subtracting their median, repeating until convergence of values.
Column and row corrections were computed separately for each experimental replicates, but across all plates within the replicate in combination. Hits were identified by the robust Z-score method under the assumption that most compounds are inactive and can serve as controls as follows: sample median and median absolute deviation (MAD) were calculated from the population of screening data points (named as sample) and used to compute robust Z-scores according to the formula: " − "=( − ) / (1.4826× ), where x corresponds to the compound-treated data point and MAD is the median of the absolute deviation from the median of the tested wells. A compound was identified as a hit if the RZ-score was < -2 or > 2 pointing in the same direction in both replicates. Compounds having a RZ-score < -2 corresponds to those considered reducing cell viability.
The same analysis pipeline was applied to each cell lines tested. Final values correspond to the mean RZ-score for each compound.
In dose-response experiments, compound activity was normalized on a per-plate basis by dividing the value in each well by the median value of the control wells (100% cell viability). For each compound, a four parameters log-logistic model was then fitted on the pooled replicate data with the R package drc. Compound activity was then summarized by computing a Drug Sensitivity Score (DSS, modified from 11 ), the area under the curve normalized by the area of an inactive compound (100% viability at all doses). Finally, compounds were scored by calculating their median dose-effect (effective dose 50, ED50) x DSS value and the top-10 compounds among which were the classical AML chemotherapies cytarabine and daunorubicin were selected for further assays.

Synergistic cell viability assays
We performed dose-range experiments of trametinib and pyrvinium single-agents or combination and assessed cell viability after 48h using the Uptiblue® reagent. We used the SynergyFinder online software to calculate synergy scores computed using the zero interaction potency (ZIP) model. All experiments were done three times separately and pooled data were analyzed.