Abstract
Background and Objective
Acute myelogenous leukemia (AML) is a common blood cancer marked by heterogeneity in disease and diverse genetic abnormalities. Additional therapies are needed as the 5-year survival remains below 30%. Trametinib is a mitogen-activated extracellular signal-regulated kinase (MEK) inhibitor that is widely used in solid tumors and also in tumors with activating RAS mutations. A subset of patients with AML carry activating RAS mutations; however, a small-scale clinical trial with trametinib showed little efficacy. Here, we sought to identify transcriptomic determinants of trametinib sensitivity in AML.
Methods
We tested the activity of trametinib against a panel of tumor cells from patients with AML ex vivo and compared this with RNA sequencing (RNA-Seq) data from untreated blasts from the same patient samples. We then used a correlation analysis between gene expression and trametinib sensitivity to identify potential biomarkers predictive of drug response.
Results
We found that a subset of AML tumor cells were sensitive to trametinib ex vivo, only a fraction of which (3/10) carried RAS mutations. On the basis of our RNA-Seq analysis we found that markers of trametinib sensitivity are associated with a myeloid differentiation profile that includes high expression of CD14 and CLEC7A (Dectin-1), similar to the gene expression profile of monocytes. Further characterization confirmed that trametinib-sensitive samples display features of monocytic differentiation with high CD14 surface expression and were enriched for the M4 subtypes of the FAB classification.
Conclusions
Our study identifies additional molecular markers that can be used with molecular features including RAS status to identify patients with AML that may benefit from trametinib treatment.
Supplementary Information
The online version contains supplementary material available at 10.1007/s40268-024-00491-5.
Key Points
| Patients with AML exhibit high variability in response to drugs such as trametinib | |
| RAS status alone is not predictive of trametinib sensitivity in AML blasts | |
| High expression of myeloid differentiation genes, such as CD14, is strongly predictive of trametinib sensitivity in AML blasts |
Introduction
Acute myeloid leukemia (AML) is a common cancer of the myeloid lineage of blood cells. There are approximately 60,000 new cases yearly in the USA. High dose chemotherapy remains the standard treatment for many patients, which frequently results in relapse. Targeted treatments are available for patients who present with mutations in FLT3, IDH1/2, or meet certain conditions related to age and comorbidities, but only a fraction of the patient population meets any of these criteria [1]. As a result, the 5-year survival rate for AML remains below 30%.
AML is a heterogeneous malignancy and the prognosis is variable based on clinical features as well as leukemia-specific genetic abnormalities. Molecular diagnostics has uncovered heterogeneity and genomic complexity within AML based on the presence or absence of cooperating mutations within functional categories, such as epigenetic regulators, cell signaling, and proliferation pathways, and master hematopoietic transcription factors. Although the treatment of AML still includes high dose chemotherapy, an enhanced understanding of this disease has led to the recent development of multiple targeted and selective treatment approaches, and our increasing awareness of functional AML subsets should inform new approaches. The application of therapies targeting known genetic and/or clinical susceptibilities could improve outcome for the large proportion of patients with AML who remain refractory to current treatments.
One potentially actionable susceptibility in AML is a set of RAS mutations that are found in 15–20% of adult cases [2]. Mitogen-activated extracellular signal-regulated kinase (MEK) inhibitors are a major class of drugs that are already approved in other cancer types to target MAPK activated tumors. Trametinib, in particular, is a MEK inhibitor that is indicated for the treatment of BRAF-mutant melanoma and non-small cell lung cancers [3, 4]. Trametinib has also been used in RAS activated tumors [5, 6]. A subset of patients with AML present with activating mutations in the MAP kinase pathway, including 15–20% of patients with AML who exhibit activation of RAS [7]. Although trametinib is not approved for use in AML, a phase II clinical trial employing trametinib to treat relapsed or refractory AML cases elicited a response in ~ 10% of patients with RAS mutation [6]. These data suggest that RAS mutations may predispose a patient to be sensitive to trametinib, but are not sufficient to predict response. Previously, an ex vivo drug profiling has shown that trametinib can show activity with variable efficacy with RAS mutant patients having the greatest sensitivity [2, 5, 8]. In addition to clinical, cytological, and genetic markers, AML cases can be stratified by their transcriptomic profile [9–11]. Recent attempts at using such transcriptomic profiles have led to some success in impacting patient outcomes [9–11]. However, thus far transcriptomic profiling and other panels of genetic, clinical, or molecular parameters fail to explain or predict all AML cases that display sensitivity to trametinib. Given the safety profile of trametinib, and that there are RAS mutations in some patients, we suggest that trametinib remains a potential candidate for a subset of patients with AML. Identifying which patient could benefit from trametinib treatment thus requires the discovery of additional biomarkers.
To find biomarkers of trametinib sensitivity in AML, we combined an ex vivo drug sensitivity assay and transcriptomic analysis on 37 primary tumor samples from patients with AML. As observed in other cohorts, ~30% of patient samples were highly sensitive to trametinib, including both patients with RAS mutations as well as those without. By combining these data with gene expression analysis, we highlight a list of genes that strongly correlate with ex vivo trametinib sensitivity. This list includes several markers of myeloid differentiation. Consistently, samples sensitive to trametinib have gene expression and cytologic profiles that resemble monocytes. In particular, we show that the hallmark monocytic differentiation marker CD14 is a robust marker of trametinib sensitivity, both at the RNA and protein level. Our results suggest that myeloid differentiation status, in particular CD14 levels, could be used as a biomarker along with RAS status to stratify patients into those that may benefit from trametinib treatment.
Material and Methods
Primary AML Samples and MEK Inhibitor Sensitivity Assay
Primary AML blasts obtained from the Penn Stem Cell and Xenograft Core (RRID: SCR_010035) were collected from peripheral blood or bone marrow samples and isolated by size-exclusion centrifugation apheresis or ficoll gradients. All samples were collected and experiments undertaken with the understanding and written consent of each subject. The study methodology confirmed to the standards set by the Declaration of Helsinki and were approved by the Penn institutional review board (IRB) committee (IRB approval #703185).
Overall, 5000 AML blasts in Iscove’s modified Dulbecco’s medium (IMDM) supplemented with 2% fetal bovine serum (FBS) were plated in assay ready plates containing a five-point dose response of trametinib in 0.2% Dulbecco’s modified Eagle’s medium (DMSO) for 72 h at 37 °C and 5% CO2. Cell viability was measured using ATPlite. Sample well data was normalized to aggregated 0.2% DMSO (n = 16) and 50 uM bortezomib control (n = 16) wells, and expressed as normalized percent inhibition [NPI = ((DMSOmean − Test well)/(DMSOmean − Bortezomibmean) × 100)]. NPI values were plotted against log10-transformed trametinib concentrations and IC50s were calculated. Drug sensitivity was defined using dose-response models generated by fitting a parametric logistic regression on lethality at five or six different drug concentration ranging from 0.0003 to 0.2 µM. We used the inflection point of the model curve, corresponding to the concentration that causes an inhibition halfway between baseline and maximum inhibition response, or IC50, as a measure of drug inhibitory activity. We assessed curve shape, IC50, and the maximum response value to determine drug potency and efficacy. Drugs with high potency have lower IC50 and drugs with high efficacy have higher maximum response values. Z-scores [z-score=(DMSOmean−Trametinibsignal/DMSOStandard Deviation] were generated from a second, independent assay and is measured as the number of standard deviations of trametinib inhibitory activity at 100 mM relative to the aggregated mean and standard deviation of 0.2% DMSO control wells (n = 32).
Transcriptome Analysis
RNA sequence (RNA-Seq) data from primary AML blasts was previously generated after total RNA extraction and poly(A) selection by paired-end Illumina sequencing (Gene Expression Omnibus dataset GSE142514). RNA-Seq data from normal hematopoietic cells were downloaded from the BLUEPRINT project (EGAD00001002308, EGAD00001002387, EGAD00001002417, EGAD00001002471, EGAD00001002507, EGAD00001002526; [12]), the Leucegene project (GSE48173; [13]), or Ranzani et al. (PRJEB5468; [14]). Sequencing adapters were removed using TrimGalore v0.6.6 (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/), reads were aligned with STAR v2.5.2a [15] against the human hg38 genome assembly and mapped reads were sorted and indexed using Samtools v1.11 [16].
Gene expression quantification was performed with Salmon v0.14.0 [17] in quasi-mapping mode using an index generated with Ensembl GRCh38 transcriptome release 94. Confounding factor correction for sequencing batch and tissue of origin was done using Combat [18]. Splicing variation quantification was performed using MAJIQ v2.0 [19] with transcriptome annotation Ensembl GRCh38 transcriptome release 94. Splicing data were corrected for sequencing batch and tissue of origin using MOCCASIN [20]. Complex splicing variations (variations that include more than two splice junctions) were removed from the analysis.
Cutoff in the variance analysis was determined from the knee-point of the curve as established with KneeLocator from the kneed package (https://kneed.readthedocs.io/en/stable/index.html) after fitting a spline over the data using the SciPy UnivariateSpline function. Hierarchical clustering was performed using SciPy’s clustering package and distance between clusters was calculated with the Ward variance minimization algorithm. Principal component analysis was performed using the decomposition function of the scikit-learn library. Gene set enrichment analysis was performed using GSEA v4.1.0.
Reverse Transcription-Quantitative Polymerase Chain Reaction (RT-qPCR)
Total RNA from primary AML blasts was reverse-transcribed using Moloney Murine Leukemia Virus reverse transcriptase and an equal ratio mix of oligo(dT) and random hexamer primers. Resulting cDNA was amplified using the Power SYBR Green PCR Master Mix (Applied Biosystems) with splice junction-spanning gene specific primers in a LightCycler 96 Instrument (Roche Life Science).
Statistics
Odds ratio in Figs. 1C and 2B were calculated using a Fisher exact test on contingency tables that only contained samples for which clinical/mutational status (Fig. 1C) or transcriptomic profile (Fig. 2B) was known. The association between splicing variations in RAS genes and trametinib sensitivity was assessed by comparing splicing levels in trametinib-resistant and sensitive samples for all splicing variations detected in these genes, applying the Benjamini/Hochberg method to correct for multiple hypothesis testing. Pearson correlations in Fig. 3B were calculated on genes that have a log(2) fold difference in expression of at least 2 between the 95th percentile and the 5th percentile of samples. The Pearson correlation coefficient cutoff corresponding to a 1% false discovery rate was obtained by calculating the value of the top 1% correlation coefficients obtained when randomly shuffling drug sensitivity data 1000 times between samples. T-tests were used in cell type-based one-on-one comparisons in Fig. 4C. Enrichment values and p-values for the gene networks in Fig. S3B were calculated using a hypergeometric test.
Fig. 1.
Trametinib sensitivity is independent of most clinical and genetic features. A Dose-response to the MEK inhibitor trametinib in primary blasts from 20 patients with AML. B Trametinib sensitivity is measured as the concentration killing 50% of blasts (IC50). Overall, 10 samples are highly sensitive to trametinib while the other 27 are highly resistant. C Association between trametinib sensitivity and clinical, demographic, or genetic features. The number of samples associated with each feature is shown in the third column over the total of samples for which information about the feature is available. The number of samples with the feature that are also sensitive to trametinib is shown in parentheses. Of all clinical and genetic factors, only RAS mutations are significantly associated with trametinib sensitivity. Fisher’s exact test. D RAS mutational status, CBFB–MYH11 fusions and M4 subtype classification shown relative to trametinib sensitivity
Fig. 2.
Trametinib sensitivity associates strongly with the expression profile of the most variable genes. A Distribution of variance in gene expression (left) or splice junction usage (right) among 37 primary AML samples; the most highly variable genes and splice junctions were used to perform hierarchical clustering analysis. B Hierarchical clustering shows that gene expression (left), but not splice junction usage (right), strongly associates with trametinib sensitivity (Odds ratio = 31.5, p value = 0.0003 for gene expression; odds ratio = 0.98, p value = 1.0 for splicing; Fisher exact test)
Fig. 3.
Analysis of a gene set whose expression levels correlate with trametinib sensitivity. A Flow chart of the analysis leading to the identification of top candidate markers of trametinib sensitivity. Pearson correlations were calculated with −log(IC50) values for the HTSC dataset or –(IC50 AUC) for the Beat-AML dataset. B Pearson correlations for the HTSC and Beat-AML datasets from gene expression variations meeting all criteria, ranked by combined correlation coefficients from both datasets. Solid black boxes mark genes with Gene Ontology annotations related to the MEK pathway. C RT-qPCR validations of genes correlating with trametinib sensitivity in seven additional AML samples. Mean expression levels are calculated with samples with undetectable expression valued at 0
Fig. 4.
Myeloid differentiation profile in trametinib-sensitive AML samples. A Top gene sets enriched among genes that correlate with trametinib sensitivity. Enrichment was calculated from the list of genes ranked in decreasing order of correlation coefficient. Only gene sets with corrected p values < 0.005 and normalized enrichment score above 2.1 are shown. The top gene set is boxed and its enrichment plot is shown in panel B. B Enrichment plot showing enrichment in genes upregulated in monocytes relative to dendritic cells among genes correlating positively with trametinib sensitivity. FDR q value = 0 at 1000 permutations. C Expression level of genes correlating with trametinib sensitivity in normal myeloid, lymphoid, and progenitor cells; ***p < 0.001, t test. D Principal component analysis performed with expression values of genes correlating with trametinib sensitivity from AML, monocyte, macrophage, dendritic cell, and CD34+ progenitor samples in our cohort (left) or the Beat-AML cohort (right). E High trametinib sensitivity among CD14+ blast samples
Results
Trametinib Sensitivity Cannot be Predicted from only Clinical and Genetic Features
To determine features that can predict trametinib sensitivity we first assessed the sensitivity of 37 primary AML blast samples (Table S1) to trametinib ex vivo. The response to trametinib was bimodal, with 10 out of 37 samples having IC50 in the low nanomolar range (< 3 nM) and the rest of samples showing no inhibition at the highest concentration of 5 µM (Fig. 1A and B, Table S2). We next assessed what clinical and genetic features are associated with trametinib sensitivity. Out of 16 clinical and 24 genetic features that we assessed, we found that only RAS mutations are significantly associated with trametinib sensitivity (Fig. 1C). We also found that RAS mutational status is significantly associated with trametinib sensitivity using data from the Beat-AML study [2], as were CBFB–MYH11 fusions and M4 FAB subtype classification (Fig. S1A). As expected, and in agreement with previous reports [6, 8], RAS mutant blasts are generally more sensitive to trametinib (Fig. 1D). However, a number of samples that are sensitive to trametinib have either no known RAS mutation, no CBFB-MYH11 fusion, and/or are not M4 subtypes. Thus, these features are insufficient to predict trametinib sensitivity alone and additional combinations of biomarkers are needed.
Trametinib Sensitivity Associates Strongly with Clusters Defined by Gene Expression
Since known clinical and genetic features are insufficient to predict trametinib sensitivity alone, we asked whether transcriptome profiling could provide additional information about what samples are sensitive to the drug. We performed bulk RNA-Seq and quantified gene expression and splicing variations on the 37 patients with primary AML samples. We decided to include splicing variations in addition to gene expression variations in our analyses because mutations in genes encoding splicing factors are commonly found in AML [21]; thus splicing variations might be prognostic factors in AML [22, 23]. Moreover, we recently observed that splicing variations frequently alter the expression level of AML-associated genes suggesting that standard DNA mutation analysis may be not be sufficient to describe the status of these genes, and that splicing variation can be used to stratify patients independent of other transcriptomic and genomic features [24, 25].
Using gene expression and splicing variation data from RNA-Seq (see Methods), we observed that a large fraction of the total variance in gene expression and splice junction usage existing between AML samples is concentrated in a small number of genes or splice junctions. Specifically, across our AML samples, 1463 out of 12,376 (11.8%) quantified protein-coding genes account for 40.6% of variance in gene expression, while 997 out of 20,765 (4.8%) binary variations in splicing (only two junctions per variation) account for 58.7% of variance in splice junction usage (Fig. 2A). Hierarchical clustering based on the expression of the 1463 most variable genes produced two well-defined groups (Fig. 2B). Strikingly, these two groups associate strongly with trametinib sensitivity (Odds ratio = 31.5, p value = 0.0003; Fisher exact test). Of note, the only trametinib-sensitive sample that clustered in group 2 has a known RAS mutation. In contrast, hierarchical clustering based on the usage of the 993 most variable splice junctions also produced two well-defined groups, but these groups do not associate with either gene expression or trametinib sensitivity (Odds ratio = 0.98, p value=1.0; Fisher exact test), and we found no splicing changes in RAS genes that are associated with trametinib sensitivity. Importantly, our analysis included a correction for batch effects (see Methods) and we indeed find that our resulting hierarchical clustering is not biased by sequencing batch, as evidenced by the absence of association between sequencing batch and cluster attribution (Fig. 2B).
Transcriptomic Markers of Trametinib Sensitivity
As hierarchical clustering by gene expression results in two groups that associate strongly with trametinib sensitivity, we sought to identify the specific genes that drive these differences and that may act as biomarkers of trametinib sensitivity. We performed a correlation analysis between gene expression and trametinib sensitivity in our cohort as well as in 433 samples from the Beat-AML cohort for which both trametinib sensitivity and RNA-Seq data are available [2] (Fig. 3A). We used a correlation analysis to accommodate the continuum of trametinib sensitivity values reported in the Beat-AML study. Starting with 12,144 protein-coding genes expressed in both cohorts, we focused on genes that have expression levels that exhibit at least a log(2) fold difference of 2 or greater between the 95th percentile and the 5th percentile of samples by expression. We then calculated the Pearson correlation coefficient between gene expression and IC50 in our assay or with the area under the IC50 curve in the Beat-AML cohort. We kept genes that correlate in the same direction in both cohorts and that have a correlation coefficient higher than that corresponding to a false discovery rate (FDR) of 1% [abs(Pearson r) > 0.414 in our assay, abs(Pearson r) > 0.174 in Beat-AML cohort] (Fig. S2A). This analysis yielded a total of 223 genes whose expression correlates with trametinib sensitivity, a large majority of which (194/223) exhibit a positive correlation between expression and drug sensitivity (Fig. 3B and Table S3). We obtained a high concordance in the list of correlating genes when we repeated the analysis using the z-score on cell survival after 72 h of 100 nM trametinib exposure as an orthogonal measurement of trametinib sensitivity (Fig. S2B), confirming the robustness of our drug sensitivity assays and the consistency of trametinib sensitivity in ex vivo AML samples.
To validate these genes, we performed RT-qPCR on all additional primary samples available with enough material (seven samples, five resistant and two sensitive) that were also subjected to the trametinib sensitivity assay, but not included in the original set and for which no RNA-Seq data were generated. As predicted, six genes selected for their strong positive correlation with trametinib sensitivity in the original analysis and known as surface markers or associated with the MAPK/MEK pathway also show higher expression in this independent analysis, although with variable consistency between genes (Fig. 3C). Notably, the correlation between CD14 expression and trametinib sensitivity was striking. CD14 is a hallmark surface marker of monocytic differentiation.
Trametinib-Sensitive AML Samples are Characterized by a Differentiated Myeloid Phenotype
Since we found a correlation of trametinib sensitivity with CD14 expression, we next investigated the transcriptomic signature associated with trametinib-sensitivity. We performed a gene set enrichment analysis (GSEA) against the entirety of the MSigDB database v7.4 for all genes whose expression positively or negatively correlates with trametinib sensitivity (Tables S4 and S5). Strikingly, many of the most enriched sets were related to monocytes or myeloid immunity, indicating that genes correlating positively with trametinib sensitivity are upregulated in differentiated myeloid cells (Fig. 4A). Furthermore, the most enriched set consists of genes that are upregulated in monocytes relative to plasmacytoid dendritic cells (Fig. 4B). We further show that genes correlating with trametinib sensitivity are expressed at a significantly higher level than the most variable genes in differentiated hematopoietic cells, where their expression is the highest in monocytes, while they are expressed at lower levels in progenitors (CD34+ cells, granulocyte-monocyte progenitors and common myeloid progenitors) (Fig. 4C). Principal component analysis with the 223 highly correlating genes confirms a high proximity in the gene expression profile of trametinib-sensitive samples to monocytes in both our cohort and in the Beat-AML cohort (Fig. 4D). This is consistent with the observation from the Beat-AML study that trametinib-sensitive samples are strongly enriched for the M4 subtype (Fig. S1), which is characterized by monocytic morphology. We note that the M4 subtype was underrepresented in our dataset so we did not have the power to observe a similar trend if one existed. Furthermore, analysis of a larger sample set of patients with AML for surface expression of monocytic surface marker CD14, demonstrates a striking correlation with higher trametinib sensitivity ex vivo (Fig. 4E). Finally, the CD14 protein-protein interaction network is enriched for components of the MAPK pathway (1.72-fold enrichment, p value = 0.0038, hypergeometric test; Fig. S3), suggesting a functional connection between CD14 expression and trametinib sensitivity. Together, these results indicate that trametinib-sensitive samples have a gene expression and cytologic profile similar to differentiated myeloid cells that most closely resembles that of monocytes.
Discussion
In this report we analyzed data from an ex vivo drug assay and RNA-Seq from primary AML patient samples to identify new molecular markers of ex vivo sensitivity to trametinib. As expected, AML blasts that carry RAS mutations are sensitive to trametinib in our assay. However, we also observed that some samples without RAS mutations are sensitive to the drug. Other than RAS mutations, CBFB-MYH11 fusions are also associated with trametinib sensitivity. We performed a transcriptomic analysis to identify additional biomarkers of sensitivity. We found that samples that are sensitive to trametinib express high levels of several genes associated with myeloid differentiation and that their gene expression profile is similar to monocytes, while resistant samples are more similar to CD34+ progenitor cells. At the cytologic level, in addition to being enriched for the M4 subtype, sensitive samples express higher levels of surface CD14 protein. M4 accounts for 15–25% of AML and is associated with a blast count of at least 20% and a monocytic cell count of at least 20% in the bone marrow with elevated circulating monocytes and 77% of M4 AML showed immunoreactivity for CD14 [26].
CD14 is of particular interest because CD14 is already included in flow cytometry immunophenotyping in AML. It remains an open question whether the high level of these myeloid differentiation markers underlie trametinib sensitivity or is a consequence of other tumor features that confer sensitivity. However, it is possible that elevated expression of CD14 is functionally involved in trametinib sensitivity as this gene has been shown to modulate ERK1/2 phosphorylation in dendritic cells and macrophages [27, 28], and we found that CD14 is part of protein–protein interaction network enriched for components of the MAPK pathway. Consistent with this hypothesis, expression of the CD14 co-receptor TLR4 also correlates positively with trametinib sensitivity (Fig. 3B, rank 197).
Given the heterogeneity in AML, we suggest that combining diverse data including genetics (e.g., RAS) with surface markers (e.g., CD14) as well as other molecular data will be necessary to guide clinical trials for trametinib treatment. Similar precision medicine analyses as the one presented here that rely on functional assays and molecular profiling may also be useful in guiding clinical trials with other drugs that have shown mixed success in AML, such as venetoclax [29], azacytidine [30], or selinexor [31]. It is also of significant interest that monocytic differentiation is identified as a mechanism of resistance to venetoclax-based therapies for AML [32]. Future studies will need to address whether trametinib is active against such venetoclax resistant cells. Overall, our studies provide novel approaches to potential precision therapy for patients with AML.
Supplementary Information
Below is the link to the electronic supplementary material.
Figure S1. A) Association between trametinib sensitivity and clinical or genetic features in the Beat-AML dataset. As for our own cohort, RAS mutations are significantly associated with trametinib sensitivity. In the Beat-AML cohort, CBFB-MYH11 fusions and the M4 subtype are also significantly enriched among trametinib-sensitive samples. P-value: Fisher exact test. B) RAS mutational status, CBFB-MYH11 fusions and M4 subtype classification shown relative to trametinib sensitivity in samples from the top and bottom 20 percentiles by trametinib sensitivity in the Beat-AML cohort. Figure S2. A) Distribution of false discovery rates corresponding to different Pearson correlation coefficients (left y-axis, black line) and number of genes passing cutoff at different Pearson correlation coefficients (right y-axis, blue line) in the HTSC or the Beat-AML datasets by gene expression. B) Venn diagram showing the overlap between gene expression variations that meet all criteria of the correlation analyses when performed with either IC50 or z-score as a measure of trametinib sensitivity. P-values and enrichment factors from hypergeometric distributions. Figure S3. A) The protein-protein interaction network consisting of genes that correlate with trametinib sensitivity involved in MAPK signaling is highly interconnected and implies direct or indirect protein interactions through a single intermediate partner between trametinib sensitivity-correlating genes with gene ontology annotations related to the MAPK pathway (dark red), other trametinib sensitivity-correlating genes (light red) and additional genes of the MAPK pathway (lavender). 1.72-fold enrichment for genes with annotations related to the MAPK/MEK pathway, p-value=0.0038, hypergeometric test. B) The protein-protein interaction network consisting of highly variable genes in the HTSC cohort involved in MAPK signaling (dark green), other highly variable genes (light green) and additional genes of the MAPK pathway (lavender) is comparatively smaller and includes fewer MAPK genes. In both networks, protein-protein interactions correspond to high-confidence physical interactions reported in the HuRI and/or STRING databases. Intermediate interactors with no MAPK function are shown as open circles. 2.84-fold depletion for genes with annotations related to the MAPK/MEK pathway, p-value=7.8X10-6, hypergeometric test. (PDF 1111 KB)
Acknowledgements
S.C. was supported by the Penn Center for Precision Medicine, and Penn Council for Discovery Science. We thank the patients who provided leukemic samples for analysis and the Hematologic Malignancies Translational Center of Excellence who helped support these studies.
Declarations
Funding
M.Q.V., Y.B., and K.W.L. were supported by U01 CA232563; K.W.L. was also supported by R35 GM118048; M.C. is supported by VA Merit Award I01BX004662.
Conflict of interest
The authors declare that they have no conflicts of interest.
Consent for publication
All samples were collected and experiments undertaken with the understanding and written consent of each subject in accordance with IRB approval #703185.
Availability of data and materials
All sequencing data used are publicly available as stated in the Transcriptomic Analysis section.
Code availability
Code is available at https://bitbucket.org/biociphers/aml_trametinib/src/main/.
Ethics approval
The study methodology conformed to the standards set by the Declaration of Helsinki and were approved by the Penn institutional review board (IRB) committee (IRB approval #703185 since 12/2001).
Author contributions
M.Q.V., Y.B., K.W.L., and S.C. conceived of the project; designed, directed, and carried out the transcriptomic analysis; and wrote the manuscript. D.C.S. and Y.L. cultured patient samples and carried out the trametinib screening. A.O., J.M. and M.R. carried out analyses of screening and other patient data. D.R. and M.C. recruited patients and collected samples.
Contributor Information
Yoseph Barash, Email: yosephb@seas.upenn.edu.
Kristen W. Lynch, Email: klync@pennmedicine.upenn.edu
Sara Cherry, Email: cherrys@pennmedicine.upenn.edu.
References
- 1.Dohner H, Wei AH, Lowenberg B. Towards precision medicine for AML. Nat Rev Clin Oncol. 2021;18(9):577–90. 10.1038/s41571-021-00509-w. [DOI] [PubMed] [Google Scholar]
- 2.Tyner JW, et al. Functional genomic landscape of acute myeloid leukaemia. Nature. 2018;562(7728):526–31. 10.1038/s41586-018-0623-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Odogwu L, et al. FDA approval summary: dabrafenib and trametinib for the treatment of metastatic non-small cell lung cancers harboring BRAF V600E mutations. Oncologist. 2018;23(6):740–5. 10.1634/theoncologist.2017-0642. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Robert C, et al. Five-year outcomes with dabrafenib plus trametinib in metastatic melanoma. N Engl J Med. 2019;381(7):626–36. 10.1056/NEJMoa1904059. [DOI] [PubMed] [Google Scholar]
- 5.Ragon BK, et al. Oral MEK 1/2 inhibitor trametinib in combination with AKT inhibitor GSK2141795 in patients with acute myeloid leukemia with RAS mutations: a phase II study. Clin Lymphoma Myeloma Leuk. 2019;19(7):431–40. 10.1016/j.clml.2019.03.015. (e13). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Borthakur G, et al. Activity of the oral mitogen-activated protein kinase kinase inhibitor trametinib in RAS-mutant relapsed or refractory myeloid malignancies. Cancer. 2016;122(12):1871–9. 10.1002/cncr.29986. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Liu X, et al. RAS mutations in acute myeloid leukaemia patients: a review and meta-analysis. Clin Chim Acta. 2019;489:254–60. 10.1016/j.cca.2018.08.040. [DOI] [PubMed] [Google Scholar]
- 8.Malani D, et al. Implementing a functional precision medicine tumor board for acute myeloid leukemia. Cancer Discov. 2022;12(2):388–401. 10.1158/2159-8290.CD-21-0410. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Cheng WY, et al. Transcriptome-based molecular subtypes and differentiation hierarchies improve the classification framework of acute myeloid leukemia. Proc Natl Acad Sci USA. 2022;119(49): e2211429119. 10.1073/pnas.2211429119. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Docking TR, et al. A clinical transcriptome approach to patient stratification and therapy selection in acute myeloid leukemia. Nat Commun. 2021;12(1):2474. 10.1038/s41467-021-22625-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Tazi Y, et al. Unified classification and risk-stratification in acute myeloid leukemia. Nat Commun. 2022;13(1):4622. 10.1038/s41467-022-32103-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Fernandez JM, et al. The BLUEPRINT data analysis portal. Cell Syst. 2016;3(5):491–5. 10.1016/j.cels.2016.10.021. (e5). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Macrae T, et al. RNA-Seq reveals spliceosome and proteasome genes as most consistent transcripts in human cancer cells. PLoS One. 2013;8(9): e72884. 10.1371/journal.pone.0072884. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Ranzani V, et al. The long intergenic noncoding RNA landscape of human lymphocytes highlights the regulation of T cell differentiation by linc-MAF-4. Nat Immunol. 2015;16(3):318–25. 10.1038/ni.3093. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Dobin A, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29(1):15–21. 10.1093/bioinformatics/bts635. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Li H, et al. The sequence alignment/map format and SAMtools. Bioinformatics. 2009;25(16):2078–9. 10.1093/bioinformatics/btp352. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Patro R, et al. Salmon provides fast and bias-aware quantification of transcript expression. Nat Methods. 2017;14(4):417–9. 10.1038/nmeth.4197. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Zhang Y, Parmigiani G, Johnson WE. ComBat-seq: batch effect adjustment for RNA-seq count data. NAR Genom Bioinform. 2020;2(3):lqaa078. 10.1093/nargab/lqaa078. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Vaquero-Garcia J, et al. RNA splicing analysis using heterogeneous and large RNA-seq datasets. Nat Commun. 2023;14(1):1230. 10.1038/s41467-023-36585-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Slaff B, et al. MOCCASIN: a method for correcting for known and unknown confounders in RNA splicing analysis. Nat Commun. 2021;12(1):3353. 10.1038/s41467-021-23608-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Papaemmanuil E, et al. Genomic classification and prognosis in acute myeloid leukemia. N Engl J Med. 2016;374(23):2209–21. 10.1056/NEJMoa1516192. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Anande G, et al. RNA splicing alterations induce a cellular stress response associated with poor prognosis in acute myeloid leukemia. Clin Cancer Res. 2020;26(14):3597–607. 10.1158/1078-0432.CCR-20-0184. [DOI] [PubMed] [Google Scholar]
- 23.Xie ZC, et al. Prognostic alternative splicing regulatory network of splicing events in acute myeloid leukemia patients based on SpliceSeq data from 136 cases. Neoplasma. 2020;67(3):623–35. 10.4149/neo_2020_190917N922. [DOI] [PubMed] [Google Scholar]
- 24.Rivera OD, et al. Alternative splicing redefines landscape of commonly mutated genes in acute myeloid leukemia. Proc Natl Acad Sci USA. 2021. 10.1073/pnas.2014967118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Wang D, et al. A Bayesian model for unsupervised detection of RNA splicing based subtypes in cancers. Nat Commun. 2023;14(1):63. 10.1038/s41467-022-35369-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Alattia L, et al. Value of CD14 immunohistochemical stain in the evaluation of acute monocytic leukemia and myelomonocytic leukemia. Am J Clin Pathol. 2016. 10.1093/ajcp/aqw151.003. [Google Scholar]
- 27.Zanoni I, et al. CD14 controls the LPS-induced endocytosis of toll-like receptor 4. Cell. 2011;147(4):868–80. 10.1016/j.cell.2011.09.051. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Elcombe SE, et al. Dectin-1 regulates IL-10 production via a MSK1/2 and CREB dependent pathway and promotes the induction of regulatory macrophage markers. PLoS One. 2013;8(3): e60086. 10.1371/journal.pone.0060086. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Pollyea DA, et al. Venetoclax for AML: changing the treatment paradigm. Blood Adv. 2019;3(24):4326–35. 10.1182/bloodadvances.2019000937. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Schuh AC, et al. Azacitidine in adult patients with acute myeloid leukemia. Crit Rev Oncol Hematol. 2017;116:159–77. 10.1016/j.critrevonc.2017.05.010. [DOI] [PubMed] [Google Scholar]
- 31.Sweet K, et al. A 2:1 randomized, open-label, phase II study of selinexor vs. physician’s choice in older patients with relapsed or refractory acute myeloid leukemia. Leuk Lymphoma. 2021;62(13):3192–203. 10.1080/10428194.2021.1950706. [DOI] [PubMed] [Google Scholar]
- 32.Pei S, et al. Monocytic subclones confer resistance to venetoclax-based therapy in patients with acute myeloid leukemia. Cancer Discov. 2020;10(4):536–51. 10.1158/2159-8290.CD-19-0710. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1. A) Association between trametinib sensitivity and clinical or genetic features in the Beat-AML dataset. As for our own cohort, RAS mutations are significantly associated with trametinib sensitivity. In the Beat-AML cohort, CBFB-MYH11 fusions and the M4 subtype are also significantly enriched among trametinib-sensitive samples. P-value: Fisher exact test. B) RAS mutational status, CBFB-MYH11 fusions and M4 subtype classification shown relative to trametinib sensitivity in samples from the top and bottom 20 percentiles by trametinib sensitivity in the Beat-AML cohort. Figure S2. A) Distribution of false discovery rates corresponding to different Pearson correlation coefficients (left y-axis, black line) and number of genes passing cutoff at different Pearson correlation coefficients (right y-axis, blue line) in the HTSC or the Beat-AML datasets by gene expression. B) Venn diagram showing the overlap between gene expression variations that meet all criteria of the correlation analyses when performed with either IC50 or z-score as a measure of trametinib sensitivity. P-values and enrichment factors from hypergeometric distributions. Figure S3. A) The protein-protein interaction network consisting of genes that correlate with trametinib sensitivity involved in MAPK signaling is highly interconnected and implies direct or indirect protein interactions through a single intermediate partner between trametinib sensitivity-correlating genes with gene ontology annotations related to the MAPK pathway (dark red), other trametinib sensitivity-correlating genes (light red) and additional genes of the MAPK pathway (lavender). 1.72-fold enrichment for genes with annotations related to the MAPK/MEK pathway, p-value=0.0038, hypergeometric test. B) The protein-protein interaction network consisting of highly variable genes in the HTSC cohort involved in MAPK signaling (dark green), other highly variable genes (light green) and additional genes of the MAPK pathway (lavender) is comparatively smaller and includes fewer MAPK genes. In both networks, protein-protein interactions correspond to high-confidence physical interactions reported in the HuRI and/or STRING databases. Intermediate interactors with no MAPK function are shown as open circles. 2.84-fold depletion for genes with annotations related to the MAPK/MEK pathway, p-value=7.8X10-6, hypergeometric test. (PDF 1111 KB)
Data Availability Statement
All sequencing data used are publicly available as stated in the Transcriptomic Analysis section.