Purpose:

RNA splicing is a fundamental biological process that generates protein diversity from a finite set of genes. Recurrent somatic mutations of splicing factor genes are common in some hematologic cancers but are relatively uncommon in acute myeloid leukemia (AML, < 20% of patients). We examined whether RNA splicing differences exist in AML, even in the absence of splicing factor mutations.

Experimental Design:

We developed a bioinformatics pipeline to study alternative RNA splicing in RNA-sequencing data from large cohorts of patients with AML.

Results:

We have identified recurrent differential alternative splicing between patients with poor and good prognosis. These splicing events occurred even in patients without any discernible splicing factor mutations. Alternative splicing recurrently occurred in genes with specific molecular functions, primarily related to protein translation. Developing tools to predict the functional impact of alternative splicing on the translated protein, we discovered that approximately 45% of the splicing events directly affected highly conserved protein domains. Several splicing factors were themselves misspliced and the splicing of their target transcripts were altered. Studying differential gene expression in the same patients, we identified that alternative splicing of protein translation genes in ELNAdv patients resulted in the induction of an integrated stress response and upregulation of inflammation-related genes. Finally, using machine learning techniques, we identified a splicing signature of four genes which refine the accuracy of existing risk prognosis schemes and validated it in a completely independent cohort.

Conclusions:

Our discoveries therefore identify aberrant alternative splicing as a molecular feature of adverse AML with clinical relevance.

See related commentary by Bowman, p. 3503

Translational Relevance

Utilizing cytogenetic and mutational information, the European LeukemiaNet (ELN) algorithm is the clinical standard for prognosis in AML. However, there is considerable room for improvement, especially in patients classified as intermediate-risk for whom treatment is challenging. The 4-gene splicing signature that we have discovered improves the accuracy of classification, converting the existing three-group risk classification (favorable, intermediate, and adverse risk) into essentially two groups with significantly different overall survival. This will facilitate improved treatment decisions to be made for patients. Our findings also reveal new molecular vulnerabilities that can be potential drug targets for the treatment of AML, a disease that currently has poor overall outcomes (<30% 5-year survival rate). Direct pharmacologic inhibition of splicing factors is potentially challenging clinically due to the toxicity of the drugs. Our data suggest that targeting integrated stress response or pathways stimulated as a consequence of missplicing in leukemic cells could be an alternative approach.

Acute myeloid leukemia (AML) is a hematologic malignancy associated with a poor prognosis and a <30% 5-year survival rate (1). With an incidence rate of 4 per 100,000 adults per year (2) and a 5-fold higher rate in people over the age of 65, AML represents approximately 40% of all new adult-onset leukemias in developed societies (3). AML is characterized by the clonal proliferation of undifferentiated myeloid precursor cells in the bone marrow and impaired hematopoiesis (4). Patients with AML have recurrent somatic driver mutations (5–7) in addition to characteristic cytogenetic and chromosomal abnormalities. These alterations have prognostic significance and are used to classify AML (5). However, not all of these mutations are exclusive to AML, with many also being detected in myelodysplastic syndrome (MDS; ref. 8) as well as in healthy individuals with age-related clonal hematopoiesis (9).

The standard-of-care treatment for AML is intensive induction chemotherapy. However, despite complete remission (CR) rates of >50%, long-term disease-free survival remains poor at <10% and a median overall survival of less than 12 months in patients aged over 60 years (10). In addition, because of significant comorbidities, intensive chemotherapy may not suit older patients (11). Alternate therapies for these individuals may include lower intensity treatments, DNA hypomethylating agents (HMA; ref. 12), or targeted therapies. However, response rates and survival benefits still remain poor (13), highlighting an important need to develop new therapeutic options for the management of AML.

To develop more effective drugs for AML, it is necessary to better understand the molecular aberrations present in leukemic cells. Aberrations in RNA splicing, a fundamental and highly conserved process occurring in >95% of multi-exon human genes (14), are increasingly being described in many cancers. Pan-cancer studies have begun to reveal that tumors have an average of approximately 20% more alternative splicing events than matched healthy tissues (15, 16). Splicing is a cotranscriptional event, orchestrated by cis-acting regulatory elements as well as trans-acting factors of the spliceosomal complex. Dysregulation of the expression of splicing factors (17) and upstream signaling pathways (18), as well as genomic mutations in cis-splice sites (19) have all been reported in cancers. In addition, in hematologic malignancies such as MDS and chronic myelomonocytic leukemia, recurrent somatic mutations in members of the E- and A- splicing complexes, such as SF3B1, U2AF1, SRSF2, and ZRSR2 are detected in >50% of patients (20). The exact mechanisms through which these mutations contribute to the malignancy remain poorly understood. U2AF1 and SF3B1 mutations might alter the 3′ splice site in target transcripts (21, 22), SRSF2 hotspot mutations affect the preferred binding motif on transcripts (23, 24) while ZRSR2 gain-of-function mutations increase intron retention (25). The different mutations have been proposed to affect unique sets of genes (21, 24) although convergence has been proposed at the level of pathways (26) or through the induction of R-loops (27).

Somatic mutations in the splicing machinery are less frequent in AML, however. Analyses of large cohorts of patients have determined the overall frequency of splicing mutations to be <20% (6, 7). However, widespread dysregulation of RNA splicing has been observed even in cancers with low frequencies of splicing factor mutations (15, 16). We therefore examined whether RNA-splicing alterations exist in AML even in the absence of somatic splicing factor mutations, and whether it correlates with disease outcome.

Patient cohorts

Data from two adult AML cohorts were used in the discovery phase of the study: The Cancer Genome Atlas (TCGA)-AML cohort (6) and the Clinseq-AML cohort (28). Data from the Beat-AML cohort (7) were used to validate significance of the splicing signature. Full details are provided as Supplementary Data.

RNA-sequencing data analyses

RNA-sequencing (RNA-seq) data were analyzed for multiple types of alternative splicing using a custom in-house bioinformatics pipeline incorporating available tools, including Mixture of Isoforms (MISO) to determine Percent Spliced In (PSI) values in each sample and rMATS for differential splicing analyses. Differential gene expression analyses were performed using DESeq2. A custom in-house pipeline was developed to identify possible changes in well-annotated protein domains due to differentially spliced events. Full details are provided as Supplementary Data.

Transcript motif analyses

Predictions of differential binding of RNA-binding proteins were made using rMAPS (29). Maximum entropy modeling was done with MaxEntScan (30). Full details are provided as Supplementary Data.

Prognostic model generation

The splicing signature was generated using LASSO Cox Regression with 10-fold cross-validation implemented in glmnet (R package v 2.0-16). The splicing risk score for each patient was calculated from the regression coefficients. Performances of prognostic models were assessed by Harrel C index. Risk contributions and variable importance of all prognostic models were estimated as described previously (31). Full details are provided as Supplementary Data.

Identification of differential alternative splicing related to outcome in patients with AML

To determine whether RNA splicing alterations might be a factor in adverse outcomes in AML, we developed a bioinformatics pipeline to quantify differential alternative splicing in RNA-seq data. We first analyzed AML transcriptomes from The Cancer Genome Atlas (TCGA) (6). To detect whether splicing alterations can exist even in the absence of any somatic mutations in splicing factors, we focused our analyses on patients who did not have any splicing factor mutations using the available somatic mutation data (6). To validate the mutational data, we queried the RNA-seq data to detect the presence of known splicing factor hotspot mutations (20) in transcripts. We confirmed the mutation data in all patients identified to have splicing factor mutations and identified an additional 16 patients with splicing factor hotspot mutations detectable in the RNA-seq data (fourteen with SRSF2 and one each with SF3B1 and ZRSR2). Our findings of TCGA-AML patients with unannotated splicing mutations are consistent with a recent independent report (32).

We stratified patients in the TCGA-AML using the widely used European LeukemiaNet (ELN) prognostic scheme (refs. 1, 4; Fig. 1A) and restricted the analysis to patients who received intensive induction chemotherapy and for whom full clinical data was available (n = 104, Supplementary Fig. S1A; Supplementary Table S1). Performing differential splicing analyses between ELNFav and ELNAdv (Fig. 1B), studying the five different types of alternative splicing events (schematized in Fig. 1C), we identified 1,288 differentially spliced events (at FDR ≤ 0.05) in 910 genes (Fig. 1D; Supplementary Table S2). A majority of the events involved the differential skipping (or retention) of exons preferentially in one set of patients (n = 716, 55.5%, Fig. 1D). Of these, 395 events involved preferential skipping of exons in ELNAdv patients (Supplementary Fig. S1C), with the remaining (n = 321) associated with exon skipping in ELNFav patients. An example of differential exon usage was the skipping of exon 37 of MYO9B in TCGA-ELNFav patients (Fig. 1E). Only reads spanning exon 36–exon 38 were detected in ELNFav patients (representative examples, patients #2914 and #2955; Fig. 1E). In ELNAdv patients (representative patients #2855 and #2817, Fig. 1E); however, there was an increase in the number of reads indicating the inclusion of exon 37 (128 and 41 reads joining exons 36 and 37, and 114 and 61 reads joining exons 37–38, in ELNAdv patients #2855 and #2817, respectively, Fig. 1E; compared with no reads in the ELNFav patients) and a concomitant decrease in reads spanning exon 36 and 38, skipping exon 37 entirely (29 and 24 reads, respectively, compared with 63 and 46 in ELNFav patients #2914 and #2955; Fig. 1E). A related phenomenon, of mutually exclusive exon usage, where adjacent exons are alternately used, contributed to 185 differential events (Fig. 1D). The retention of introns was the next most prevalent class (n = 201, 15.6%, Fig. 1D) in TCGA-ELNAdv patients, as seen in the representative example of the retention of intron 8 in CDK10 (increased intron-specific reads and a decrease in exon–exon reads in ELNAdv patients, Fig. 1F). Additional examples of differential 3′ or 5′ splice site usages are shown in Supplementary Figs. S1D and S1E, respectively.

Figure 1.

Identification of differential alternative splicing in patients with AML. A, Kaplan–Meier survival analysis of ELN stratification in the TCGA AML cohort showing significant survival differences between ELNFav and ELNAdv patients. P value computed using log-rank (Mantel–Cox) test. B, Schematic outline to analyze RNA-seq data for differential alternative splicing. C, Different alternative splicing events detected. Exons affected are represented in gray, while up- and downstream exons are shown in brown and green, respectively. Introns are represented as a black line, and depicted as a solid thick black line when retained. D, Distribution of differentially spliced events identified comparing ELNFav and ELNAdv in the TCGA AML cohort. SE, skipped exons; RI, retained introns; MXE, mutually exclusive exons; A5′SS, alternative 5′ splice sites; A3′SS, alternative 3′ splice sites. E, Sashimi plots of an alternative exon skipping event in the MYO9B gene in the TCGA data, with examples from representative patients shown. Sequencing reads support the skipping of exon 37 (boxed) in two representative ELNFav patients (#2914, #2955, orange tracks), while indicating exon inclusion in ELNAdv patients (#2855, #2817, red tracks). Lines connecting each exon represent splice junctions and numbers on each line indicate the number of supporting RNA-seq reads. F, Sashimi plots of a differential intron retention event in the CDK10 gene in the TCGA data from representative example patients. Differential retention of intron 8 (boxed) is observed in ELNAdv patients (exemplified by representative patients #2014, #2849, red tracks) and not observed in ELNFav patients (as seen in representative patients #2811, #2835, orange tracks). Lines connecting each exon represent splice junctions and numbers on each line indicate the number of supporting RNA-seq reads. G, Kaplan–Meier survival analysis of ELN stratification in the Clinseq AML cohort showing significant survival differences between three ELN subgroups. P value computed using log-rank (Mantel–Cox) test. H, Distribution of differentially spliced events identified comparing ELNFav and ELNAdv in the Clinseq AML cohort. SE, skipped exons; RI, retained introns; MXE, mutually exclusive exons; A5′SS, alternative 5′ splice sites; A3′SS, alternative 3′ splice sites. I, Venn diagram depicting the overlap of differentially spliced genes in both cohorts. Bar plots represent the distribution of alternative splicing events in the shared set of genes. SE, skipped exons; RI, retained introns; MXE, mutually exclusive exons; A5′SS, alternative 5′ splice sites; A3′SS, alternative 3′ splice sites. J, Bubble plots of Ingenuity Pathway Analysis of the differentially spliced genes, in TCGA (red), Clinseq (green), and shared genes (gray). The size of each bubble corresponds to significance of enrichment.

Figure 1.

Identification of differential alternative splicing in patients with AML. A, Kaplan–Meier survival analysis of ELN stratification in the TCGA AML cohort showing significant survival differences between ELNFav and ELNAdv patients. P value computed using log-rank (Mantel–Cox) test. B, Schematic outline to analyze RNA-seq data for differential alternative splicing. C, Different alternative splicing events detected. Exons affected are represented in gray, while up- and downstream exons are shown in brown and green, respectively. Introns are represented as a black line, and depicted as a solid thick black line when retained. D, Distribution of differentially spliced events identified comparing ELNFav and ELNAdv in the TCGA AML cohort. SE, skipped exons; RI, retained introns; MXE, mutually exclusive exons; A5′SS, alternative 5′ splice sites; A3′SS, alternative 3′ splice sites. E, Sashimi plots of an alternative exon skipping event in the MYO9B gene in the TCGA data, with examples from representative patients shown. Sequencing reads support the skipping of exon 37 (boxed) in two representative ELNFav patients (#2914, #2955, orange tracks), while indicating exon inclusion in ELNAdv patients (#2855, #2817, red tracks). Lines connecting each exon represent splice junctions and numbers on each line indicate the number of supporting RNA-seq reads. F, Sashimi plots of a differential intron retention event in the CDK10 gene in the TCGA data from representative example patients. Differential retention of intron 8 (boxed) is observed in ELNAdv patients (exemplified by representative patients #2014, #2849, red tracks) and not observed in ELNFav patients (as seen in representative patients #2811, #2835, orange tracks). Lines connecting each exon represent splice junctions and numbers on each line indicate the number of supporting RNA-seq reads. G, Kaplan–Meier survival analysis of ELN stratification in the Clinseq AML cohort showing significant survival differences between three ELN subgroups. P value computed using log-rank (Mantel–Cox) test. H, Distribution of differentially spliced events identified comparing ELNFav and ELNAdv in the Clinseq AML cohort. SE, skipped exons; RI, retained introns; MXE, mutually exclusive exons; A5′SS, alternative 5′ splice sites; A3′SS, alternative 3′ splice sites. I, Venn diagram depicting the overlap of differentially spliced genes in both cohorts. Bar plots represent the distribution of alternative splicing events in the shared set of genes. SE, skipped exons; RI, retained introns; MXE, mutually exclusive exons; A5′SS, alternative 5′ splice sites; A3′SS, alternative 3′ splice sites. J, Bubble plots of Ingenuity Pathway Analysis of the differentially spliced genes, in TCGA (red), Clinseq (green), and shared genes (gray). The size of each bubble corresponds to significance of enrichment.

Close modal

To validate these findings, we analyzed an independent cohort of patients with AML from the Scandinavian Clinseq study (28). Selecting the patients similarly to the TCGA cohort (Supplementary Fig. S1B; Supplementary Table S1), we performed differential splicing analyses between Clinseq-ELNFav (n = 47) and ELNAdv (n = 75) patients (Fig. 1G). We detected a total of 2,484 alternative splicing events (FDR ≤0.05), affecting 1,566 genes (Fig. 1H; Supplementary Table S3). As in the TCGA cohort, the majority of the events in the Clinseq data were skipped exons (n = 1217, 48.9%, Fig. 1H; Supplementary Fig. S1C). Mutually exclusive exon usage was the next most prevalent (n = 854, 34.4%) followed by intron retention (n = 232, 9.25%, Fig. 1H). Comparing both cohorts, we found differential splicing events occurring in the same direction, that is, enriched either in ELNAdv patients in both cohorts, or in ELNFav patients in both cohorts, in 222 genes (Fig. 1I; Supplementary Table S4). Of these, 93 splicing events (in 78 genes) were identical in both cohorts, which we define as Class A events. A second class, Class B (244 events/173 genes in TCGA, 424 events/182 genes in Clinseq), affected the same gene and with the same directionality but represented different splicing events or occurred at different locations within the gene in the two cohorts. In 19 genes, we observed splicing occurring in opposite directions between the two cohorts.

To determine the molecular impact of this alternative splicing, we performed pathway analyses. Ingenuity Pathway Analysis of the Class A plus Class B genes revealed enrichment for a number of pathways, including those with functions related protein translation or intracellular signaling (Fig. 1J; Supplementary Table S5). Pathways related to protein translation remained enriched when considering Class A or B genes separately (Supplementary Fig. S1F). Orthogonal gene ontology–based analyses also supported these findings, with enrichment for pathways related to protein translation and RNA processing (Supplementary Fig. S1G). Our data reveals recurrent and shared alternative splicing differences between patients with AML with good or poor prognosis in two independent cohorts, converging on specific molecular pathways.

Prediction of the functional consequences of alternative splicing

Analogous to genetic mutations, we expected that while some splicing events would have potentially deleterious effects on subsequent protein translation, others might be silent. To identify deleterious splicing events, we developed a custom bioinformatics pipeline (described in Materials and Methods). Briefly, the chromosomal coordinates of each splicing event were used to generate nucleotide sequences for the spliced and unspliced transcripts. These were then in silico translated and the generated primary sequences were scanned to predict the protein effect (Fig. 2A).

Figure 2.

Analysis of the predicted impact of alternative splicing on protein function. A, Schematic of the analytic pipeline to identify potentially deleterious alternative splicing events. B, Pie chart distribution of protein domain prediction results. Bar plots on the right indicating the distribution of alternative splicing events predicted to lead to a complete loss of protein domains. C, Sashimi plot of a representative protein domain disruption event caused by intron retention in HNRNPH1 gene with examples from representative patients shown. Intron 11 is differentially retained in patients with ELNFav AML (exemplified by the two representative tracks at the top, orange), disrupting the RRM1 domain. Lines connecting flanking exons represent splice junctions and the numbers on each line indicate the number of supporting RNA-seq reads. D, Bubble plot of Ingenuity Pathway Analysis of genes with predicted complete loss of domains. The size of each bubble corresponds to significance of enrichment. E, Nonnegative matrix factorization clustering of patients with AML based on similarity of splicing [percent spliced in (PSI) values] of the shared splicing events, in the TCGA (left) and Clinseq (right) cohorts. Patients were classified as “adverse-like” (red), or “favorable-like” (orange) based on clustering. Oncoprints below denote somatic mutations identified in the patients. P values (Fisher exact test) are shown for TCGA (left) and Clinseq (right), with events with P < 0.05 in bold.

Figure 2.

Analysis of the predicted impact of alternative splicing on protein function. A, Schematic of the analytic pipeline to identify potentially deleterious alternative splicing events. B, Pie chart distribution of protein domain prediction results. Bar plots on the right indicating the distribution of alternative splicing events predicted to lead to a complete loss of protein domains. C, Sashimi plot of a representative protein domain disruption event caused by intron retention in HNRNPH1 gene with examples from representative patients shown. Intron 11 is differentially retained in patients with ELNFav AML (exemplified by the two representative tracks at the top, orange), disrupting the RRM1 domain. Lines connecting flanking exons represent splice junctions and the numbers on each line indicate the number of supporting RNA-seq reads. D, Bubble plot of Ingenuity Pathway Analysis of genes with predicted complete loss of domains. The size of each bubble corresponds to significance of enrichment. E, Nonnegative matrix factorization clustering of patients with AML based on similarity of splicing [percent spliced in (PSI) values] of the shared splicing events, in the TCGA (left) and Clinseq (right) cohorts. Patients were classified as “adverse-like” (red), or “favorable-like” (orange) based on clustering. Oncoprints below denote somatic mutations identified in the patients. P values (Fisher exact test) are shown for TCGA (left) and Clinseq (right), with events with P < 0.05 in bold.

Close modal

Using this methodologic framework, we predict that 26% of the Class A plus Class B events (n = 87, Fig. 2B) cause a complete loss of well-annotated protein domains. The majority of these events involved intron retention events that alter the reading frame (Fig. 2B). An example of this is the retention of intron 8 of the splicing regulator HNRPH1 disrupting the RNA Recognition Motif (RRM) protein domain (Fig. 2C) and an altered transcript predicted to trigger nonsense-mediated decay. An additional 20% of events (n = 67) lead to a partial loss of protein domains (Fig. 2B). The functional consequences of a partial loss of a domain are harder to predict a priori and likely to be protein-specific. Furthermore, of the remaining 181 events (∼54%) of unknown consequence (Fig. 2B), we cannot rule out that some may also affect protein function, through altering protein secondary structure or unannotated domains. Focusing on the events leading to a complete domain loss, pathway analysis revealed that proteins affected by aberrant splicing are still enriched for specific molecular functions, including protein translation (Fig. 2D), which we previously observed (Fig. 1J). Our results suggest that alternative splicing changes leading to predicted protein dysfunction in genes involved in protein translation recurrently occur in patients with AML.

Analysis of the upstream drivers of alternative splicing differences

We next sought to understand the underlying reasons for the splicing differences in these risk groups. We investigated whether any of the somatic driver mutations (in the absence of splicing factor mutations) are correlated with the alternative splicing differences. We clustered patients based on the similarity of their splicing of Class A and B events and assessed the enrichment for somatic mutations within the clusters (Fig. 2E). Apart from NPM1, TP53 and CEBPA mutations, which are intrinsic to the ELN classification algorithm, no other somatic driver mutation showed any statistically significant correlation with the splicing groups (Fig. 2E).

Among the alternatively spliced genes, we observed there were RNA-binding proteins including factors with known roles in RNA splicing (Fig. 3A). These factors form a tightly interconnected network with multiple known protein–protein interactions (Fig. 3B), suggesting that missplicing of these factors could trigger a cascade of splicing alterations in patients with AML. To find evidence for this, we performed motif-scanning analyses (29) of the Class A plus B set of commonly differentially spliced transcripts to determine whether they might be targets for the misspliced splicing factors. We focused on the exon skipping events that formed the majority of the shared differential splicing events across both cohorts (Fig. 1I). Eighty-seven exons were recurrently differentially skipped in ELNAdv patients compared with ELNFav patients in both cohorts, while 94 exons were differentially retained in ELNAdv patients. To determine motif overenrichment, we identified a control set of no-differentially spliced exons (n = 9,986). We then performed motif scanning analyses of the transcript sequences flanking the differentially spliced exons (n = 181), comparing against the nondifferentially spliced set. These analyses indicated that a well-conserved binding motif for HNRNPA1 is significantly overrepresented flanking the 94 exons differentially retained in ELNAdv AML compared with nondifferentially spliced exons (blue dotted line compared with black line, P < 0.0002; Fig. 3C). Our informatics pipeline prediction was that the detected alternative splicing in HNRNPA1, a multifunctional splicing regulator that is known to act as a splicing repressor (33), would produce a nonfunctional protein in ELNAdv patients. Consistent with this, exon inclusion in ELNAdv patients was higher within transcripts where it would normally bind and repress splicing (Fig. 3C). We extended the motif scan analyses to HNRNPC, a splicing factor whose physiologic function is to repress exon inclusion (34) and which is predicted by our informatics analyses to be nonfunctional in ELNFav AML. Consistent with our predictions, we observed a significant overenrichment for HNRNPC-binding motifs flanking the 87 exons that were differentially retained in ELNFav AML patients compared with the background set of nondifferentially spliced exons (dotted red line vs. black line, Fig. 3D).

Figure 3.

Analysis of the upstream drivers of differential alternative splicing in AML. A, List of splicing factors that are commonly differentially spliced across both cohorts. The type of splicing, predicted effect on the protein and FDR are shown. B, Interaction network indicating validated protein–protein interactions (edges) between the differentially spliced, with predicted functional impairment, splicing factor genes (nodes). C, Motif scanning analysis for HNRNPA1 binding sites across a meta-exon generated from the differentially spliced events, with an arrow indicating a peak of significant overenrichment. Motif enrichment scores (left axis) and P values (right axis) are shown. The dashed lines indicate scores of skipped (red) and retained (blue) exons, while the black solid line indicates that of a background score from all nondifferentially spliced exons. The green horizontal line is set at P = 0.05. D, Motif scanning analysis for HNRNPC across a meta-exon generated from all differentially spliced events. Representation similar to C. Arrows indicate peaks of significant overenrichment. E, LOGO analyses of splice donor sites of exons differentially retained (left) or skipped (right) in ELNAdv patients. Analysis is within a 9-base window across the intron–exon junction (3 bases in exon and 6 bases in intron). F, Smoothened density estimates of the position weight matrices (Shapiro score) of the splice donor sites of all differential exon-skipping events. Skipped exons (blue) and background exons (gray) are displayed, illustrating weaker splice sites in the skipped exons.

Figure 3.

Analysis of the upstream drivers of differential alternative splicing in AML. A, List of splicing factors that are commonly differentially spliced across both cohorts. The type of splicing, predicted effect on the protein and FDR are shown. B, Interaction network indicating validated protein–protein interactions (edges) between the differentially spliced, with predicted functional impairment, splicing factor genes (nodes). C, Motif scanning analysis for HNRNPA1 binding sites across a meta-exon generated from the differentially spliced events, with an arrow indicating a peak of significant overenrichment. Motif enrichment scores (left axis) and P values (right axis) are shown. The dashed lines indicate scores of skipped (red) and retained (blue) exons, while the black solid line indicates that of a background score from all nondifferentially spliced exons. The green horizontal line is set at P = 0.05. D, Motif scanning analysis for HNRNPC across a meta-exon generated from all differentially spliced events. Representation similar to C. Arrows indicate peaks of significant overenrichment. E, LOGO analyses of splice donor sites of exons differentially retained (left) or skipped (right) in ELNAdv patients. Analysis is within a 9-base window across the intron–exon junction (3 bases in exon and 6 bases in intron). F, Smoothened density estimates of the position weight matrices (Shapiro score) of the splice donor sites of all differential exon-skipping events. Skipped exons (blue) and background exons (gray) are displayed, illustrating weaker splice sites in the skipped exons.

Close modal

Extending our analyses of the sequence determinants of differential splicing further, we observed increased usage of noncanonical bases at the donor (Fig. 3E) and acceptor (Supplementary Fig. S2A) splice sites adjacent to exons differentially spliced in ELNAdv patients (n = 181) when compared with the background set of nondifferentially spliced exons (n = 9,986). Exons differentially skipped in ELNAdv patients (n = 87) also have weaker splice donor sites compared with background exons (Fig. 3F). To determine whether dysregulation of other RNA-binding factors (in addition to the ones we have already predicted above) might be contributing to differential splicing, we performed a systematic evaluation of 114 RNA-binding factors using a catalog of well-characterized binding motifs (29). We extended the analyses to include the entire set of Class A and B events beyond just skipped exons. We found overenrichment of motifs for PABPC1, a RNA-binding protein recently proposed to have roles in RNA splicing (35), and for RBM46 specifically adjacent to the 94 exons that were differentially retained in ELNAdv patients compared with ELNFav patients in both the TCGA and Clinseq cohorts (blued dotted line, Supplementary Fig. S2B and S2C). Similarly, introns that were preferentially retained in ELNAdv patients were enriched for SRSF3 binding at the 3′ end (Supplementary Fig. S2D). These RNA-binding factors are also known to have protein–protein interactions with other splicing proteins predicted by our analyses to be affected by differential splicing (Supplementary Fig. S2E). Our results suggest that missplicing of splicing factors, together with specific biophysical properties of cis-factors, contribute to the alternative splicing differences we have observed in patients with AML.

Induction of an integrated stress response in patients with ELNAdv AML

Our analyses had indicated that genes related to protein translation were differentially spliced (Fig. 1J), with predicted functional impairment (Fig. 2D) in patients with ELNAdv AML. A cellular consequence of impaired protein translation would be the induction of the Integrated Stress Response (ISR) within cells (36). To find evidence in support of this, we performed differential gene expression analyses between ELNFav and ELNAdv patients (Fig. 4A). A total of 2,219 genes were differentially expressed in the TCGA cohort at FDR < 0.05 (Fig. 4B; Supplementary Table S6) and 1,710 genes in Clinseq (Fig. 4B; Supplementary Table S6). GSEA analyses of the differentially expressed genes clearly indicate a strong upregulation of ISR genes (37) in patients with ELNAdv AML in both cohorts (Fig. 4C). In addition, individual patient analyses revealed a proportional trend between the strength of the induction of ISR gene expression and the extent of missplicing of protein translation genes within the same patient (Fig. 4D). ELNAdv patients, who had higher levels of expression of ISR target genes, tended to have higher levels of missplicing of protein translation genes (Fig. 4D).

Figure 4.

Analysis of the impact of alternative splicing on the transcriptome. A, Schematic outline of the informatics methodology used to identify differentially expressed genes. B, Volcano plots of differentially expressed genes in TCGA (left) and Clinseq (right). Genes highlighted in red have FDR < 0.05 and in orange with FDR < 0.05 and log2 (fold change) > |1|. C, Gene set enrichment analyses (GSEA) showing upregulation of a set of previously reported ATF4-regulated integrated stress response genes (38) in ELNAdv patients from the TCGA (left) and Clinseq (right) cohorts. D, Hierarchical clustering of individual patients (columns) based on the expression of integrated stress response genes (rows) with core enrichment from GSEA for the TCGA and Clinseq cohorts, respectively. Rows were scaled on the basis of expression. A scaled z-score of the PSI values of protein translation genes was calculated in each patient and is represented below. E, Ingenuity Pathway Analysis results of the differentially expressed genes. Venn diagram indicates differentially expressed genes shared by both cohorts. Inflammation-related pathways with associated enrichment values are shown. F, GSEA analyses of a published (39) set of inflammation genes upregulated as a result of decreased protein synthesis. Results show upregulation in ELNAdv patients in both the TCGA (left) and Clinseq (right) cohorts. G, Integrated network analysis of differentially spliced translation genes (green) and differentially expressed inflammation genes (upregulated in red, and downregulated in blue). Experimentally validated protein–protein interactions are depicted as lines, connecting the proteins (nodes). DEG, differentially expressed genes.

Figure 4.

Analysis of the impact of alternative splicing on the transcriptome. A, Schematic outline of the informatics methodology used to identify differentially expressed genes. B, Volcano plots of differentially expressed genes in TCGA (left) and Clinseq (right). Genes highlighted in red have FDR < 0.05 and in orange with FDR < 0.05 and log2 (fold change) > |1|. C, Gene set enrichment analyses (GSEA) showing upregulation of a set of previously reported ATF4-regulated integrated stress response genes (38) in ELNAdv patients from the TCGA (left) and Clinseq (right) cohorts. D, Hierarchical clustering of individual patients (columns) based on the expression of integrated stress response genes (rows) with core enrichment from GSEA for the TCGA and Clinseq cohorts, respectively. Rows were scaled on the basis of expression. A scaled z-score of the PSI values of protein translation genes was calculated in each patient and is represented below. E, Ingenuity Pathway Analysis results of the differentially expressed genes. Venn diagram indicates differentially expressed genes shared by both cohorts. Inflammation-related pathways with associated enrichment values are shown. F, GSEA analyses of a published (39) set of inflammation genes upregulated as a result of decreased protein synthesis. Results show upregulation in ELNAdv patients in both the TCGA (left) and Clinseq (right) cohorts. G, Integrated network analysis of differentially spliced translation genes (green) and differentially expressed inflammation genes (upregulated in red, and downregulated in blue). Experimentally validated protein–protein interactions are depicted as lines, connecting the proteins (nodes). DEG, differentially expressed genes.

Close modal

It has recently been shown that metabolic stresses, including amino acid deprivation that decrease protein synthesis, trigger a proinflammatory response (38). Pathway analyses of the 602 genes that were commonly differentially expressed in the same direction in both cohorts (i.e., either upregulated in ELNAdv both cohorts, or downregulated in both cohorts, Supplementary Table S5) revealed an enrichment for a number of inflammation-related pathways (Fig. 4E). We find an upregulation of these stress-induced inflammatory genes in patients with ELNAdv AML in whom protein translation is impacted due to splicing (Fig. 4F). Network analyses further confirm strong interconnections between the misspliced translation-related genes and the differentially expressed proinflammation genes in ELNAdv AML (Fig. 4G). Our data support a scenario where missplicing of protein translation genes triggers a proinflammatory stress response in ELNAdv patients.

Determining the prognostic relevance of alternative splicing events

Given these findings, we examined whether alternative splicing could serve as a prognostic marker for adverse outcomes in AML. While gene expression and epigenetic studies have been previously linked to AML outcome (39, 40), these analyses would have missed the impact of alternative splicing. Utilizing machine-learning techniques (schematized in Fig. 5A; see Supplementary methods for more details), we identified four genes (MYO9B, GAS5, GIGYF2, RPS9, Fig. 5B) whose differential alternative splicing could stratify AML patients with good and poor prognosis. The differential splicing of these four genes (“splicing signature”) performs comparably with the ELN in both cohorts (Fig. 5C), with similar Harrell C-index (31; Supplementary Fig. S3A) where a C-index of 50% is equivalent to a random assignment and 100% represents a correct ranking of the survival times of all patients.

Figure 5.

Evaluating the prognostic significance of alternatively splicing in AML. A, Machine learning approach used to identify splicing markers. Input patients were randomly classified into two sets, training (80%) and test set (20%). LASSO Cox regression with 10-fold cross-validation was applied to the training set to identify markers. Identified markers were internally validated on the test set. B, The four splicing markers identified to have prognostic significance in both AML cohorts. Regression coefficient and event type are displayed. C, Kaplan–Meier analysis of TCGA-AML (top row) or Clinseq-AML patients (bottom row) stratified either by the ELN or the splicing signature. P values were computed using log-rank (Mantel–Cox) test. D, TCGA-AML (top row) or Clinseq-AML patients (bottom row) classified initially by ELN (left) and reclassified by adding the LSC17 and the Splicing Signature (right). Sankey flow diagrams (middle) illustrate the redistribution of patients, with the widths of the lines proportional to numbers of patients redistributed (number also denoted). P values were computed using log-rank (Mantel–Cox) test. E, Independent validation of the splicing signature in the Beat-AML (7) cohort. Patients were stratified by the splicing signature-based risk score. P values were computed using log-rank (Mantel–Cox) test.

Figure 5.

Evaluating the prognostic significance of alternatively splicing in AML. A, Machine learning approach used to identify splicing markers. Input patients were randomly classified into two sets, training (80%) and test set (20%). LASSO Cox regression with 10-fold cross-validation was applied to the training set to identify markers. Identified markers were internally validated on the test set. B, The four splicing markers identified to have prognostic significance in both AML cohorts. Regression coefficient and event type are displayed. C, Kaplan–Meier analysis of TCGA-AML (top row) or Clinseq-AML patients (bottom row) stratified either by the ELN or the splicing signature. P values were computed using log-rank (Mantel–Cox) test. D, TCGA-AML (top row) or Clinseq-AML patients (bottom row) classified initially by ELN (left) and reclassified by adding the LSC17 and the Splicing Signature (right). Sankey flow diagrams (middle) illustrate the redistribution of patients, with the widths of the lines proportional to numbers of patients redistributed (number also denoted). P values were computed using log-rank (Mantel–Cox) test. E, Independent validation of the splicing signature in the Beat-AML (7) cohort. Patients were stratified by the splicing signature-based risk score. P values were computed using log-rank (Mantel–Cox) test.

Close modal

More accurate stratification and improved prognosis would especially benefit patients with AML classified as intermediate-risk, a group of patients with response and survival rates intermediate to ELNFav and ELNAdv. Accurately identifying ELNInt patients with the most severe risk prognosis would aid in treatment decisions made in the clinic. Equally, ELNInt patients with a predicted favorable prognosis could be treated appropriately. As the mutations and cytogenetics-based ELN, gene expression–based LSC17 signature (40), and our splicing signature represent complementary biological measurements all with the potential to contribute to disease severity, we investigated their combined potential to more accurately classify patients with AML. Addition of the splicing signature to the ELN or LSC17 alone improved the accuracy in both the TCGA (Supplementary Fig. S3B and S3C) and Clinseq cohorts (Supplementary Fig. S3D and S3E), with higher C-indices for the combined signatures (Supplementary Fig. S3F and S3G). Applied together, the combination of the three signatures improved the accuracy of classification of patients with AML, converting the three-group risk classification to essentially two groups with significantly different overall survival in both cohorts (Fig. 5D).

To independently validate the prognostic significance of the splicing signature, we also analyzed data from the BEAT-AML cohort (7). Selecting patients for the TCGA and Clinseq cohorts (Supplementary Fig. S4A), we performed RNA-splicing analyses on the transcriptomic data and calculated the splicing risk score (see Supplementary Methods for details). The prognostic significance of the splicing signature was better than the ELN 2017 (P = 0.0018 vs. P = 0.035, Fig. 5E) within this independent cohort of patients. Addition of the splicing signature to the ELN 2017 (Supplementary Fig. S4B) or the LSC17 (Supplementary Fig. S4C), and combining the signatures (Supplementary Fig. S4D), further improved the accuracy of classification (Supplementary Fig. S4E) consistent with what we had observed in the TCGA and Clinseq cohorts.

Recurrent somatic mutations in RNA splicing factors have been reported in some hematologic malignancies (20). Analyzing AML transcriptomes, we have discovered recurrent alternative RNA splicing differences between ELNFav and ELNAdv patients even in the absence of splicing factor mutations. Many of these alternatively spliced events are predicted to alter protein function, including members of the spliceosomal complex and protein translation genes. Integration with gene expression revealed that ELNAdv patients had an induction of the ISR and a proinflammatory transcriptional program that was proportional to the degree of missplicing of protein translation genes. Furthermore, using machine learning, we identified four alternatively spliced genes that could be used to refine current mutation and transcriptome-based prognostic classification of patients with AML.

The origin of the missplicing that we have detected in patients with AML remains unknown. It is possible that aberrant transcriptional programs initiated by oncogenic driver mutations might dysregulate splicing networks through the misexpression of splicing cofactors. The splicing factors are also subject to a number of regulatory posttranslational modifications. Phosphorylation of splicing factors by kinases of the SRPK and CLK families control their enzymatic activity and subcellular localization (41) and AML cells are sensitive to pharmacologic inhibition of these kinases (42). Many RNA-binding proteins are also methylated by the PRMT family of protein arginine methyltransferases and PRMT inhibition kills leukemic cells (43). Furthermore, splicing alterations due to epigenetic or chromatin changes due to somatic mutations (32, 44) or possibly as a consequence of aging (45) have also been recently reported. It is possible that some or all of these mechanisms could contribute to the splicing alterations we have detected in AML. A cascade of missplicing would then be predicted to arise because of the highly interconnected regulatory networks involving a number of splicing factors and RNA-binding proteins (33, 46).

Decreased protein translation induces the ISR, a conserved pathway which serves promotes cell survival by modulating cellular homeostasis during cellular stress (36). Protein translational stress leads to the efficient translation of the ISR effector ATF4 and upregulation of its target genes (36). Increased ISR and ATF4 activity have been recently shown to be marker of leukemic stem cells in patients with AML (37). Our data indicates aberrant alternative splicing of protein translation genes and an induction of the ISR in patients with AML with poor outcomes. Recently, a second cellular stress response, induction of a proinflammatory transcriptional program, has been identified as a result of decreased protein synthesis (38). Our data are consistent with this, where upregulation of inflammatory genes is seen in ELNAdv patients.

Induction of inflammatory genes and the NFκB pathway have also been reported as a consequence of SF3B1 and SRSF2 mutations in MDS (47). It is possible that the functional consequences of aberrant RNA splicing, through somatic mutations or otherwise, might converge on common downstream consequences. The upregulation of inflammation could induce a leukemic microenvironment that supports the growth of AML clones. AML cells have been recently reported to be dependent on signaling from the proinflammatory cytokine IL1 (48). Furthermore, IL1 signaling suppressed the growth of healthy leukemic cells, thereby promoting leukemogenesis and influencing clonal selection of neoplastic cells (48). While pharmacologic inhibition of splicing factors has been proposed as a targetable vulnerability of leukemic cells (45, 49), the narrow therapeutic window for these drugs due to toxicity poses a potential challenge to using them clinically. Our data suggest that targeting integrated stress response or inflammation-promoting pathways that might be stimulated in leukemic cells as a consequence of missplicing could be an alternative approach.

While cytogenetic and mutational information have become the clinical standard for prognosis in AML, there is still significant heterogeneity that remains unresolved. Assessing additional molecular parameters, including gene expression (40) and DNA methylation (39) have been proposed to improve stratification of patients. However, these analyses would have missed capturing an important molecular feature of AML, aberrant alternative splicing. By complementing existing schema with splicing information, we were able to improve the accuracy of risk stratification, including for ELNInt patients, which should aid in treatment decisions in the clinic. Demographically, the three cohorts (TCGA and BEAT-AML, North American; Clinseq, European) do share some broad commonalities, including a high proportion of patients of white ethnicity (>80% in all) and older patients (median age >55 years in the TCGA and >60 years in the Clinseq and BEAT-AML cohorts; refs. 6, 7, 28). Further assessment of the splicing signature in additional AML cohorts with broader representation of ethnic diversity and younger patients should bring clarity to the wider utility of the splicing signature.

Key points

  • Widespread and recurrent alternative splicing differences exist between patients with AML with good or poor prognosis

  • Missplicing of RNA splicing factors leads to altered splicing of their target transcripts

  • Aberrant splicing of protein translation genes triggers the induction of an integrated stress response and concomitant inflammatory response

  • Alternative RNA splicing information can be used to improve the accuracy of existing prognostic algorithms in AML

  • The addition of the splicing signature accurately classifies patients with ELNInt AML, effectively converting a three-group classification system into a two-group one, facilitating better treatment decisions to be made.

No potential conflicts of interest were disclosed.

Conception and design: G. Anande, A. Unnikrishnan, J.E. Pimanda

Development of methodology: G. Anande, N.P. Deshpande, A. Unnikrishnan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Herold, S. Lehmann

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Anande, N.P. Deshpande, S. Mareschal, A.M.N. Batcha, H.R. Hampton, T. Herold, J.W.H. Wong, A. Unnikrishnan, J.E. Pimanda

Writing, review, and/or revision of the manuscript: N.P. Deshpande, S. Mareschal, T. Herold, S. Lehmann, M.R. Wilkins, J.W.H. Wong, A. Unnikrishnan, J.E. Pimanda

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Anande, N.P. Deshpande, T. Herold

Study supervision: N.P. Deshpande, M.R. Wilkins, J.W.H. Wong, A. Unnikrishnan, J.E. Pimanda

The authors would like to thank Dr. Ling Zhong (Mark Wainwright Analytical Centre, University of New South Wales, New South Wales, Australia) for technical assistance rendered. The authors also thank Dr. Annatina Schnegg (University of New South Wales, New South Wales, Australia) for critical review and discussion of the manuscript. This work was facilitated by infrastructure provided by the NSW Government co-investment in the National Collaborative Research Infrastructure Scheme (NCRIS, Australia). The authors acknowledge the following funding support: G. Anande was supported by a postgraduate scholarship from the University of New South Wales, with additional funding from the Translational Cancer Research Network. A. Unnikrishnan acknowledges funding support from the National Health and Medical Research Council of Australia (APP1163815), Leukemia & Lymphoma Society (USA) and Anthony Rothe Memorial Trust. J.E. Pimanda acknowledges funding from National Health and Medical Research Council of Australia (APP1024364, 1043934, 1102589), Cancer Institute of New South Wales/Translational Cancer Research Network and Anthony Rothe Memorial Trust. T. Herold is supported by the Wilhelm Sander Foundation (2013.086.2), the Physician Scientists Grant (G-509200-004) from the Helmholtz Zentrum München and the German Cancer Consortium (Deutsches Konsortium für Translationale Krebsforschung, Heidelberg, Germany) and a grant from Deutsche Forschungsgemeinschaft (DFG SFB 1243). A.M.N. Batcha was partially funded by the BMBF grant 01ZZ1804B (DIFUTURE).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Dohner
H
,
Estey
E
,
Grimwade
D
,
Amadori
S
,
Appelbaum
FR
,
Buchner
T
, et al
Diagnosis and management of AML in adults: 2017 ELN recommendations from an international expert panel
.
Blood
2017
;
129
:
424
47
.
2.
SEER Cancer Stat Facts: acute myeloid leukemia
.
Bethesda, MD
:
National Cancer Institute
.
Available from:
https://seer.cancer.gov/statfacts/html/amyl.html.
3.
Shallis
RM
,
Wang
R
,
Davidoff
A
,
Ma
X
,
Zeidan
AM
. 
Epidemiology of acute myeloid leukemia: recent progress and enduring challenges
.
Blood Rev
2019
;
36
:
70
87
.
4.
Dohner
H
,
Estey
EH
,
Amadori
S
,
Appelbaum
FR
,
Buchner
T
,
Burnett
AK
, et al
Diagnosis and management of acute myeloid leukemia in adults: recommendations from an international expert panel, on behalf of the European LeukemiaNet
.
Blood
2010
;
115
:
453
74
.
5.
Papaemmanuil
E
,
Gerstung
M
,
Bullinger
L
,
Gaidzik
VI
,
Paschka
P
,
Roberts
ND
, et al
Genomic classification and prognosis in acute myeloid leukemia
.
N Engl J Med
2016
;
374
:
2209
21
.
6.
Cancer Genome Atlas Research Network
,
Ley
TJ
,
Miller
C
,
Ding
L
,
Raphael
BJ
,
Mungall
AJ
, et al
Genomic and epigenomic landscapes of adult de novo acute myeloid leukemia
.
N Engl J Med
2013
;
368
:
2059
74
.
7.
Tyner
JW
,
Tognon
CE
,
Bottomly
D
,
Wilmot
B
,
Kurtz
SE
,
Savage
SL
, et al
Functional genomic landscape of acute myeloid leukaemia
.
Nature
2018
;
562
:
526
31
.
8.
Haferlach
T
,
Nagata
Y
,
Grossmann
V
,
Okuno
Y
,
Bacher
U
,
Nagae
G
, et al
Landscape of genetic lesions in 944 patients with myelodysplastic syndromes
.
Leukemia
2014
;
28
:
241
7
.
9.
Genovese
G
,
Kahler
AK
,
Handsaker
RE
,
Lindberg
J
,
Rose
SA
,
Bakhoum
SF
, et al
Clonal hematopoiesis and blood-cancer risk inferred from blood DNA sequence
.
N Engl J Med
2014
;
371
:
2477
87
.
10.
DiNardo
CD
,
Cortes
JE
. 
Mutations in AML: prognostic and therapeutic implications
.
Hematology Am Soc Hematol Educ Program
2016
;
2016
:
348
55
.
11.
Kantarjian
H
,
Ravandi
F
,
O'Brien
S
,
Cortes
J
,
Faderl
S
,
Garcia-Manero
G
, et al
Intensive chemotherapy does not benefit most older patients (age 70 years or older) with acute myeloid leukemia
.
Blood
2010
;
116
:
4422
9
.
12.
Dombret
H
,
Seymour
JF
,
Butrym
A
,
Wierzbowska
A
,
Selleslag
D
,
Jang
JH
, et al
International phase 3 study of azacitidine vs. conventional care regimens in older patients with newly diagnosed AML with >30% blasts
.
Blood
2015
;
126
:
291
9
.
13.
Estey
E
,
Levine
RL
,
Lowenberg
B
. 
Current challenges in clinical development of "targeted therapies": the case of acute myeloid leukemia
.
Blood
2015
;
125
:
2461
6
.
14.
Wang
ET
,
Sandberg
R
,
Luo
S
,
Khrebtukova
I
,
Zhang
L
,
Mayr
C
, et al
Alternative isoform regulation in human tissue transcriptomes
.
Nature
2008
;
456
:
470
6
.
15.
Kahles
A
,
Lehmann
KV
,
Toussaint
NC
,
Huser
M
,
Stark
SG
,
Sachsenberg
T
, et al
Comprehensive analysis of alternative splicing across tumors from 8,705 patients
.
Cancer Cell
2018
;
34
:
211
24
.
16.
Climente-Gonzalez
H
,
Porta-Pardo
E
,
Godzik
A
,
Eyras
E
. 
The functional impact of alternative splicing in cancer
.
Cell Rep
2017
;
20
:
2215
26
.
17.
Luo
C
,
Cheng
Y
,
Liu
Y
,
Chen
L
,
Liu
L
,
Wei
N
, et al
SRSF2 regulates alternative splicing to drive hepatocellular carcinoma development
.
Cancer Res
2017
;
77
:
1168
78
.
18.
Vu
NT
,
Park
MA
,
Shultz
JC
,
Goehe
RW
,
Hoeferlin
LA
,
Shultz
MD
, et al
hnRNP U enhances caspase-9 splicing and is modulated by AKT-dependent phosphorylation of hnRNP L
.
J Biol Chem
2013
;
288
:
8575
84
.
19.
Jayasinghe
RG
,
Cao
S
,
Gao
Q
,
Wendl
MC
,
Vo
NS
,
Reynolds
SM
, et al
Systematic analysis of splice-site-creating mutations in cancer
.
Cell Rep
2018
;
23
,
270
81
.
20.
Yoshida
K
,
Sanada
M
,
Shiraishi
Y
,
Nowak
D
,
Nagata
Y
,
Yamamoto
R
, et al
Frequent pathway mutations of splicing machinery in myelodysplasia
.
Nature
2011
;
478
:
64
9
.
21.
Ilagan
JO
,
Ramakrishnan
A
,
Hayes
B
,
Murphy
ME
,
Zebari
AS
,
Bradley
P
, et al
U2AF1 mutations alter splice site recognition in hematological malignancies
.
Genome Res
2015
;
25
:
14
26
.
22.
Dolatshad
H
,
Pellagatti
A
,
Fernandez-Mercado
M
,
Yip
BH
,
Malcovati
L
,
Attwood
M
, et al
Disruption of SF3B1 results in deregulated expression and splicing of key genes and pathways in myelodysplastic syndrome hematopoietic stem and progenitor cells
.
Leukemia
2015
;
29
:
1092
103
.
23.
Zhang
J
,
Lieu
YK
,
Ali
AM
,
Penson
A
,
Reggio
KS
,
Rabadan
R
, et al
Disease-associated mutation in SRSF2 misregulates splicing by altering RNA-binding affinities
.
Proc Natl Acad Sci U S A
2015
;
112
,
E4726
34
.
24.
Kim
E
,
Ilagan
JO
,
Liang
Y
,
Daubner
GM
,
Lee
SC
,
Ramakrishnan
A
, et al
SRSF2 mutations contribute to myelodysplasia by mutant-specific effects on exon recognition
.
Cancer Cell
2015
;
27
:
617
30
.
25.
Madan
V
,
Kanojia
D
,
Li
J
,
Okamoto
R
,
Sato-Otsubo
A
,
Kohlmann
A
, et al
Aberrant splicing of U12-type introns is the hallmark of ZRSR2 mutant myelodysplastic syndrome
.
Nat Commun
2015
;
6
:
6042
.
26.
Qiu
J
,
Zhou
B
,
Thol
F
,
Zhou
Y
,
Chen
L
,
Shao
C
, et al
Distinct splicing signatures affect converged pathways in myelodysplastic syndrome patients carrying mutations in different splicing regulators
.
RNA
2016
;
22
:
1535
49
.
27.
Chen
L
,
Chen
JY
,
Huang
YJ
,
Gu
Y
,
Qiu
J
,
Qian
H
, et al
The augmented R-Loop is a unifying mechanism for myelodysplastic syndromes induced by high-risk splicing factor mutations
.
Mol Cell
2018
;
69
:
412
425
.
28.
Wang
M
,
Lindberg
J
,
Klevebring
D
,
Nilsson
C
,
Mer
AS
,
Rantalainen
M
, et al
Validation of risk stratification models in acute myeloid leukemia using sequencing-based molecular profiling
.
Leukemia
2017
;
31
:
2029
2036
.
29.
Park
JW
,
Jung
S
,
Rouchka
EC
,
Tseng
YT
,
Xing
Y
. 
rMAPS: RNA map analysis and plotting server for alternative exon regulation
.
Nucleic Acids Res
2016
;
44
:
W333
8
.
30.
Yeo
G
,
Burge
CB
. 
Maximum entropy modeling of short sequence motifs with applications to RNA splicing signals
.
J Comput Biol
2004
;
11
:
377
94
.
31.
Gerstung
M
,
Pellagatti
A
,
Malcovati
L
,
Giagounidis
A
,
Porta
MG
,
Jadersten
M
, et al
Combining gene mutation with gene expression data improves outcome prediction in myelodysplastic syndromes
.
Nat Commun
2015
;
6
:
5901
.
32.
Yoshimi
A
,
Lin
KT
,
Wiseman
DH
,
Rahman
MA
,
Pastore
A
,
Wang
B
, et al
Coordinated alterations in RNA splicing and epigenetic regulation drive leukaemogenesis
.
Nature
2019
;
574
:
273
7
.
33.
Huelga
SC
,
Vu
AQ
,
Arnold
JD
,
Liang
TY
,
Liu
PP
,
Yan
BY
, et al
Integrative genome-wide analysis reveals cooperative regulation of alternative splicing by hnRNP proteins
.
Cell Rep
2012
;
1
:
167
78
.
34.
Konig
J
,
Zarnack
K
,
Rot
G
,
Curk
T
,
Kayikci
M
,
Zupan
B
, et al
iCLIP reveals the function of hnRNP particles in splicing at individual nucleotide resolution
.
Nat Struct Mol Biol
2010
;
17
:
909
15
.
35.
Horan
L
,
Yasuhara
JC
,
Kohlstaedt
LA
,
Rio
DC
. 
Biochemical identification of new proteins involved in splicing repression at the Drosophila P-element exonic splicing silencer
.
Genes Dev
2015
;
29
:
2298
311
.
36.
Pakos-Zebrucka
K
,
Koryga
I
,
Mnich
K
,
Ljujic
M
,
Samali
A
,
Gorman
AM
. 
The integrated stress response
.
EMBO Rep
2016
;
17
:
1374
95
.
37.
van Galen
P
,
Mbong
N
,
Kreso
A
,
Schoof
EM
,
Wagenblast
E
,
Ng
SWK
, et al
Integrated stress response activity marks stem cells in normal hematopoiesis and leukemia
.
Cell Rep
2018
;
25
:
1109
1117
.
38.
Gameiro
PA
,
Struhl
K
. 
Nutrient deprivation elicits a transcriptional and translational inflammatory response coupled to decreased protein synthesis
.
Cell Rep
2018
;
24
:
1415
24
.
39.
Figueroa
ME
,
Lugthart
S
,
Li
Y
,
Erpelinck-Verschueren
C
,
Deng
X
,
Christos
PJ
, et al
DNA methylation signatures identify biologically distinct subtypes in acute myeloid leukemia
.
Cancer Cell
2010
;
17
:
13
27
.
40.
Ng
SW
,
Mitchell
A
,
Kennedy
JA
,
Chen
WC
,
McLeod
J
,
Ibrahimova
N
, et al
A 17-gene stemness score for rapid determination of risk in acute leukaemia
.
Nature
2016
;
540
:
433
7
.
41.
Stamm
S
. 
Regulation of alternative splicing by reversible protein phosphorylation
.
J Biol Chem
2008
;
283
:
1223
7
.
42.
Tzelepis
K
,
De Braekeleer
E
,
Aspris
D
,
Barbieri
I
,
Vijayabaskar
MS
,
Liu
WH
, et al
SRPK1 maintains acute myeloid leukemia through effects on isoform usage of epigenetic regulators including BRD4
.
Nat Commun
2018
;
9
:
5378
.
43.
Fong
JY
,
Pignata
L
,
Goy
PA
,
Kawabata
KC
,
Lee
SC
,
Koh
CM
, et al
Therapeutic targeting of RNA splicing catalysis through inhibition of protein arginine methylation
.
Cancer Cell
2019
;
36
:
194
209
.
44.
Inoue
D
,
Chew
GL
,
Liu
B
,
Michel
BC
,
Pangallo
J
,
D'Avino
AR
, et al
Spliceosomal disruption of the non-canonical BAF complex in cancer
.
Nature
2019
;
574
:
432
6
.
45.
Crews
LA
,
Balaian
L
,
Delos Santos
NP
,
Leu
HS
,
Court
AC
,
Lazzari
E
, et al
RNA splicing modulation selectively impairs leukemia stem cell maintenance in secondary human AML
.
Cell Stem Cell
2016
;
19
:
599
612
.
46.
Liang
Y
,
Tebaldi
T
,
Rejeski
K
,
Joshi
P
,
Stefani
G
,
Taylor
A
, et al
SRSF2 mutations drive oncogenesis by activating a global program of aberrant alternative splicing in hematopoietic cells
.
Leukemia
2018
;
32
:
2659
71
.
47.
Lee
SC
,
North
K
,
Kim
E
,
Jang
E
,
Obeng
E
,
Lu
SX
, et al
Synthetic lethal and convergent biological effects of cancer-associated spliceosomal gene mutations
.
Cancer Cell
2018
;
34
:
225
41
.
48.
Carey
A
,
Edwards
DK
,
Eide
CA
,
Newell
L
,
Traer
E
, et al
Identification of interleukin-1 by functional screening as a key mediator of cellular expansion and disease progression in acute myeloid leukemia
.
Cell Rep
2017
;
18
:
3204
18
.
49.
Lee
SC
,
Dvinge
H
,
Kim
E
,
Cho
H
,
Micol
JB
,
Chung
YR
, et al
Modulation of splicing catalysis for therapeutic targeting of leukemia with mutations in genes encoding spliceosomal proteins
.
Nat Med
2016
;
22
:
672
8
.