Improving AML Classification Using Splicing Signatures

In this issue of Clinical Cancer Research, Anande and colleagues (1) show that altered splicing is prevalent in acute myeloid leukemia (AML) and correlates with an elevated integrated stress response. Stratification of patients according to their splicing signature, independent of splicing factor mutational status, improved prognostic classification, which could better inform treatment decisions. Splicing is an essential RNA processing step to generate mature messenger RNAs that is seamlessly coordinated by hundreds of proteins and five small nuclear RNAs. The assortment of RNA species produced by alternative splicing increases proteome diversity and permits plasticity in cellular responses to internal and external cues. Since the discovery of recurrent somatic mutations in spliceosomal components in myelodysplastic syndromes (MDS) and AML less than 10 years ago, there has been a surge of studies examining how splicing dysfunction could contribute to disease pathophysiology (2). The prevalence of these mutations in myeloid malignancies led to the hypothesis that production of defective splice isoforms is a common disease driver across numerous MDS/AML subtypes. Subsequent studies by many groups revealed that mutations in specific splicing factors correlate with different prognosis and result in distinct alternative splicing signatures, suggesting the mechanism could be more factor-specific and complex. Upon deeper examination, although different transcripts and splicing events are altered, the current data suggest there is some convergence at the pathway level for most cells carrying splicing factor mutations. But what about the role of splicing dysfunction in the absence of these mutations?

Although common in MDS, mutations in spliceosomal components are found in less than 20% of AML cases (3). Does this mean that alterations to splicing are not significantly contributing to AML pathophysiology? Anande and colleagues addressed this question by examining the splicing patterns in human AML samples. To identify whether specific splicing alterations were associated with clinical outcome independent of splicing factor mutations, they examined RNA splicing in samples lacking detectable mutations in any of the common malignancy-associated splicing factors. Specifically, they examined differential splice isoform usage between patients with AML with favorable versus adverse outcomes as defined by European Leukemia Network (ELN) parameters. Differential splicing was defined in two cohorts: The Cancer Genome Atlas (TCGA) and Scandanavian ClinSeq. There was high concordance in the differentially spliced transcripts between the two cohorts, increasing the likelihood that the changes might be relevant to AML pathology. In particular, factors involved in RNA processing and protein translation were significantly enriched. The observed alternative isoforms often resulted in loss of domains, suggesting the aberrant splicing would lead to diminished function of the affected factor. In agreement with this observation, there was an increase in the expression of integrated stress response genes consistent with defective protein translation (Fig. 1). Such metabolic stress can trigger a proinflammatory response, which was also observed in their analysis. These findings are in-line with recent studies demonstrating that elevated inflammatory signaling is a common outcome in myeloid malignancies when splicing is defective (2).

If these AML samples lacked mutations in spliceosomal components, then what could explain the splicing alterations? There was not a significant correlation with any other known AML driver mutations, suggesting a novel mechanism for the observed alternative splicing. To identify potential RNA-binding proteins that could underlie the splicing alterations, Anande and colleagues performed motif enrichment analysis comparing the regulatory sequences surrounding alternatively spliced elements with commonly spliced events in the same dataset. Intersection of the candidates from this motif analysis with RNA-processing factors that were themselves alternatively spliced revealed a short list of novel RNA-binding proteins that could be involved in promoting or suppressing AML. These include the splicing repressors hnRNPA1, which is predicted to be nonfunctional in adverse AML cases, and hnRNPC, which is predicted to have diminished function in favorable cases. Along with splicing, RNA-binding proteins linked to all stages of mRNA biogenesis are implicated in AML suggesting RNA processing as a focal node for targeting this disease (4).

Therapeutic interventions are heavily guided by established stratification approaches which are based on cellular and molecular diagnostics, such as genomics and gene expression. One of the best established is the ELN classification primarily guided by cytogenetics and mutational analysis. It is known that splicing factor mutations hold independent prognostic value in MDS, thus can the newly defined AML splicing signature predict outcome? To address this question, Anande and colleagues first refined the splicing list using LASSO to define a robust splicing signature comprised of four genes that showed a strong correlation with patient outcome. Stratification of the TCGA training dataset or the independent ClinSeq dataset with this splicing signature alone largely recapitulated the prognostic classification based on ELN definitions. Moreover, categorizing of the BEAT-AML cohort according to the splicing signature improved patient stratification of an ELN intermediate outcome group. Combining stratification strategies driven by distinct biological processes can improve their predictive power. As such, Anande and colleagues further refined the well-established ELN classification with the gene expression–driven LSC17 and their novel splicing signature and demonstrated improved accuracy in categorizing patients with AML (Fig. 1). This new stratification could better inform decisions on patient treatment.

These results illustrate that splicing aberrations are prevalent in AML and not restricted to cells carrying splicing factor mutations. What still remains unclear is what drives these changes? Their data suggest that alternative splice isoform usage of RNA-binding proteins can promote promiscuous splicing of numerous transcripts, but what initiates the alternative splicing of the RNA-binding proteins? Splicing patterns are highly cell type specific, thus it is possible that the novel splice isoforms in the RNA-binding proteins represent natural variants in the altered AML cellular state. Splicing decisions are also regulated by transcriptional elongation, a process often altered in AML. Several new epigenetic drugs targeting elongation such as BET and DOT1L inhibitors are currently in clinical and preclinical testing for leukemias (5). In the future, it will be intriguing to determine whether there is a connection between the efficacy of these inhibitors and splicing alterations. Beyond the specific splicing alterations, the current data suggest that defects in splicing trigger integrated stress and proinflammatory responses, which are both linked to AML pathophysiology. Pharmacologic targeting of splicing in leukemia is tricky as it is an essential process for all cells, which could result in a small therapeutic window or high morbidity. These new findings suggest that targeting the downstream pathways could provide a more tenable treatment for AML.

Overall, this work provides novel insight into splicing dysfunction in AML, potential pathways impacted, and the value of this finding in patient stratification. Numerous new ideas and questions arise from this analysis of primary clinical data that can better inform future preclinical testing on the function and targetability of splicing in AML.

No potential conflicts of interest were disclosed.

T.V. Bowman was funded by American Cancer Society RSG-129527-DDC, the Evans MDS Foundation, DOD BM180109, and NIH 1R56DK121738-01.