RNA Splicing Factor Mutations Drive Aberrant Canonical and Cryptic Circular RNA Biogenesis in Leukemia

RNA splicing factors (SFs) are recurrently mutated in liquid cancers. Mutations on SFs such as SF3B1, SRSF2, U2AF1, ZRSR2, and the U1 snRNA promote leukemogenesis by driving aberrant pre-mRNA splicing in critical pathways such as DNA damage response (DDR) and cell cycle regulation (CCR). Circular RNAs (circRNAs) are covalently closed RNAs that form via backsplicing ( Fig A). As a novel RNA species, circRNAs are now known to play diverse functional roles in cancer biology. Because backsplicing is a type of alternative splicing, we hypothesized that SF mutations also influence circRNA biogenesis in leukemia. To date, such causal links have not been clearly established.

We first performed circRNA-seq on a panel of lymphoid and myeloid cell lines modeling the most common SF mutations ( SF3B1 K700E, SRSF2 P95H, U2AF1 S34F/Q157R, ZRSR2 KO, and U1 snRNA 3A>C). We observed that SF mutations tend to globally upregulate circRNAs with each mutant SF driving its own mutually exclusive set of circRNAs that we have termed SF mutation-associated circRNAs (SF-circRNAs). Each set of SF-circRNAs differs from each other along several features such as average lengths and selection of backsplicing exons. Additionally, while SF-circRNAs were generally mutation-specific, the majority of SF-circRNAs arose from commonly shared genes enriched for the DDR and CCR pathways. These results support a model in which SF mutations drive divergent splicing events convergent on a few pathways as a common mechanism in leukemias.

Because SF3B1 is the most recurrently mutated SF in cancers, we concentrated on studying SF3B1 mutation-associated circRNAs (SF3B1-circRNAs). During linear RNA splicing, hotspot mutations on SF3B1 typically promote selection of cryptic 3' splice site (SS). Cryptic 3'SS are weak, inauthentic 3'SS proximal to the canonical site that can be activated for splicing by mutant SF3B1. Reasoning that cryptic 3'SS selection also occurs during backsplicing, we developed a pipeline called CrypticCIRC to call for these events. Through CrypticCIRC, we discovered the existence of cryptic circRNAs that account for nearly 1 in 10 of all SF3B1-circRNAs. Approximately 25% of cryptic circRNAs emanated from de novo 3'SS previously unannotated in the genome. Furthermore, we found that cryptic circRNAs were especially enriched on genes involved in the DDR and CCR pathways, furthering the notion that these pathways are preferentially targeted by SF mutations. Finally, by comparing with linear RNA splicing data, we also observed that some cryptic 3'SS may undergo either aberrant linear splicing or aberrant backsplicing but not both, indicating mutual exclusivity in cryptic splice site selection between linear and backsplicing.

Finally, to mechanistically dissect SF-circRNA biology, we focused our study on one candidate circRNA, circZEB1. The parental ZEB1 gene, a master transcription and DDR factor, undergoes both aberrant canonical and cryptic backsplicing at exon 2 with exon 4 to give rise to two circZEB1 isoforms ( Fig B). In our full SF mutant panel, we found that the canonical and cryptic isoforms were consistently upregulated only in SF3B1 mutant cell lines (Nalm-6, K-562, HEK293T, HG-3, and MEC-1); We further confirmed circZeb1 upregulation in mouse models (CD19-Cre Sf3b1 K700E) where the canonical isoform is evolutionarily conserved. These results confirm circZEB1 as a true bona fide SF3B1-circRNA. Focusing on the function of this circRNA, we found that CasRx-mediated knockdown of canonical circZEB1 resulted in increased ZEB1 protein, but not mRNA, expression while overexpression of circZEB1 decreased ZEB1 protein levels. These results show that circZEB1 translationally regulates its parental gene. Finally, consistent with ZEB1's role in genome stability, we also observed a global reduction in γH2AX levels concomitant with circZEB1 ablation, highlighting DDR as a common pathway for SF-circRNAs ( Fig B).

Altogether, we demonstrate that SF mutations drive widespread aberrant circRNA biogenesis mainly on DDR and CCR genes. We provide evidence for the existence of cryptic circRNAs as a novel class of circRNAs. We show that circZEB1 promotes DNA damage via regulating ZEB1 translation. Our results collectively highlight circRNAs as an additional regulatory modality alongside linear RNAs through which SF mutations exert function, supporting a unifying mechanism for the pan-cancer impact of SF mutations in leukemia.