Mitochondrial sequencing identifies long noncoding RNA features that promote binding to PNPase
Andrew D Taylor,Quincy A Hathaway,Amina Kunovac,Mark V Pinti,Mackenzie S Newman,Chris C Cook,Evan R Cramer,Sarah A Starcovic,Michael T Winters,Emily S Westemeier-Rice,Garrett K Fink,Andrya J Durr,Saira Rizwan,Danielle L Shepherd,Aaron R Robart,Ivan Martinez,John M Hollander
DOI: https://doi.org/10.1152/ajpcell.00648.2023
2024-08-01
Abstract:Extranuclear localization of long noncoding RNAs (lncRNAs) is poorly understood. Based on machine learning evaluations, we propose a lncRNA-mitochondrial interaction pathway where polynucleotide phosphorylase (PNPase), through domains that provide specificity for primary sequence and secondary structure, binds nuclear-encoded lncRNAs to facilitate mitochondrial import. Using FVB/NJ mouse and human cardiac tissues, RNA from isolated subcellular compartments (cytoplasmic and mitochondrial) and cross-linked immunoprecipitate (CLIP) with PNPase within the mitochondrion were sequenced on the Illumina HiSeq and MiSeq, respectively. lncRNA sequence and structure were evaluated through supervised [classification and regression trees (CART) and support vector machines (SVM)] machine learning algorithms. In HL-1 cells, quantitative PCR of PNPase CLIP knockout mutants (KH and S1) was performed. In vitro fluorescence assays assessed PNPase RNA binding capacity and verified with PNPase CLIP. One hundred twelve (mouse) and 1,548 (human) lncRNAs were identified in the mitochondrion with Malat1 being the most abundant. Most noncoding RNAs binding PNPase were lncRNAs, including Malat1. lncRNA fragments bound to PNPase compared against randomly generated sequences of similar length showed stratification with SVM and CART algorithms. The lncRNAs bound to PNPase were used to create a criterion for binding, with experimental validation revealing increased binding affinity of RNA designed to bind PNPase compared to control RNA. The binding of lncRNAs to PNPase was decreased through the knockout of RNA binding domains KH and S1. In conclusion, sequence and secondary structural features identified by machine learning enhance the likelihood of nuclear-encoded lncRNAs binding to PNPase and undergoing import into the mitochondrion.NEW & NOTEWORTHY Long noncoding RNAs (lncRNAs) are relatively novel RNAs with increasingly prominent roles in regulating genetic expression, mainly in the nucleus but more recently in regions such as the mitochondrion. This study explores how lncRNAs interact with polynucleotide phosphorylase (PNPase), a protein that regulates RNA import into the mitochondrion. Machine learning identified several RNA structural features that improved lncRNA binding to PNPase, which may be useful in targeting RNA therapeutics to the mitochondrion.