Long-read direct RNA-seq reveals dynamic MM transcriptome made of novel isoforms and post-transcriptional modifications Free

Increasing evidence indicates considerable heterogeneity in multiple myeloma (MM) clinical course and treatment response was driven by the complex and dynamic nature of its transcriptome. Thus a deep understanding of post-transcriptional regulation that was limited by short-read RNA-seq is crucial for overcoming drug resistance and relapse.

We conducted direct RNA-seq for 14 patient-derived xenograft (PDX) samples on a Oxfordnanopore ‘PromethION’ sequencer. We conducted base-calling, read alignment, poly(A) tail and RNA modification calling using ‘Dorado’. Transcriptome reconstruction, quantification, reads and poly(A) tail assignment was performed with ‘IsoQuant’. Reads uniquely assigned to transcripts were further used for expression quantification. Transcripts were reported if >70% of assigned reads had poly(A) tails. RNA modifications including N6-methyadenosine (m6A), 5-methylcytosine (m5C) and pseudouridine (Ψ) were identified using deep-learning models in ‘Dorado’. We obtained on average 18.7 million reads per sample. 94% of reads were detected with poly(A) tails. We identified on average 2.64 million, 30.71 million and 36.43 million bases with m6A with DRACH motif, m5C and Ψ per sample, respectively.

>92% of all sequenced reads were uniquely aligned and their average length was 1,045 bp, with a maximum of 219 kb. A transcriptomic profile was established, containing 65,862 expressed unique transcripts, including 26,535 novel transcripts. 9.3% of all reads were assigned to novel transcripts, suggesting an underestimation of the MM transcriptome. 2,592 novel transcripts were expressed (TPM>1) in >50% samples and included 1,208 and 1,241 transcripts with novel splicing junctions and splicing sites. 143 novel transcripts were derived from genomic regions that were thought to be transcriptionally silent or from independent transcriptional events at known gene loci. 370, 240 and 284 novel transcripts were found to be t(4;14)-, t(11;14)-, and hyperdiploid-specific, respectively, including from previously undescribed gene loci. 1,086 myeloma and subtype-specific transcripts were predicted with high coding potential (p<0.05, complete ORF and 'Fickett'>0.4), including those in IRF4, PRDM1, MYC, BCL9, CCNE1, BAX and another 70 key genes. RNA modifications were reported be involved in MM proliferation and therapy resistance. We found 49,457, 24,257 and 45,283 differential modifications (>25% difference |Log2FC|>1, ≥5 reads) between t(11;14), t(14;16), t(4;14) and hyperdiploid samples, suggesting a dynamic RNA modification landscape at the subtype levels. E.g. A m5C modification at the 3'-UTR of CD79B was found in t(11;14) (33% vs. 0%), suggesting its role in mRNA stabilization and subtype-specific pathogenesis.

Integrative analysis revealed novel features of these transcripts: an isoform of IRF4 (IRF4-V) was found to undergo intron retention (836bps) after exon 7 of the canonical IRF4 transcript(IRF4-201) and terminated before exon 8. IRF4-V was consistently expressed (TPM>1) in all samples but not well covered in short-read RNA-seq. Poly(A) tails of IRF4-V were significantly longer compared to IRF4-201 in all samples (+25nt, p<0.002, Mann Whitney U test), suggesting a longer half-life. RNA modification analysis further revealed 3 m6A DRACH sites (>50%) that were detected in either exon 7 or the retained intron in all samples. Such modifications were significantly enriched in this region compared to the rest of IRF4 (p<0.000001, Fisher's exact test). Meanwhile, neither m5C nor Ψ were enriched. We further identified several m6A readers (e.g. RBM45, r=0.54, Pearson correlation) and splicing factors (e.g. RBM4, r=0.75) whose expression was highly correlated with the IRF4-V usage (dPSI), suggesting their roles in regulating an IRF4-201 to IRF4-V switch. Compared to IRF4-201, IRF4-V contained a truncated IRF association domain (IAD), whose inactivation would affect the transcriptional efficiency of its targets.

This study demonstrates that long-read direct RNA-seq is a powerful tool for uncovering the complex MM (epi-) transcriptome. We identified numerous novel transcripts, including those with altered coding potential, poly(A) abnormalities and specific patterns of RNA modifications in key drivers like IRF4. This integrated transcriptomic and epitranscriptomic analysis offers novel insights into MM biology and highlights potential new avenues for therapeutic intervention.