Full-length direct RNA sequencing reveals extensive remodeling of RNA expression, processing and modification in aging Caenorhabditis elegans

Summary Organismal aging is marked by decline in cellular function and anatomy, ultimately resulting in death. To inform our understanding of the mechanisms underlying this degeneration, we performed standard RNA sequencing and Nanopore direct RNA sequencing over an adult time course in Caenorhabditis elegans. Long reads allowed for identification of hundreds of novel isoforms and age-associated differential isoform accumulation, resulting from alternative splicing and terminal exon choice. Genome-wide analysis reveals a decline in RNA processing fidelity and a rise in inosine and pseudouridine editing events in transcripts from older animals. In this first map of pseudouridine modifications for C. elegans, we find that they largely reside in coding sequences and that the number of genes with this modification increases with age. Collectively, this analysis discovers transcriptomic signatures associated with age and is a valuable resource to understand the many processes that dictate altered gene expression patterns and post-transcriptional regulation in aging.


Introduc=on
Most organisms experience funcWonal decline over Wme due to accumulaWon of cellular damage.The process of aging is known to elicit both coordinated and stochasWc changes in gene expression, which can either promote decline or aid in maintenance of cellular homeostasis.The nematode, Caenorhabdi+s elegans is a useful animal model for understanding transcripWonal changes underlying aging due to its short lifespan and well-annotated genome 1- 3 .C. elegans share a high geneWc homology to humans and, like humans, experience a funcWonal and anatomical decline in aging [3][4][5] .
RNA sequencing (RNA-seq) has been used extensively to characterize the aging transcriptome.Pharmacological suppression of gene expression changes that coincide with aging in C. elegans promotes lifespan, highlighWng the important contribuWon of the transcriptome in maintenance of proper cellular funcWon 6 .Many age-associated gene expression changes are regulated; however, a large proporWon arise from loss of transcripWonal and post-transcripWonal regulaWon, including changes to transcripWonal elongaWon rate, splicing fidelity, and mRNA surveillance [7][8][9] .Loss of other post-transcripWonal regulators, including RNA ediWng enzymes that convert adenosine to inosine in C. elegans, also alter lifespan, suggesWng a role for RNA modificaWons in maintenance of a normal lifespan 10,11 .This hints at potenWal important roles for other RNA modifying enzymes in maintenance of lifespan, warranWng further exploraWon.
Previous transcriptomic studies in aging have been limited by typical RNA-seq methodologies, which rely on transcript assembly from short cDNA fragments.Oxford Nanopore Technologies Direct RNA sequencing (Nanopore DRS) overcomes these limitaWons by reading naWve RNA strands, eliminaWng the requirement for fragmentaWon, reverse transcripWon and PCR amplificaWon, and allowing for sequencing of full-length mRNAs, with no theoreWcal upper limit to read length 12 .This method improves transcript annotaWons, which are difficult to confidently assemble with short reads and has been used to idenWfy novel splice isoforms and 3'UTRs in C. elegans at developmental Wme points 13,14 .Transcript annotaWons with full-length support aid in idenWfying alternaWve splicing events that may be biologically relevant.
Nanopore DRS also facilitates the idenWficaWon of mRNA features that require specialized library preparaWon methods to be detected with standard RNA-seq, including modified RNA nucleoWdes and poly(A) tail lengths 15,16 .ModificaWons like pseudouridine (Ψ) are increasingly being idenWfied in mammalian mRNA, though the difficulty of detecWng these modificaWons has limited our understanding of their funcWonal roles 17,18 .Leveraging Nanopore DRS to idenWfy such modificaWons will enhance our understanding of their impact on mRNA genome-wide.Similarly, poly(A) tails have historically required complex experimental and computaWonal methods to sequence 19,20 .Using these methods, it was determined that short poly(A) tails are a feature of highly expressed, well translated genes in C. elegans and other species examined, suggesWng that poly(A) tails are subject to co-or post-transcripWonal regulaWon 21 .Characterizing poly(A) tail lengths across a variety of experimental condiWons will further strengthen our understanding of their regulaWon and impact on mRNA stability and translaWon.
Despite its many advantages, Nanopore DRS remains limited by its relaWvely low basecalling accuracy and depth of sequencing, though by both of these metrics the methodology is improving 22 .Previous studies in C. elegans demonstrated a requirement for stringent filtering of raw Nanopore DRS data to idenWfy high confidence mRNA isoforms 13,14 .Isoform annotaWons can be further improved using higher accuracy short reads to correct splice juncWon sequences 23 .Using Nanopore DRS in tandem with short read RNA-seq, therefore, is opWmal for performing in depth, comprehensive transcriptomic profiling.
In this study we performed Nanopore DRS in conjuncWon with short read Illumina RNAseq across an adult Wme course in wild-type C. elegans.We generated an extensive dataset, which we used to idenWfy novel transcript isoforms and 3'UTRs with full-length support.With these data, we also characterized signatures of the aging transcriptome, including changes to gene expression, poly(A) tail lengths, and an increase in detected inosine and pseudouridine RNA modificaWons.

Long-read and short-read sequencing over an adult =me course in C. elegans
To be_er understand how the RNA transcriptome changes during aging in an intact animal, we have profiled RNA expression at eight Wme points in wild-type (WT) C. elegans adults, spanning reproducWve (adult days 1-4) and post-reproducWve (adult days 5, 7, 10, 15)   periods.We chose to profile WT animals, as chemical and geneWc manipulaWons to prevent reproducWon alter gene expression and aging pathways 24,25 .By day 15, about 50% of the populaWon is alive, so we ended collecWons then to avoid escalaWon of survivor bias in our gene expression analyses (Figure 1A).Care was also taken to minimize contaminaWon of the adult samples with eggs, progeny, and deceased animals.
RNA samples from three independent collecWons for each Wme point were subjected to Nanopore direct RNA sequencing (Nanopore DRS) to obtain full-length reads and Illumina RNA sequencing (Illumina RNA-seq) (Figure 1B).As expected, more genes were detected with the Illumina RNA-seq method compared to Nanopore DRS (Figure 1C; Table S1; Table S2), as longread sequencing methods are known to have a relaWvely lower sequencing depth 26 .The methods are, however, well correlated when examining normalized gene expression, despite the highly divergent sequencing protocols (Figure 1D; Figure S1).This high correlaWon between Nanopore DRS and Illumina RNA-seq is observed at each Wme point sequenced (Figure 1E).
We next examined gene expression changes at each adult Wme point relaWve to our earliest Wme point, day 1.Many genes are significantly differenWally expressed at each Wme point relaWve to day 1 and the number of differenWally expressed genes increases over the aging Wme course (Figure 1F).Due to limited sequencing depth, fewer differenWally expressed genes are detected with Nanopore DRS, but most of these genes overlap with differenWally expressed genes detected by Illumina RNA-seq (Figure 1G).

Iden=fica=on of hundreds of novel isoforms and 3'UTRs with Nanopore DRS
A key strength of Nanopore DRS is assignment of reads to individual, full-length isoforms, which facilitates idenWficaWon of novel isoforms.To assign reads to annotated isoforms and idenWfy novel isoforms, we first needed to apply stringent filters to our Nanopore DRS reads (Figure 2A).One limitaWon of the Nanopore DRS method is read truncaWon resulWng in reads that are not full-length, which are predominantly 3' biased 27 .To remove these reads, we used a filtering pipeline to subtract reads that do not correspond to annotated transcripWon start sites 14 , then reads lacking poly(A) tails.This filtering pipeline removed ~50% of reads at each Wme point (Figure S2A), but over 19 million reads remained to be used for isoform analysis (Figure S2B; Table S3), far exceeding read numbers obtained in earlier studies using Nanopore DRS in C. elegans at developmental Wme points 13,14 .The filtered reads do not show the 3' bias observed for reads prior to filtering, demonstraWng the uWlity of these filtering steps in removing reads that are not full-length (Figure 2B).
Filtering steps are necessary for transcript annotaWon, however they do introduce significant bias, as some genes show a much higher proporWon of truncated reads than others.
To assess this potenWal source of bias, we compared filtered Nanopore DRS normalized counts to normalized Illumina RNA-seq counts and found that the methods are no longer highly correlated amer filtering (Figure S2C).One example of a gene with an extremely high proporWon of truncated reads is daf-2, an important regulator of aging 28 (Figure S2D).It is not yet clear what causes these truncaWon events during sequencing, but the observed filtering biases reduce the inherent quanWtaWve nature of the Nanopore DRS approach.
Amer filtering for full-length reads we used FLAIR to correct Nanopore DRS read splice juncWons with higher accuracy Illumina RNA-seq reads, then assigned reads to novel and annotated isoforms 23 .We applied stringent filters for isoform annotaWon to eliminate sequences that may represent incompletely spliced transcripts.For this, we required that each isoform represent at least 10% of the total expression for a given gene and set a 20 read minimum for each isoform across all Wme points.Through this method, we detected over 14,000 isoforms, including 813 novel isoforms represenWng 782 genes, that are not present in the WS279 WormBase annotaWon 29 (Figure 2C; Table S4).The vast majority of novel isoforms are detected at all adult Wme points, as low abundance isoforms are likely filtered out by our stringent cutoffs, making it more difficult to detect isoforms only expressed at a single Wme point.Less stringent filters could be used to pull out more novel isoforms from this dataset, if desired.We also used our isoform annotaWons idenWfied with FLAIR to extract 3'UTR coordinates and idenWfied 617 novel 3'UTRs in our data, whose 3' ends do not fall within 10 bp of annotated WS279 WormBase 3'UTR coordinates (Figure 2D; Table S5).
We used stringent filtering criteria and supporWng read requirements to provide confidence that our idenWfied novel isoforms represent detectable gene products.We validated new splice isoforms for three genes using RT-PCR with primers flanking novel alternaWve splicing events.Through this method, we were able to detect PCR products corresponding to annotated as well as novel isoforms for each of the three genes tested, supporWng the validity of our detecWon and annotaWon method (Figure 2E).
We were also curious to see if any isoforms arise from a fusion of two or more genes, as these isoforms would be difficult to detect with short-read sequencing.Gene fusions are usually aberrant and have known roles in promoWng tumorigenesis 30,31 and in some cases arise from transcripWonal readthrough under stress condiWons [32][33][34] .Long reads from Nanopore DRS capture full-length transcripts, so are useful in idenWficaWon of gene fusions, which may span large genomic regions.To screen for this genome-wide, we searched our long-read supported isoforms for introns longer than the median C. elegans gene and an overlap with two or more annotated genes.This analysis revealed two interesWng fusion isoforms of detectable abundance, both of which we were able to validate using RT-PCR with isoform-specific primers and Sanger sequencing of PCR amplicons (Figure 2F).InteresWngly, the fusion isoform spanning the genes lat-1 and enol-1 contains a very long (32kb) intron, which overlaps several other genes.This interesWng isoform appears to be a novel splicing event rather than an artefact of genomic misannotaWon, as we do not detect a PCR product with genomic DNA.
To detect more mulW-gene isoforms, we deployed LongGF, which searches Nanopore DRS data for reads with mulWple genome alignments to idenWfy predicted gene fusions 35 .This tool idenWfied several gene fusion events across all Wme points, including inter-and intrachromosomal gene fusions, which may result from chromosome structural variants or transsplicing, respecWvely (Figure S3A).Expression of these fusion isoforms is varied over aging (Figure S3B).SupporWng this detecWon method, we were able to validate the most highly abundant gene fusion isoform, which maps to genes on chromosomes II (C18H9.6)and IV (clec-173), by RT-PCR and Sanger sequencing of the resulWng amplicon (Figure S3C).AddiWonally, one of the gene fusion isoforms detected in our analysis, eri-6:eri-7, was previously reported to arise from trans-splicing of two separate pre-mRNAs 36 .
DetecWon of these rare gene fusion isoforms, novel splice isoforms, and novel 3'UTRs highlights a disWnct advantage of the Nanopore DRS method.Our use of adult Wme points and our high depth of sequencing build upon previous Nanopore DRS studies in C. elegans and will serve to strengthen genome annotaWons for the community.

Accumula=on of dis=nct transcript isoforms during aging
Isoforms with full-length support help validate and update exisWng gene annotaWons, but the low read numbers and bias introduced from filtering Nanopore DRS reads make it difficult to examine differenWal isoform usage at a large scale with these data.Therefore, we leveraged our Illumina RNA-seq data to perform alternaWve splicing analysis with SUPPA2 37 at each of our Wme points relaWve to day 1, using isoform annotaWons with full-length Nanopore DRS support to define exon boundaries.We detected several alternaWve splicing events at later days relaWve to day 1, including in genes required for stress resistance and proper locomoWon (Figure 3A; Table S6).These alternaWve splicing events include exon skipping, use of mutually exclusive exons, intron retenWon, and alternaWve first exon usage (Figure 3B).They may be important for regulaWon of expression of these genes or protein funcWon over the course of aging.

Global changes in RNA processing and poly(A) tail lengths during aging
To forge a deeper understanding of features of mRNA processing that may change with aging, we examined global trends in young animals compared to old animals.We combined our three earliest Wme points (days 1, 2, and 3) and our three latest Wme points (days 7, 10, and 15) to compare young and old animals, respecWvely.To get a be_er idea of how splicing and transcripWon may be altered in aging, we sought to examine the percent of reads in introns and intergenic regions, as well as the use of unannotated splice juncWons.For these analyses, we used Illumina RNA-seq reads due to their higher accuracy and read numbers.
We reasoned that changes to the percent of total reads mapping to introns may indicate splicing defects, altered splicing efficiency, or a change in transcripWon rate.When we strictly defined introns, requiring they not overlap with any WormBase WS279 annotated exon sequence, we did not see evidence of increased intron inclusion in old animals compared to those in young, as the percent reads mapping to introns is not significantly different (Figure 4A; Table S7).This result is contrary to a previous report where a slight increase in intronic read assignments during aging was observed 8 .This disparity may result from our more stringent definiWon of introns or the use of combined mulWple young and old Wme points.A similar analysis examining the percent of reads assigned to intergenic regions, which can serve as a proxy for inefficient transcripWon terminaWon or improper transcripWon start site selecWon, revealed a modest but significant increase in intergenic region assignment in aging (Figure 4B).
To examine splicing fidelity, we looked at the percent of splice juncWon reads that do not match annotated splice juncWons, which was previously reported to increase in day 15 adults 8 .
Here, we also observe a slight but significant increase in the percent of reads that do not match WormBase WS279 annotated splice juncWons (Figure 4C), suggesWng a modest accumulaWon of aberrant transcripts in old animals compared to young.We observe no bias for change in 3' or 5' mismatches for unannotated splice juncWon reads (Figure S4), which may indicate that this accumulaWon is due to a reducWon in their targeted degradaWon, or that both 5' and 3' splice site recogniWon is affected in later adulthood.
Another feature of mRNA processing that we hypothesized may be altered in aging is polyadenylaWon.Due to limitaWons of typical sequencing methods, poly(A) tail lengths have not yet been examined in this context.Here, we again compared young and old animals, first looking at bulk poly(A) tail length distribuWons from our Nanopore DRS data.We find that poly(A) tails display a modest increase in size in old animals (median=53 nt) compared to young (median=50 nt) (Figure 4D).We note that adult poly(A) tail length distribuWons and median tail lengths closely resemble those of larval stage C. elegans 13,14,21 .
There is a known anWcorrelaWon between transcript abundance and poly(A) tail length in larval C. elegans and other species 21 .To determine if more highly expressed transcripts maintain shorter poly(A) tails in adults, we binned transcripts by abundance and examined median poly(A) tail length distribuWons in these groups.This analysis reveals that despite the overall tail lengthening observed in older animals, highly abundant transcripts maintain shorter poly(A) tails than less abundant RNAs (Figure 4E).This implies that poly(A) tail length is likely subject to the same regulatory processes in aging as in larval stages.
To determine if any specific transcripts undergo large changes to median poly(A) tail length over aging, we compared the change in tail lengths for individual isoforms in old animals compared to young (Figure 4F; Table S8).This analysis revealed that the vast majority (81%) of isoforms undergo very li_le change to poly(A) tail lengths, with a change in median poly(A) of less than 10 nt in old animals compared to young.Only 3.8% of isoforms undergo large changes to median poly(A) tail lengths of more than 20 nt.Most large changes to poly(A) tail lengths can be explained by changes to transcript abundance, in agreement with the previous observaWons.
However, some genes are notable excepWons, with large age-dependent changes to poly(A) tail lengths without corresponding changes in abundance.This subset includes genes in stress response, fa_y acid binding, and heme transport pathways (Figure S5).Together, these results show there are disWncWve global changes to post-transcripWonal RNA processing, including decreased splice site fidelity, increased intergenic RNA expression, and poly(A) tail lengthening, that coincide with aging.However, these important regulatory programs appear to be largely maintained into late adulthood despite the immense changes to physiology and gene expression programs that occur in aging.

Abundance and frequency of inosine edits increase in aging
Finally, we leveraged our Nanopore DRS and Illumina RNA-seq data to idenWfy modified RNA nucleoWdes and determine if their abundance and distribuWon change in aging.We specifically searched for signatures associated with adenosine to inosine (A-to-I) ediWng and pseudouridine (Ψ).A-to-I ediWng can be easily detected with Illumina RNA-seq data, as inosine is reverse transcribed as guanosine during library preparaWon 38 .So, we queried for adenosine to guanosine (A-to-G) mismatches in our Illumina RNA-seq with SAILOR 39 , to take advantage of the method's high depth of sequencing.We combined young (days 1, 2, 3) and old (days 7, 10, 15) Wme points and required high confidence edits present in more than 5% of reads covering each site and in all three biological replicates.
With these stringent filters applied, we idenWfied over 3,800 A-to-G edits in both young and old animals, which we consider putaWve A-to-I ediWng sites (Table S9).The majority of these sites lie in 3'UTRs (Figure 5A), consistent with previous reports in human and C. elegans [40][41][42] .
Though A-to-G sites are detected in a variety of classes of RNA, most sites lie within protein coding genes (Figure 5B).Most genes with inosine have edits in both young and old animals (Figure S6A), but there are more inosine sites detected in old animals (Figure 5A), suggesWng there may be age-dependent regulaWon of inosine ediWng.
Inosine in mRNA coding regions has the capacity to alter translaWon through recoding or translaWonal stalling 43 .We wanted to assess whether detected inosine edits in coding regions could change the amino acid encoded at each posiWon.Inosine is most omen interpreted as guanosine by translaWon machinery 43,44 , so we asked whether subsWtuWon of guanosine at edited sites would alter the encoded amino acid.Many inosines indeed give rise to missense variants (Figure 5C), with the potenWal to alter protein funcWon.Among the predicted missense variants resulWng from inosine ediWng, the majority at both Wme points change the chemical properWes of the encoded amino acid (Figure S6B).Therefore, while less abundant than their 3'UTR counterparts, many inosine edits in coding sequence may impact protein funcWon.
Many genes contain mulWple inosine edits in both old and young animals (Figure 5D), consistent with previous reports in C. elegans, where clusters of inosine edits in 3'UTRs overlapping with repeWWve sequence elements were observed 45 .We also detect a significant enrichment for overlap of inosine edits with repeat sequences at both young and old Wme points, when compared to the abundance of repeWWve elements genome-wide (Figure 5E).Furthermore, over 20% of the inosine edits located in 3'UTRs overlap potenWal microRNA binding sites (Table S9).As inosine has different base-pairing properWes than adenosine 46 , these changes may alter the efficiency of microRNA targeWng and regulaWon of these genes.
Because we observed an overall increase in the number of high confidence inosine edits in aging, we also asked if individual sites are more frequently edited in old animals.To this end, we looked at sites that are detected in young and old animals and compared the percentage of reads with A-to-G edits at each Wme point.Of the 2,659 inosine sites detected at both young and old Wme points, 904 inosine sites have more than a 10% change in ediWng in old compared to young animals.Of these, almost all (98%) increase in ediWng percentage (Figure 5F).This result suggests that there is an age-dependent increase in inosine ediWng genome-wide in C. elegans.This global trend is also apparent for individual genes with mulWple inosine edits (Figure 5G).Overall, our results reveal an increase in the total number of adenosine to inosine conversions in aging, as well as a striking increase in the percentage of edited transcripts at many sites.

Evidence for pseudouridine edi=ng in C. elegans
Pseudouridine is not detectable with Illumina RNA-seq without special methods for library preparaWon 47 , and its presence has not yet been examined genome-wide in C. elegans.
We used a newly developed program, NanoPsu 48 , to detect Ψ in our Nanopore DRS sequencing data, as it does not require specialized sample preparaWon.To ensure we only consider high confidence Ψ sites, we required that sites have a ≥0.90 confidence score and be detected in all three biological replicates.With these criteria, we detect 312 Ψ sites in young and 476 sites in old animals (Table S10).Most of these sites lie within coding regions, where they may potenWally affect translaWon efficiency or fidelity (Figure 6A).Almost all detected Ψ sites are within protein coding genes (Figure 6B), but we also see Ψ sites in ribosomal RNA (rRNA) genes, which we recover to some extent despite the Nanopore DRS selecWon for RNAs with poly(A) tails due to their extremely high abundance.rRNAs are known to be modified with Ψ in many species 49 , so their presence in this analysis strengthens our confidence in the detecWon method.
Most genes with Ψ only have a single edited site, based on our strict criteria.Less than 5% of these genes have mulWple sites and there is no significant moWf enrichment for sequences surrounding Ψ sites.Ψ sites in C. elegans small nuclear RNAs (snRNAs) have been mapped 50 , but these genes lack poly(A) tails, so are not well detected with Nanopore DRS and thus do not emerge in this analysis.Around half of the genes with Ψ in young animals also have Ψ in old animals.Strikingly, however, half of detected Ψ genes are unique to the older Wme points (Figure 6C).Higher detecWon of Ψ in aging cannot be explained by increased expression of these genes with edits only detected in the older samples, as they have varied expression pa_erns over aging (Figure S7A).
While this is the first study to idenWfy Ψ in C. elegans mRNA, the C. elegans genome has numerous predicted pseudouridine synthase genes that are homologous to human, including enzymes that edit mRNA in human 51 (Figure S7B).These genes have dynamic expression pa_erns in aging (Figure S7C).Our detecWon of Ψ genome-wide and evidence for pseudouridine synthase enzymes in the C. elegans genome led us to validate Ψ sites supported by Nanopore DRS.For this, we used a Sanger sequencing-based approach wherein treatment of RNA with sulfite and bisulfite introduces a deleWon at Ψ sites during reverse transcripWon 52 .With this approach, we detected a deleWon at predicted Ψ sites in the coding region of the rnh-2 mRNA and in the rRNA rrn-1.1 upon sulfite/bisulfite treatment (Figure 6D).This validaWon provides strong evidence of Ψ in C. elegans mRNA.

New features of known aging genes
To summarize, we idenWfied new characterisWcs that define the aging transcriptome, including a global increase in inosine and pseudouridine modificaWons and a small, but significant, decline in RNA processing fidelity.We also idenWfied novel splice isoforms and 3'UTRs and instances of age-dependent accumulaWon of specific isoforms.ParWcularly interested in annotaWng new features of genes with known roles in aging, we compared our findings to a list of 1,484 genes previously reported to promote or antagonize longevity 29,53,54 .
We find that many newly uncovered features idenWfied in this study do indeed overlap with known regulators of aging (Figure 7; Table S11).In all, we annotated previously unknown features of 227 genes with published roles in aging.IdenWficaWon of such features may be important for understanding how these important genes are regulated.For example, aps-3, which is involved in intracellular membrane trafficking and antagonizes longevity 29 , has over 100 inosine edits and a pseudouridine site within its 3'UTR.Similarly, some genes that are alternaWvely spliced are known regulators of lifespan, including erm-1 and prdx-2 55,56 , which both use alternaWve first exons in aging relaWve to day 1.We idenWfied and validated a novel gene fusion isoform of enol-1, which promotes longevity in C. elegans.Human orthologs of this gene are implicated in age-related diseases including Alzheimer's disease and cancer 57 .These examples highlight some of the new informaWon learned about important aging regulators through this study.

Discussion
In this study we performed an in-depth invesWgaWon of transcriptome changes that occur over an aging Wme course in wild type C. elegans, which we grew under standard laboratory condiWons.We generated a comprehensive dataset using the well-established Illumina RNA sequencing method in tandem with long-read Oxford Nanopore Technologies direct RNA sequencing.With Nanopore DRS we generated over 19 million full-length reads, allowing for the idenWficaWon of 813 novel isoforms, including novel gene fusion isoforms, and 617 novel 3'UTRs, which will improve exisWng gene annotaWons and serve as a resource for C. elegans researchers.By comparing older to younger populaWons to idenWfy signatures of aging, we observed age-dependent alternaWve splicing and measurable changes to RNA processing, including an increase in unannotated splice juncWons and reads in intergenic regions.Nanopore DRS reads allowed us to assay poly(A) tail lengths genome-wide, where overall we observed adult poly(A) tails similar in length to those in larval stages, with a modest tail lengthening in aging.The most striking changes coinciding with aging were uncovered when we invesWgated RNA modificaWons.We detected an increase in adenosine to inosine ediWng, both in the number of genes with these edits and the ediWng frequency at over one third of sites.Using Nanopore DRS to detect pseudouridine sites genome-wide, we present the first map of Ψ sites in C. elegans mRNA.Like with inosine, we observe an age-dependent increase in Ψ sites.With our comprehensive dataset and analyses, we idenWfied novel isoforms, splicing events, and modificaWons for hundreds of pro-and anW-longevity genes.These findings reveal new features of the aging transcriptome and will serve as a valuable resource for other scienWsts interested in the transcripWonal and post-transcripWonal effectors of aging.

Advantages of tandem long-read and short-read sequencing
In our study, we combined long-read sequencing data with short-read sequencing.This allowed us to answer more biological quesWons and strengthened our novel isoform annotaWons.While short-read RNA sequencing has been applied in C. elegans and other organisms to assess age-dependent changes to gene expression, these methods were limited by read length and their requirement for reverse transcripWon and PCR amplificaWon of libraries.
We generated reads exceeding 10kb and leveraged the ability of this method to sequence RNA directly to invesWgate poly(A) tail lengths and pseudouridine modificaWons genome-wide.
Despite major improvements to Nanopore DRS in recent years, reads generated through this method remain lower in accuracy as compared to short reads 58,59 .CorrecWng splice juncWons of long reads with supporWng short reads, as we did with our Nanopore DRS data, accounts for this and allows for higher confidence isoform annotaWons 23 .Similarly, filtering for full-length reads was necessary for isoform assignment, as we observed a large proporWon of reads that are not full-length.Here, we relied on current genome annotaWons to define full-length reads, as was previously established 14 .The 5' read truncaWons in Nanopore DRS remain among the biggest limitaWons of this technology.Specific genes are more suscepWble to these truncaWons, likely due to difficult to read sequences or structures.This biased our quanWficaWon of full-length isoforms, which led us to use our short-read data to invesWgate alternaWve splicing, further demonstraWng the uWlity of applying both sequencing methods in tandem.

Altera=ons to RNA processing in aging are detectable but not widespread
Our adult Wme course, with collecWon Wme points ranging from young adult to day 15, allowed us to ask how RNA processing changes during aging.We idenWfied several genes with alternaWve splicing in later days relaWve to day 1, including ret-1, which was previously shown to be alternaWvely spliced in aging 8 .While only a small proporWon of expressed genes were found to be alternaWvely spliced later in adulthood, alternaWve isoform usage for individual genes can be highly relevant for longevity.Isoforms generated through alternaWve splicing, alternaWve transcripWon start sites, or alternaWve polyadenylaWon can regulate longevity differenWally.For example, loss of funcWon alleles of daf-2 famously result in an extended lifespan in C. elegans 28 , whereas overexpression of a non-signaling insulin receptor isoform, daf-2b, extends lifespan 60 .This example highlights the contribuWon of alternaWve splicing in maintenance of a normal lifespan and provides jusWficaWon for further invesWgaWon of the alternaWve splicing events idenWfied in this study.While detectable shims in alternaWve splicing were not widespread, individual alternaWvely spliced isoforms can play important cellular roles.
To look at global splicing changes we assayed reads in introns and unannotated splice juncWon reads as in Heintz et al 8 .Our results agree with their finding that unannotated splice juncWon use modestly increases with age.We do not, however, see the same increased intron retenWon they reported in our samples from pooled young and old Wme points.We believe this is due to our stringent definiWon of introns, which removed any annotated intron sequence overlapping with an exon of another transcript.Regardless, the effects on splicing we do observe in aging are modest, so it remains unclear whether these small changes are biologically relevant under normal growth condiWons.Loss of the key splicing factor SFA-1 does not affect lifespan when animals are fed ad libitum but is required for lifespan extension under dietary restricWon 8 .Further, the increase in unannotated splice juncWon reads does not necessarily indicate a decrease in splicing fidelity in aging.AccumulaWon of improperly spliced mRNAs could result from a decline in recogniWon or degradaWon of aberrant transcripts.For example, nonsense mediated decay (NMD) acWvity declines with age in C. elegans 61 , which could explain these observaWons.
We also observed a modest overall change to poly(A) tail lengths in aging.Poly(A) tail lengths are anWcorrelated to transcript abundance and codon opWmality 21 , hinWng at a link between regulaWon of poly(A) tail length and translaWon.We do observe tail lengthening in aging, but fluctuaWons in tail length resemble similar fluctuaWons observed between larval stages in C. elegans and, thus, likely do not represent meaningful deterioraWon of tail length regulaWon 13,14 .Some mRNAs experience larger fluctuaWons to poly(A) tail lengths over our Wme course and, thus, may be interesWng candidates for further invesWgaWon of poly(A) tail length dynamics in an intact animal.

Inosine edi=ng increases in an age-dependent manner
We observed a striking increase in the number and frequency of inosine edits during aging in this study.This is parWcularly interesWng because loss of funcWon alleles of the C. elegans adenosine deaminase acWng on RNA (ADAR) enzymes, which catalyze conversion of adenosine to inosine, alter lifespan 10,11,62 .Specifically, animals lacking ADR-2, which acWvely deaminates RNA, are long-lived, and animals lacking ADR-1, which regulates ediWng efficiency through interacWons with ADR-2 and RNA, are short-lived 10 .Based on the adr-2 loss of funcWon phenotype and the increase in inosine ediWng that correlates with age, it may stand to reason that increased inosine edits in older animals promote degeneraWve aging phenotypes.Inosine ediWng has also been linked to longevity in humans, where single nucleoWde polymorphisms (SNPs) in ADAR enzymes ADARB1 and ADARB2 are associated with extreme old age 11 .While the funcWonal consequences of these SNPs have not been explored, this associaWon provides further evidence for the importance of A-to-I ediWng in aging.Our study goes one step beyond previous work focusing solely on ADAR enzymes by demonstraWng an age-dependent increase in target RNA ediWng genome-wide.With this informaWon, it is also important to consider the effects of A-to-I ediWng on individual target mRNAs.We idenWfied dozens of genes with edits in protein coding sequence that may alter protein funcWon or the speed of translaWon 43,44 .Many more inosine sites were detected within non-coding regions of mRNA, where their funcWon is not as well understood, but may affect secondary structure, microRNA targeWng, or splicing 46 .Many edited nucleoWdes reside within genes that promote or antagonize longevity.It is therefore possible that loss of A-to-I ediWng at specific sites could explain the lifespan phenotypes of adr-1 and adr-2 loss of funcWon mutants.

Detec=on and age-dependent increase of pseudouridine in mRNA
This study is the first to idenWfy and validate pseudouridine edits in C. elegans mRNA.IdenWficaWon of this modificaWon was made possible with Nanopore DRS, highlighWng one of the key advantages of this sequencing method.This exciWng discovery suggests a previously unknown mode of post-transcripWonal regulaWon in this broadly used model organism and paves the way for future studies invesWgaWng the funcWonal roles of Ψ.The age-dependent increase in Ψ and presence of this modificaWon in many aging genes are also intriguing.We idenWfied mulWple putaWve pseudouridine synthase genes in C. elegans based on homology to human, which would be exciWng targets for future studies invesWgaWng the funcWonal roles of Ψ in mRNA and the effect of this modificaWon within specific aging genes.Due to its difficulty to detect and lack of known reader proteins, the role of Ψ in mRNA currently remains elusive.
There is some evidence that Ψ synthase enzymes may recognize structural moWfs 63 and that Ψ stabilizes RNA-RNA interacWons and affects translaWon 64 .Ψ in coding regions was shown to alter mRNA translaWon rate and promote low level synthesis of mulWple pepWdes from a single mRNA in human cells 65 .As the majority of Ψ sites found in our study are in protein coding regions of mRNA, these modificaWons could affect protein producWon for the several hundred genes idenWfied.While the funcWonal consequence of Ψ in mRNA is not well understood, aging C. elegans provides a strong model for future studies on the biological relevance of this modificaWon.

Annota=ng new features of genes that promote and antagonize longevity
While previous studies have explored changes to the transcriptome that coincide with aging, our datasets and analyses provide an in-depth characterizaWon that goes beyond standard differenWal gene expression comparisons.Our approach with eight adult collecWon Wme points and paired Nanopore DRS and Illumina RNA-seq allowed us to ask new quesWons with high resoluWon, leading to idenWficaWon of novel isoforms, 3'UTRs, and modified RNA nucleoWdes.The process of aging is mulWfaceted and complex.Maintenance of normal lifespan is regulated through many pathways important for normal cellular processes, including metabolism, autophagy, apoptosis, cell proliferaWon, and protein synthesis 66 .PerturbaWon of individual genes that control these processes, therefore, can lead to a shortened or extended lifespan.As alternaWve splicing and RNA modificaWons can alter transcript abundance, translaWonal efficiency, and funcWon of the encoded protein, this work serves as a valuable resource for advancing a deeper understanding of aging.

Limita=ons of the study
This study remains limited by our inability to rely fully on Nanopore DRS data for all the analyses conducted.As aforemenWoned, the read truncaWons, lower depth of sequencing, and higher error rates forced us to rely on short reads for some of our analyses.With our approach using two different sequencing methods, however, we were able to overcome many of the challenges posed by Nanopore DRS and make use of this exciWng new technology.Because of the newness of this technology, we were careful to use stringent filtering criteria and conservaWve cutoffs for definiWon of novel isoforms and pseudouridine.Indeed, we were able to validate our computaWonal predicWons experimentally for examined genes.There are likely many more isoforms and sites of interest to be uncovered in our publicly available data.
Recent studies have demonstrated that invesWgaWng aging in whole animals may limit our understanding of how gene expression changes relate to funcWonal decline.New technologies have enabled Wssue-specific or single-cell RNA-seq in aging C. elegans and have shown that the transcriptomes of different Wssues or cell types exhibit disWnct changes in aging 67,68 .Therefore, there are likely trends that we miss due to the relaWvely lower resoluWon of whole animals.We look forward to future iteraWons of long-read sequencing that require lower inputs and allow for adaptaWons of exciWng new technologies.Table S7.Percent of reads mapping to introns or intergenic regions and percent of splice juncWon reads that do not match WS279 WormBase annotaWons in young (days 1, 2, 3) and old (days 7, 10, 15) animals.Table S8.Median, mean, minimum, and maximum poly(A) tail lengths for isoforms with fulllength support.
Table S9.High confidence A-to-G edit site coordinates in BED format, features that overlap with A-to-G sites, and predicted codon changes resulWng from ediWng.Table S10.High confidence pseudouridine edit site coordinates in BED format and overlapping features.Table S11.Annotated aging genes with novel features from this study indicated.genes with new features (novel isoforms, novel 3'UTRs, age-dependent alternaWve splicing, inosine, pseudouridine) idenWfied in this study.An iniWal list of 1,484 aging genes was generated from RNAi or loss of funcWon (LOF) allele phenotypes from WormBase (anW-longevity=extended lifespan, pro-longevity=reduced lifespan) 29 and addiWonal non-overlapping annotated lifespan regulaWng genes from the literature 53,54 .Colored side bars on the heatmaps indicate the annotaWon supporWng each gene.

C. elegans strains and collec=ons
C. elegans N2 animals were grown under standard laboratory condiWons on NGM plates seeded with OP50 E. coli 69 .SynchronizaWon with hypochlorite treatment was performed and embryos were plated and grown at 20ºC.CollecWons were performed at days 1-5, 7, 10, and 15 starWng at day 1 of adulthood, 96 h amer plaWng.

RNA extrac=on and sequencing
IsolaWon of total RNA was performed using a standard Trizol protocol.Total RNA was treated with proteinase K (NEB) to increase purity.12 µg of total RNA were prepared for sequencing using the Direct RNA sequencing kit (cat# SQK-RNA002) and was sequenced on MinION (ONT).
The same RNA was also prepared with the Illumina Stranded TruSeq RNA library prep kit.Prior to library preparaWons, ribosomal RNA was removed using RiboZero Gold (Illumina).cDNA libraries were sequenced on Illumina's NovaSeq 6000 (100 bp paired-end reads).

Illumina RNA-seq mapping and read coun=ng
Illumina RNA sequencing reads were aligned to the Wormbase WS279 reference genome using STAR version 2.7.9a 70 with parameters --runThreadN 24 --readFilesCommand zcat --outSAMtype BAM SortedByCoordinate --outFileNamePrefix.Counting was performed with featureCounts 71 with parameters -p -s 2. Read normalization and differential gene expression analyses were conducted with DESeq2 72 , comparing all time points to day 1.Genes with a Benjamini-Hochberg adjusted P-value ≤0.05 were considered differentially expressed.

Nanopore DRS mapping and filtering
Nanopore RNA sequencing reads were mapped to the Wormbase WS279 reference genome using minimap2 version 2.22-r1101 73 with parameters -a -x splice -uf -k14.To assign unfiltered reads to genes and perform precursory differential expression analysis, featureCounts 71 was used with parameters --fracOverlap 0.8 -primary -R CORE -L.Read normalization and differential gene expression analyses were conducted with DESeq2, comparing all time points to day 1.Genes with a Benjamini-Hochberg adjusted P-value ≤0.05 were considered differentially expressed.To perform read filtering for full-length isoform identification, primary reads were filtered using samtools version 1.15.1 74 with parameters -F0x100 -F0x900.The TSS_filter.py script 14 was used to filter full length reads from primary alignments.Then, filtered reads with a "PASS" flag from Nanopolish poly(A) tail length estimation were used for downstream analyses.

Splice isoform and 3'UTR analysis
Assigning reads to isoforms and splicing analysis was conducted using FLAIR 23 , with WS279 genome annotaWons.Splice juncWons from Illumina RNA-seq were obtained using the juncWons_from_sam.pyscript.Nanopore splice juncWons were corrected with the flair.pycorrect module and collapsed using the flair.pycollapse module with the -s 0.10 parameter, then isoforms were filtered for a minimum of 20 total assigned reads.The predictProducWvity module was used and 3'UTR coordinates were defined from the stop codon to the 3' end of producWve transcripts.3'UTRs whose 3' ends did not fall with 10 bp of WS279 WormBase 3'UTR annotaWons were considered novel.For idenWficaWon of fusion isoforms, all full-length novel isoforms containing intron(s) longer than the median gene length in C. elegans (1,956 bases) and that overlap with exons from two or more annotated WormBase WS279 genes were further examined.

Long gene fusion analysis
Unfiltered Nanopore DRS alignments were used to run LongGF with default parameters 35 .
Predicted gene fusions with ≥20 supporWng reads were kept.

RT-PCR for novel isoforms
Primers were designed to flank novel splicing events.Reverse transcripWon of day 1 RNA was performed with SuperScript™III Reverse Transcriptase (Thermo Fisher) and PCR was performed using GoTaq® G2 DNA Polymerase (Promega).PCR products were visualized by agarose gel electrophoresis.Genomic DNA was isolated using a standard Trizol protocol.

Alterna=ve splicing analysis
A GTF file of all transcripts idenWfied by Nanopore DRS was created and SUPPA2 37 was used to generate an alternaWve splicing events file and perform PSI calculaWons with Illumina RNA-seq data.

Assignment of reads to introns and intergenic regions
Stringent intronic genomic coordinates were derived from the Wormbase WS279 annotaWon by pulling regions that lie within transcripts between annotated exons then removing regions that overlap with any annotated exon.Stringent intergenic genomic coordinates were derived from the WS279 annotaWon by pulling regions that lie between annotated transcripts and removing regions that overlap with any annotated exon.A SAF file was generated for each feature type and Illumina RNA sequencing STAR alignments were assigned using featureCounts with parameters -p -s 2 -T 8 -F SAF --fracOverlap 0.5.

Splice junc=on fidelity analysis
A database of intronic genomic coordinates was created from the Wormbase WS279 annotaWon by pulling regions that lie within transcripts between annotated exons.High confidence collapsed splice juncWon coordinates from STAR mapping (SJ.out.tabfiles) were compared to the intronic database.Uniquely mapping reads that spanned annotated or unannotated splice juncWons were counted and a percentage of unannotated splice juncWons was derived by dividing unannotated splice juncWon counts by the total number of uniquely mapping reads spanning splice juncWons.

Poly(A) tail length analysis
Nanopolish version 0.13.3 75 was used to index fast5 and fastq files and perform poly(A) tail length calling.Filtered reads with "PASS" tags were used for analysis of poly(A) tail lengths with read assignments from FLAIR.

Inosine detec=on genome-wide
The SAILOR pipeline from the FLARE package (h_ps://github.com/YeoLab/FLARE)was used to idenWfy A-to-G edits genome-wide.Sites with a proporWon of edited reads ≥0.05 and a quality score ≥0.75 in three independent biological replicates were used for downstream analyses.
Sites with a probability score ≥0.90 in three independent biological replicates were used for downstream analyses.

Data and code availability
• All datasets have been deposited at Sequence Read Archive (SRA) and will be freely available at the Wme of publicaWon.• All original code has been deposited at Zenodo under: h_ps://zenodo.org/doi/10.5281/zenodo.11672730• Any addiWonal informaWon required to reanalyze data from this study is available upon request from the lead contact, Amy Pasquinelli (apasquinelli@ucsd.edu)

Materials availability
This study did not generate new unique reagents.

Figure 2 .
Figure 2. Iden=fica=on of hundreds of novel isoforms and 3'UTRs.A. Approach for filtering and

Figure 3 .
Figure 3. Alterna=ve transcript isoforms in aging.A.Table showing all genes with significant

Figure 4 .
Figure 4. Altered RNA processing in aging.A. Boxplot (thick horizontal bar represents the mean,

Figure 5 .
Figure 5. Genome-wide detec=on of inosine edi=ng.A. Stacked bar plot showing the number

Figure 6 .
Figure 6.An age-dependent increase in pseudouridine edi=ng.A. Stacked bar plot showing the

Figure 7 .
Figure 7. New features of hundreds of known aging genes.Heatmaps of pro-or anW-longevity