Ageing-associated changes in transcriptional elongation influence longevity

Physiological homeostasis becomes compromised during ageing, as a result of impairment of cellular processes, including transcription and RNA splicing1–4. However, the molecular mechanisms leading to the loss of transcriptional fidelity are so far elusive, as are ways of preventing it. Here we profiled and analysed genome-wide, ageing-related changes in transcriptional processes across different organisms: nematodes, fruitflies, mice, rats and humans. The average transcriptional elongation speed (RNA polymerase II speed) increased with age in all five species. Along with these changes in elongation speed, we observed changes in splicing, including a reduction of unspliced transcripts and the formation of more circular RNAs. Two lifespan-extending interventions, dietary restriction and lowered insulin–IGF signalling, both reversed most of these ageing-related changes. Genetic variants in RNA polymerase II that reduced its speed in worms5 and flies6 increased their lifespan. Similarly, reducing the speed of RNA polymerase II by overexpressing histone components, to counter age-associated changes in nucleosome positioning, also extended lifespan in flies and the division potential of human cells. Our findings uncover fundamental molecular mechanisms underlying animal ageing and lifespan-extending interventions, and point to possible preventive measures.

Ageing impairs a wide range of cellular processes, many of which affect the quality and concentration of proteins. Among these, transcription is particularly important, because it is a main regulator of protein levels [7][8][9] . Transcriptional elongation is critical for proper mRNA synthesis, owing to the co-transcriptional nature of pre-mRNA processing steps such as splicing, editing and 3′ end formation 2,10 . Indeed, dysregulation of transcriptional elongation results in the formation of erroneous transcripts and can lead to numerous diseases 1,3 . During ageing, animal transcriptomes undergo extensive remodelling, with large-scale changes in the expression of transcripts involved in signalling, DNA damage responses, protein homeostasis, immune responses and stem cell plasticity 11 . Furthermore, some studies uncovered an age-related increase in variability and errors in gene expression [12][13][14] . Such previous work has provided insights into how the transcriptome adapts to, and is affected by, ageing-associated stress. However, it is not known if, or to what extent, the transcription process itself affects or is affected by ageing.
In this study, we used high-throughput transcriptome profiling to investigate how the kinetics of transcription are affected by ageing, how such changes affect mRNA biosynthesis and to elucidate the role of these changes in age-related loss of function at the organismal level. We show an increase in RNA polymerase II (Pol II) elongation speed with age across five metazoan species, a speed reduction under lifespan-extending conditions and a causal contribution of Pol II elongation speed to lifespan. We thus reveal an association of fine-tuning Pol II speed with genome-wide changes in transcript structure and chromatin organization. gradient of read coverage along an intron, it is possible to determine the elongation speeds of Pol II at individual introns (Fig. 1a,b). Note that this measure is only weakly associated with the expression level of the transcript 20 (Supplementary Table 2). To monitor how the kinetics of transcription changes during ageing, we quantified the distribution of intronic reads resulting from RNA-seq in five animal species-the worm Caenorhabditis elegans, the fruitfly Drosophila melanogaster, the mouse Mus musculus, the rat Rattus norvegicus 21 and the human Homo sapiens-at different adult ages (see Methods; Supplementary Table 1), and using diverse mammalian tissues (the brain, liver, kidney and whole blood), fly brains and whole worms. Human samples originated from whole blood (healthy donors, 21-70 years of age) and from two primary human cell lines (fetal lung fibroblasts (IMR90) and human umbilical vein endothelial cells (HUVECs)) driven into replicative senescence. After filtering, we obtained between 518 and 6,969 introns that passed quality criteria for reliable Pol II speed quantification (see Methods). These different numbers of usable introns mostly result from interspecies variation in intron size and number, and to some extent from variation in sequencing depth. To rate the robustness of Pol II speed changes across biological replicates, we clustered samples based on their 'speed signatures', that is, on the detected elongation speeds across all introns that could be commonly quantified across each set of experiments. We observed largely consistent co-clustering of samples from the same age across species, whereas young and old samples mostly separated from each other (Extended Data Fig. 1). This suggests that age-related speed changes were consistent across biological replicates and reliably quantifiable in independent measurements. We observed an increase in average Pol II elongation speed with age in all five species and all tissue types examined ( Fig. 1c and Extended Data Fig. 2a). Changes in Pol II speed did not correlate with either the length of the intron or with its position within the gene (Supplementary Table 2). The observed increase in Pol II elongation speed was even more pronounced after selecting introns with consistent speed changes across all replicates (that is, always up or down with age; Extended Data Fig. 3). This result is non-trivial because our analysis also revealed introns with a consistent reduction in Pol II speed. To confirm our findings with an orthogonal assay, we monitored transcription kinetics in IMR90 cells using 4-thiouridine (4sU) labelling of nascent RNA. After inhibiting transcription with 5,6-dichloro-1-β-d-ribofuranosyl benzimidazole (DRB), we conducted a pulse-chase-like experiment quantifying 4sU-labelled transcripts at four time points after transcription release (that is, at 0, 15, 30 and 45 min). This enabled us to quantify Pol II progression into gene bodies (see Methods for details) and confirmed our results based on intronic slopes using proliferating (young) and senescent (old) IMR90 cells. Measurements of Pol II speed from the 4sU-based assay showed significant correlation with those from the slope-based assay (Fig. 1d), with Pol II speed increasing on average in both approaches ( Fig. 1e and Extended Data Fig. 4). Note that, although many individual genes showed a decrease in elongation speed with ageing in both assays, the majority exhibited increased speed.
To assess whether known lifespan-extending interventions-inhibition of insulin-IGF signalling and dietary restriction-affected Pol II speeds, we sequenced RNA from insulin-IGF signalling mutants, using daf-2-mutant worms at day 14 and fly brains from dilp2-3,5 mutants at day 30 and day 50, as well as the hypothalamus from aged wild-type and Irs1-null mice. We also sequenced RNA from the kidney and liver of dietary-restricted and ad libitum-fed mice. In all comparisons, except Irs1-null mice and the livers from 26-month-old dietary-restricted mice, lifespan-extending interventions resulted in a significant reduction of Pol II speed. Pol II elongation speeds thus increased with age across a wide range of animal species and tissues, and this increase was, in most cases, reverted under lifespan-extending conditions (Fig. 1).
Although Pol II speed changed consistently with age across replicates (Extended Data Fig. 1), we did not observe specific classes of genes to be affected across models. To determine whether genes with Article particular functions were more strongly affected by age-related Pol II speed changes, we performed gene set enrichment analysis on the 200 genes with the highest increase in Pol II speed during ageing in worms, fly brains, mouse kidneys and livers, and rat livers. Only very generic functional classes, such as metabolic activity, were consistently enriched across three or more species (Extended Data Fig. 5). Thus, no specific cellular process appeared to be consistently affected across species and tissues. Next, we examined age-associated gene expression changes of transcriptional elongation regulators. We observed that some regulators (for example, PAF1 and THOC1) were consistently downregulated across species during ageing (Extended Data Fig. 6), which was also confirmed using gene set enrichment analysis (Extended Data Fig. 7). These expression changes potentially represent a compensatory cellular response to a detrimental increase in transcriptional elongation speeds.

Reducing Pol II elongation speed extends lifespan
To determine whether changes in Pol II speed are causally involved in the ageing process, we used genetically modified worm and fly strains carrying point mutations in a main Pol II subunit that reduce its elongation speed (the ama-1(m322) mutant in C. elegans 5 and the RpII215 C4 mutant in D. melanogaster) 6 . We sequenced total RNA from wild-type and 'slow' Pol II-mutant worms (whole animal at day 14) or fly heads (at day 10 and day 50). Measurements of elongation speeds confirmed the expected reduction of average Pol II speeds in both C. elegans ama-1(m322) and D. melanogaster RpII215 C4 (Fig. 2a). To assess whether Pol II speed and its associated maintenance of transcriptional fidelity also affected ageing of the whole organism, we measured survival of these animals. Slowing down Pol II increased lifespan in both worms and fruitflies (median lifespan increase of approximately 20% in C. elegans and approximately 10% in D. melanogaster; Fig. 2b and Extended Data Fig. 8a). CRISPR-Cas9-engineered reversal of the Pol II mutations in worms restored lifespan essentially to wild-type levels (Extended Data Fig. 8b). Furthermore, mutant worms displayed higher pharyngeal pumping rates at older age than wild-type worms, suggesting that health span was also extended by slowing down Pol II elongation speed (Extended Data Fig. 8c).

Changes of transcript structure and sequence
Optimal elongation rates are required for fidelity of alternative splicing for some exons 22,23 . Slow elongation favours weak splice sites that lead to exon inclusion, whereas these exons are skipped if elongation is faster 10,24,25 . Faster elongation rates can also promote intron retention leading to the degradation of transcripts via nonsense-mediated decay 26 and possibly contributing to disease phenotypes 27 . Therefore, we next quantified changes in splicing. The first measure that we used was splicing efficiency, which is the fraction of spliced reads from all reads aligning to a given splice site 28 . In most datasets, from total and nascent RNA-seq, we observed an increase of the spliced exon junctions relative to unspliced junctions during ageing, and a decrease of the percent spliced junctions under lifespan-extending conditions (Fig. 3a). Consistent with earlier work 29 , we observed more spliced transcripts under conditions of increased Pol II speed, that is, greater splicing efficiency. For co-transcriptional splicing to occur, Pol II first needs to transcribe all parts relevant to the splicing reaction (that is, 5′ donor, branch point and 3′ acceptor), which are located at the opposite ends of an intron 30,31 . Our data suggest that accelerated transcription shortens the interval in which splicing choices are made, thus shortening the time between nascent RNA synthesis and intron removal. Accelerated transcription and splicing carries the risk of increasing the frequency of erroneous splicing events, which has been associated with advanced age and shortened lifespan 4,32-34 . It is non-trivial to deduce whether a specific splice isoform is the product of erroneous splicing or created in response to a specific signal. Simply checking whether an observed isoform is annotated in some database can be problematic for multiple reasons. For instance, most databases have been created on the basis of data from young animals or embryonic tissue. Thus, a detected isoform that only may be functionally relevant in old animals will not be reported in such databases. Moreover, an annotated isoform might be the result of erroneous splicing if its expression is normally suppressed at a particular age or cellular context. We therefore based our analysis on the notion that extremely rare isoforms (rare with respect to all other isoforms of the same gene in the same sample) are more likely erroneous than frequent isoforms 35 which performs de novo quantification of exon-exon junctions based on split-mapped RNA-seq reads. Owing to its ability to identify alternatively excised intron clusters, Leafcutter is particularly suitable to study rare exon-exon junctions 38 . We defined rare splicing events as exon-exon junctions supported by 0.7% or less of the total number of reads in a given intron cluster, and the gene-specific fraction of rare clusters was computed as the number of rare exon-exon junctions divided by the total number of detected exon-exon junctions in that gene. We observed that such rare exon-exon junctions often resulted from exon skipping or from the usage of cryptic splice sites (Extended Data Fig. 9). The average fraction of rare splicing events increased during ageing in the fly and worm, and this effect was reverted under most lifespan-extending conditions (Fig. 3b). However, we did not observe a consistent age-associated increase of the fraction of rare splice variants across all species, which may at least in part be due to the more complex organization of splice regulation in mammalian cells.
Another potential indicator of transcriptional noise is the increased formation of circular RNAs (circRNAs) 39,40 , that is, of back-spliced transcripts with covalently linked 3′ and 5′ ends 41 . Increased speed of Pol II has previously been associated with increased abundance of circRNA 42 .
Thus, we quantified the fraction of circRNAs as the number of back-spliced junctions normalized by the sum of back-spliced fragments and linearly spliced fragments. We observed either increased or unchanged average circRNA fractions during ageing, whereas reducing the speed of Pol II also reduced the formation of circRNAs (Fig. 3c). This suggests that faster Pol II elongation correlates with a general increase in the level of circRNAs. Nevertheless, our data do not provide evidence that increased levels of circRNA directly result from increased Pol II speed, despite it being a consequence of the overall reduced quality in RNA production.
Increased speeds of Pol II can lead to more transcriptional errors because the proofreading capacity of Pol II is challenged 12 . To assess the potential effect of accelerated elongation on transcript quality beyond splicing, we measured the number of mismatches in aligned reads for each gene. For this, we normalized mismatch occurrence to individual gene expression levels and excluded mismatches that were probably due to genomic variation or other artefacts (see Methods for details). We observed that the average fraction of mismatches increased with age, but decreased under most lifespan-extending treatments (Fig. 3d). Consistent with previous findings 12 Table 3 for details on the number of genes and back-spliced junctions used per comparison. In all panels, the dashed line at 0 indicates no change as a visual aid. In all panels, the silhouettes of the organs were created using BioRender (https://biorender.com), and the silhouettes of species are from PhyloPic (https://phylopic.org).
exhibited reduced numbers of mismatches compared with wild-type control levels in three out of four comparisons.

Changes in chromatin structure associated with Pol II elongation
Subsequently, we explored alterations in chromatin structure as a possible cause of the age-associated changes in Pol II speeds. Nucleosome positioning along DNA is known to affect both Pol II elongation and splicing 19,[43][44][45] . Furthermore, aged eukaryotic cells display reduced nucleosomal density in chromatin and 'fuzzier' core nucleosome positioning 46,47 . Thus, age-associated changes in chromatin structure could contribute to the changes in Pol II speed and splicing efficiency that we observed. To test this, we performed micrococcal nuclease (MNase) digestion of chromatin from early (proliferating) and late-passage (senescent) human IMR90 cells, followed by approximately 400 million paired-end read sequencing of mononucleosomal DNA (MNase-seq). Following mapping, we examined nucleosome occupancy. In senescent cells, introns were less densely populated with nucleosomes than proliferating cells 48 (Fig. 4a). In addition, we quantified peak 'sharpness', reflecting the precision of nucleosome positioning in a given MNase-seq dataset (see Methods), as well as the distances between consecutive nucleosomal summits as a measure of the spacing regularity 48,49 . Principal component analysis of the resulting signatures indicated consistent changes of nucleosome sharpness and distances upon entry into senescence as the samples clearly separated by condition (Fig. 4b,c).
Both measures were significantly, but moderately, altered in senescent cells (Fig. 4d,e): average sharpness was slightly decreased (along both exons and introns) and average internucleosomal distances slightly increased in introns. In conclusion, the transition from a proliferating cell state to replicative senescence was associated with small but significant changes in chromatin structure, involving nucleosome density and positioning changes that were previously shown to affect Pol II elongation 43,47,50 .

Overexpression of histone proteins decreases Pol II elongation speed and extends fly lifespan
The organization of nucleosomes is severely influenced by histone availability 45,46 . For example, histone H3 depletion reduces nucleosomal density and renders chromatin more accessible to MNase digestion 51 . Such global loss of histones constitutes a hallmark of ageing and senescence 52 . Consistent with this, our senescent IMR90 cells and HUVECs carry significantly reduced protein levels of histone H3 (Fig. 5a). Conversely, elevated levels of histones promote lifespan extension in yeast 46 , C. elegans 53 and D. melanogaster 54 . To assess whether Pol II elongation speed and senescence entry in human cells are causally affected by changes in nucleosomal density, we generated IMR90 cell populations homogeneously overexpressing GFP-tagged H3 or H4 in an inducible manner ( Fig. 5b and Extended Data Fig. 10a,b).
Overexpression of either histone resulted in significant reduction of Pol II speed, confirming the causal connection between chromatin structure and transcriptional elongation (Fig. 5c). Reduction of Pol II speed was accompanied by markedly reduced senescence-associated β-galactosidase staining in H3-overexpressing or H4-overexpressing cells compared with both control (GFP only) and uninduced cells (Fig. 5d). Moreover, both H3-overexpressing and H4-overexpressing cells did not display induction of p21 or depletion of HMGB1, which are both hallmarks of senescence entry, compared with control IMR90 cells ( Fig. 5e and Extended Data Fig. 10c). Finally, MTT assays showed that viability and proliferation were improved in H3-overexpressing and, to a lesser extent, in H4-overexpressing cells compared with control cells Article (Fig. 5f). Together, these results suggest that H3 or H4 overexpression decelerates Pol II and compensates for the ageing-induced core histone loss 44,50 to restrict senescence entry. The average speed reduction following H4 overexpression was significantly larger than that obtained upon H3 overexpression, yet H4-overexpressing near-senescent IMR90 cells only marginally outperformed control cells in MTT assays (Fig. 5f). This raises the possibility of excessive reduction in Pol II speed negatively affecting aspects of cell function 55 . To address the role of nucleosome density in organismal lifespan, we used UAS-Histone 3 (ref. 54) to overexpress His3, specifically in Drosophila glial cells using Repo-Gal4. H3 overexpression led to significantly increased numbers of mononucleosomes in aged (60 days of age) compared with wild-type fly heads (Fig. 5g), thus possibly compensating for age-associated loss of histone proteins. Furthermore, H3 overexpression in glial cells increased fruitfly lifespan (Fig. 5h). These in vivo results are consistent with our in vitro data from IMR90 cells, demonstrating that H3 overexpression partially reverts the ageing effects on chromatin density and promotes longevity in flies. As this was linked to a reversal in Pol II elongation speed, our findings, together with earlier findings in yeast 46,51 , C. elegans 53 and D. melanogaster 54 , demonstrate how the structure of the chromatin fibre probably modulates Pol II elongation speed and lifespan.

Discussion
We found a consistent increase in the average intronic Pol II elongation speed with age across four animal models, two human cell lines and human blood, and could revert this trend by using lifespan-extending treatments. We also documented ageing-related changes in splicing and transcript quality, such as the elevated formation of circRNAs and increased numbers of mismatches with genome sequences, which probably contribute to age-associated phenotypes. Furthermore, we observed a consistent increase in the ratios of spliced to unspliced transcripts (splicing efficiency) with age across species (Fig. 3a), which has been reported to be a result of increased elongation speed 29 . However, we cannot exclude the possibility that this increase resulted from changes in RNA half-lives. Although average speed changes were predominantly significant, they remained small in absolute terms. This is expected, as drastic, genome-wide changes of RNA biosynthesis would quickly be detrimental for cellular functions and would probably lead to early death. Instead, what we monitored here is a gradual reduction of cellular fitness characteristic for normal ageing. Critically, we were able to increase lifespan in two species by decelerating Pol II. Thus, despite being small in magnitude and stochastically emerging in tissues or cell populations, these effects are clearly relevant for organismal lifespan.
Genes exhibiting accelerated Pol II elongation were not enriched for specific cellular processes, indicating that speed increase is probably not a deterministically cell-regulated response, but rather a spontaneous age-associated defect. Yet, the genes affected were not completely random, as we observed consistent changes across replicates for a subset of introns. Thus, there must be location-specific factors influencing which genomic regions are more prone to increases in Pol II speed and which are not. This observation is consistent with earlier findings and our data, indicating that chromatin structure may causally contribute to age-associated Pol II speed increase. Although we still lack a complete understanding of the molecular events driving Pol II speed increase, our findings indicate that ageing-associated changes in chromatin structure have an important role.
Our work establishes Pol II elongation speed as an important contributor to molecular and physiological traits with implications beyond ageing. Misregulation of transcriptional elongation reduces cellular and organismal fitness and may therefore contribute to disease phenotypes 53,56,57 . Together, the data presented here reveal a molecular mechanism contributing to ageing and serve as a means for assessing the fidelity of the cellular machinery during ageing and disease.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-05922-y. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Worm strains and demography assays
Nematodes were cultured using standard techniques at 20 °C on nematode growth media (NGM) agar plates and were fed with Escherichia coli strain OP50. The DR786 strain carrying the ama-1(m322) IV mutation in the large subunit of Pol II (RBP1), which confers α-amanitin resistance, was obtained from the Caenorhabditis Genetics Center 58,59 . The DR786 strain was then outcrossed into wild-type N2 strain four times and the mutation was confirmed by sequencing. The 5′-3′ AGAA-GGTCACACAATCGGAATC primer was used for sequencing. For each genotype, a minimum of 120 age-matched day 1 young adults were scored every other day for survival and transferred to new plates to avoid starvation and carry-over progeny. Lifespan analyses using the C. elegans Lifespan Machine were conducted as previously described 60 .
In brief, wild-type N2 and mutant worms were synchronized by egg-prep (hypochlorite treatment) and grown on NGM agar plates seeded with OP50 at 20 °C. Upon reaching L4 stage, these worms were transferred onto plates containing 0.1 g ml −1 5-fluoro-2′-deoxyuridine (FUDR) and placed into the modified flatbed scanners (35 worms per plate). The scan interval was 30 min. Objects falsely identified as worms were censored. Time of death was automatically determined by the C. elegans Lifespan Machine 60 . Demography experiments were repeated multiple times. For all experiments, genotypes were blinded. Statistical analyses were performed using the Mantel-Cox log-rank method.

Measurements of pharyngeal pumping rates in worms
Synchronized wild-type and ama-1(m322) animals were placed on regular NGM plates seeded with OP50 bacteria on day 1 and day 8 of adulthood, and the pharyngeal pumping rate was assessed by observing the number of pharyngeal contractions during a 10-s interval using a dissecting microscope and Leica Application Suite X imaging software. The pharyngeal pumping rate was then adjusted for the number of pharyngeal pumping per minute. Animals that displayed bursting, internal hatching and death were excluded from the experiments. Experiments were repeated three independent times in a blinded manner, scoring a minimum of 15 randomly selected animals per genotype and time point for each experiment. One-way ANOVA with Tukey's multiple comparison test was used for statistical significance testing, with ****P < 0.0001, and error bars representing standard deviation.

Fly strains and fly maintenance
The RpII215 C4 fly strain (RRID:BDSC_3663), which carries a single point mutation (R741H) in the gene encoding the Drosophila Pol II 215 kDa subunit (RBP1), was received from the Bloomington Drosophila Stock Center. Flies carrying the RpII215 C4 allele 61 are homozygous viable but show a reduced transcription elongation rate 6 . RpII215 C4 mutants were backcrossed for six generations into the outbred white Dahomey (wDah) wild-type strain. A PCR screening strategy was used to follow the RpII215 C4 allele during backcrossing. Therefore, genomic DNA from individual flies was used as a template for a PCR with primers SOL1064 (CCGGATCACTGCTGCATATTTGTT) and SOL1047 (CCGCGCGACTCAGG ACCAT). The 582-bp PCR product was restricted with BspHI, which specifically cuts only in the RpII215 C4 allele, resulting in two bands of 281 bp and 300 bp. At least 20 individual positive female flies were used for each backcrossing round. Long-lived insulin-mutant flies, which lack three of the seven Drosophila insulin-like peptides, dilp2-3,5 mutants (RRID:BDSC_30889) 62 , were also backcrossed into the wDah strain, which was used as the wild-type control in all fly experiments. Flies were maintained and experiments were conducted on 1,0 SY-A medium at 25 °C and 65% humidity on a 12-h light-12-h dark cycle 62 .

Fly lifespan assays
For lifespan assays, fly eggs of homozygous parental flies were collected during a 12-h time window and the same volume of embryos was transferred to each rearing bottle, ensuring standard larval density. Flies that eclosed during a 12-h time window were transferred to fresh bottles and were allowed to mate for 48 h. Subsequently, flies were sorted under brief CO 2 anaesthesia and transferred to vials. Flies were maintained at a density of 15 flies per vial and were transferred to fresh vials every 2-3 days and the number of dead flies was counted. Lifespan data were recorded using Excel and were subjected to survival analysis (log-rank test) and presented as survival curves.

Mouse maintenance and dietary restriction protocol
The dietary restriction (DR) study was performed in accordance with the recommendations and guideline of the Federation of the European Laboratory Animal Science Association (FELASA), with all protocols approved by the Landesamt für Natur, Umwelt und Verbraucherschutz, Nordrhein-Westfalen, Germany. Details of the mouse liver DR protocol have been previously published 63 . For the mouse kidney, male C57BL/ 6 mice were housed under identical specific pathogen-free conditions in group cages (five or fewer animals per cage) at a relative humidity of 50-60% and a 12-h light and 12-h dark rhythm. For DR versus control, 8-week-old mice were used. DR was applied for 4 weeks. Control mice received food and water ad libitum. Mice were killed at 12 weeks. For comparison of young versus aged mice, 14-week-old and 96-week-old mice were used. Food was obtained from ssniff (Art. V1534-703) and Special Diet Services. The average amount of food consumed by a mouse was determined by daily weighing for a period of 2 weeks and was on average 4.3 g daily. DR was applied for 4 weeks by feeding 70% of the measured ad libitum amount of food. Water was provided ad libitum. Mice were weighed weekly to monitor weight loss. Neither increased mortality nor morbidity was observed during DR.

RNA extraction
Wild-type N2 strain, α-amanitin-resistant ama-1(m322) mutants and long-lived insulin-IGF signalling mutants, daf-2(e1370), were sent for RNA-seq. For each genotype, more than 300 aged-matched adult worms at desired time points were collected in TRIzol (Thermo Fisher Scientific) in three biological replicates. Total RNA was extracted using the RNAeasy Mini kit (Qiagen). The RNA-seq data for brains of 30-day-old and 50-day-old dilp2-3,5 and wDah control flies have been previously published 64 . Ten-day-old and 50-day-old RpII215 C4 mutants and wDah control flies were snap frozen and fly heads were isolated by vortexing and sieving on dry ice. Total RNA from three biological samples per treatment group was prepared using TRIzol reagent according to the manufacturer's instructions, followed by DNase treatment with the TURBO DNA-free Kit (Thermo Fisher Scientific). Mouse liver samples were isolated from 5-month-old, 16-month-old and 27-month-old ad libitum and DR animals, which corresponded to 2, 13 and 24 months of DR treatment, respectively. RNA was isolated by TRIzol and was treated with DNase. The RNA-seq data for 5-month-old and 27-month-old liver DR samples have been previously published 63 , whereas the data for 16 months are first published here. The RNeasy Mini Kit and TRIzol were used to isolate RNA from snap-frozen kidneys as per manufacturer's instructions. Hypothalamus tissue of long-lived insulin receptor substrate 1 (Irs1 −/− ) knockout mice 65 and C57BL/6 black control animals was dissected manually at the age of 27 months. RNA was isolated by TRIzol with subsequent DNase treatment. For blood samples, globin RNA was removed using the GLOBINclear Kit mouse/rat/human for globin mRNA depletion.

Human whole-blood sample acquisition and RNA extraction
Human samples were obtained as part of a clinical study on ageing-associated molecular changes (German Clinical Trials Register: DRKS00014637) at University Hospital Cologne. The study cohort consisted of healthy participants between 21 and 70 years of age. Whole-blood samples were obtained using the PAXgene Blood RNA system (Becton Dickinson GmbH) directly after informed consent.
After storage at −80 °C for at least 24 h, RNA extraction was performed by usage of the PAXgene Blood RNA Kit (Qiagen) according to the manufacturer's protocol. The study was operated in accordance with the Declaration of Helsinki and the good clinical practice guidelines by the International Conference on Harmonization. All patients provided informed consent and approval of each study protocol was obtained from the local institutional review board (Ethics committee of the University of Cologne (17-362, 2018-01-17)).
Total RNA and nascent RNA-seq From 1 µg input of total RNA, ribosomal RNA was removed using the Ribo-Zero Human/Mouse/Rat kit (Illumina). Sequencing libraries were generated according to the TruSeq stranded total RNA (Illumina) protocol. To generate the final complementary DNA (cDNA) library, products were purified and amplified by PCR for 15 cycles. After validation and quantification of the library on an Agilent 2100 Bioanalyzer, equimolar amounts of libraries were pooled. Pools of five libraries were sequenced per lane on an Illumina HiSeq 4000 sequencer. For a description of all the RNA-seq datasets used in this study, see Supplementary Table 1. The same protocol was used to sequence cDNA libraries from human cell 'factory' RNA, which was isolated as previously described 66 .

RNA-seq alignments and gene expression analysis
Raw reads were trimmed with trimmomatic 67  '-outFilterType BySJout-outWigNorm None' on the genome version mm10, rn5, hg38, dm6 and ce5 for M. musculus, R. norvegicus, H. sapiens, D. melanogaster and C. elegans, respectively. We estimated transcript counts using Kallisto version 0.42.5 for each sample. To determine differentially expressed genes, we used DESeq2 version 1.8.2 (ref. 69) with RUVr normalization version 1.6.2 (ref. 70). For the differential analysis of transcriptional elongation regulators, we downloaded the list of positive and negative regulators from the GSEA/MSigDB 71 . Gene ontology (GO) term enrichment analysis of differentially expressed genes or genes with increased Pol II elongation speed was carried out using TopGO version 2.20.0. For GO enrichment analysis of differentially expressed genes, we identified 4,784 genes as evolutionarily conserved from each species of our study to humans: genes were either direct orthologues (one2one) or fusion genes (one2many) of H. sapiens were retrieved from ENSEMBL database using biomaRt 2.24.1 (ref. 72). Using our 4,784 genes evolutionary conserved, we further divided into consistently upregulated or downregulated genes across species during ageing or ageing intervention (as the target set for GO enrichment: 92 genes (ageing upregulated) and 71 genes (ageing downregulated); and 164 genes (ageing intervention upregulated) and 473 genes (ageing intervention downregulated) as background set of 4,784 orthologue genes between R. norvegicus, M. musculus, D. melanogaster, C. elegans and H. sapiens). For GO enrichment analysis of genes harbouring increasing Pol II speed, we used as the target set the top 200 or 300 genes with an increase in Pol II speed change for each species. Quantification of transcript abundance for ITPR1 and AGO3 was obtained by using StringTie 73 . For circRNAs, we aligned the reads using STAR version 2.5.1b 68 with the following parameters: '-chimSegmentMin 15-outSJfilterOverhangMin 15 15 15 15-alignSJover-hangMin 15-alignSJDBoverhangMin 15-seedSearchStartLmax 30-out-FilterMultimapNmax 20-outFilterScoreMin 1-outFilterMatchNmin 1-outFilterMismatchNmax 2-chimScoreMin 15-chimScoreSeparation 10-chimJunctionOverhangMin 15′. We then extracted back-spliced reads from the STAR chimeric output file and normalized the number of back-spliced reads by the sum of back-spliced (BS i ) and spliced reads from linear transcripts (S1 i , S2 i ) for an exon i (8): Here, S1 i refers to the number of linearly spliced reads at the 5′ end of the exon and S2 i refers to the respective number of reads at the 3′ end of the exon. Thus, this score quantifies the percent of transcripts from this locus that resulted in circRNA. Finally, we quantified the significance of the average change in circRNA formation between two conditions using the Wilcoxon rank test.

Definition of intronic regions
All annotation files for this analysis were downloaded from the Ensembl website 74 using genome version ce10 for C. elegans, mm10 for M. musculus, hg38 for H. sapiens, rn5 for R. norvegicus and dm6 for D. melanogaster. The following filtering steps were applied on the intronic ENSEMBL annotation files. First, we removed overlapping regions between introns and exons to avoid confounding signals due to variation in splicing or transcription initiation and termination.
Overlapping introns were merged to remove duplicated regions from the analysis. In the next step, we used STAR 68 to detect splice junctions and compared them with the intronic regions. Introns with at least five split reads bridging the intron (that is, mapping to the flanking exons) per condition were kept for subsequent analyses. As a result, we ensured a minimum expression level of the spliced transcript. When splice junctions were detected within introns, we further subdivided those introns accordingly. Introns with splice junction straddling were discarded. The above-mentioned steps were performed using Bedtools version 2.22.1 using subtract and merge commands. After these filtering steps, the number of usable introns per sample varied between a few hundred (n = 546, C. elegans, total RNA) to over 10,000 (n = 13,790, H. sapiens, nascent RNA-seq). These large differences resulted from different sequencing depths, sequencing quality (number of usable reads) and from the complexity of the genome (numbers and sizes of introns, number of alternative isoforms, among others). To avoid artefacts due to the different numbers of introns used per sample, we always contrasted the same sets of introns for each comparison of different conditions (for example, old versus young; treatment versus control). Note that certain comparisons were not possible for all species owing to variations in the experimental design. For instance, for mouse kidney, only a single time point after lifespan intervention (DR, age 3 months) was available, which prevented a comparison of old versus young DR mice, but allowed comparison with ad libitum-fed mice at the young age.

Transcriptional elongation speed based on intronic read distribution
To calculate Pol II speeds, we used RNA-seq data obtained from total RNA 75 and nascent RNA 65,76 enrichment. In contrast to the widely used polyA enrichment method 77 , which primarily captures mature, spliced mRNAs and is therefore not suitable to estimate Pol II speeds based on intronic reads, these methods yield sufficient intronic coverage to quantify elongation rates. To analyse the distribution of intronic reads between conditions, we devised a score for each intron. We fitted the read gradient (slope) along each of the selected introns (5′→3′; see above for the filtering criteria). Note that the intron gradient is not influenced by exonucleolytic degradation of excised intron lariats 20,78 and that this measure is only weakly associated with the expression level of the transcript 20 (Supplementary Table 2).
To transform slopes to Pol II elongation speed, we used the following formalism. We assume an intron of length L and we assume that at steady state, a constant number of polymerases is initiating and the same number of polymerases is terminating at the end of the intron; that is, we assume that premature termination inside the intron can be ignored. Polymerases are progressing at a common speed of k (base pairs per minute). The average time that it takes a polymerase to traverse the whole intron is hence: Transcription is initiated at a rate of n polymerases per unit time (1/ minute). Hence, the number of polymerases N initiating during t ∆ is: The slope s is the number of transcripts after the distance L minus the number of transcripts at the beginning divided by the length of the intron: Hence, slope and speed are inversely related and the speed depends also on the initiation rate (that is, the expression rate). However, we observed empirically only a small dependency between expression and slope 20 (Supplementary Table 2).
To validate our estimates of Pol II speeds, we compared our data with experimental values estimated via GRO-seq 19 and tiling microarray data 15 . There was a significant correlation (GRO-seq: R = 0.38, P = 4 × 10 −5 , compared with time point 25-50 min (see Jonkers et al. 19 ); tilling array: R = 0.99, P ≤ 2.6 × 10 −16 (data not shown)) between our data and experimentally measured transcriptional elongation values. We noted that our Pol II speed estimates for different introns of the same gene were more similar than Pol II speed estimates for random pairs of introns, implying that gene-specific factors or local chromatin structure influence Pol II speed (Extended Data Fig. 2b).

4sUDRB labelling, TUC conversion and elongation rate calculation
The estimation of transcription elongation speed using RNA labelling was based on the measurement of nucleotides added per time unit in a newly synthesized nascent transcript. First, transcription was reversibly inhibited by DRB to achieve accumulation of Pol II at the transcription start sites and synchronized transcriptional elongation initiation upon DRB removal. Simultaneously with the DRB removal, cells were pulsed for different time points with the uridine analogue 4sU to enrich for newly synthesized transcripts. Last, total RNA was isolated per each time point and the Pol II speed was determined by calculating the 4sU nucleotides added to the nascent transcript per time point. To estimate Pol II speed change in ageing cells, human fetal lung fibroblasts (IMR90) in proliferating and in senescent state were treated using this experimental procedure.
To select the time points to be used in the experiment, validate the DRB treatment and removal and check the enrichment efficiency of 4sU, a control experiment was set according to a previous protocol 79 . Two million proliferating cells (passage 14) were treated with 100 µM DRB (D1916, Merck) in their medium for 3 h at 37 °C and, upon DRB removal, they were pulsed with 1 mM 4sU (T4509, Sigma-Aldrich) for 0, 5,15,30,45,60,90 and 120 min. Immediately after the completion of each time point, cells were lysed in TRIzol (15596018, Thermo Fisher) and RNA was isolated with the Direct-Zol RNA mini-prep kit (R2052, Zymo Research). To validate DRB treatment, quantitative PCR with reverse transcription was performed in cDNA from all time points using the primers designed by Fuchs et al. 79 in proximal and distant introns of the OPA1 gene. Furthermore, to estimate 4sU enrichment, the RNA collected in each time point was biotinylated using the EZ-Link biotin HPDP kit (21341, Thermo Fisher) and biotinylated RNA was enriched with streptavidin-coated beads (DYNAL Dynabeads M-280 streptavidin; 11205D, Thermo Fisher). Evaluation of quantitative PCR with reverse transcription was performed also with the primers suggested by Fuchs et al. 79 against TTC-17 nascent and mature mRNA and 18S rRNA.
For the actual experiment, we performed the thiouridine-to-cytidine conversion sequencing (TUC-seq) protocol developed by Lusser et al. 80 to detect the 4sU-labelled transcripts in different time points. In this method, the thiol group of 4sU is quantitatively converted to cytidine via oxidation by OsO 4 in aqueous NH 4 Cl solution. The OsO 4 -treated RNA samples are submitted to RNA-seq to quantify labelled and non-labelled transcripts and define the number of reads containing uridine-to-cytidine conversions. To this aspect, 9 million proliferating (passage 9) cells and 9 million cells that had entered senescence (passage 35) were treated with 100 µM DRB for 3 h at 37 °C. Immediately after DRB removal, cells were pulsed with 1 mM 4sU for 0, 5 80 . Subsequent sample concentration and purification were also done according to this protocol. 4sU-labelled and OsO 4 -treated RNA samples derived from proliferating and senescent IMR90 cells in all five time points were subjected to RNA-seq. As a negative control for the TUC conversion, we used a mixture 1:1 of 4sU-labelled but not OsO 4 -treated samples from the time points 30 min and 45 min. The RNA-seq was performed in two biological replicates per condition.
Detection of labelled transcripts was performed based on the Lusser protocol, modified for Illumina RNA-seq: (1) FASTQ files were aligned to the genome using STAR to produce BAM files. (2) Sam2tsv 81 was then used to identify single-nucleotide mismatches.
(3) A custom R script was used to count the number of A-G or T-C mismatches per read. Only read pairs with at least three A-G or T-C mismatches were assumed to be 4sU-labelled and thus retained for subsequent analyses. Because the 5-min samples contained a very low number of reads with conversions, they were discarded from the rest of the analysis. We used two approaches for estimating the elongation rate per gene from the 4sU-labelling data. For the first approach, we tracked the progress of Pol II complexes constructing single-gene coverage profiles using 4sU-labelled reads. Progression was determined by picking the 99th percentile of gene body coverage in each sample to determine the front of elongating RNA polymerases. (We did not use the last converted read to determine the front because this measure would be too sensitive to noise in the data.) Elongation rates were calculated by fitting a linear model on the front positions of Pol II in 0, 15, 30 and 45 min in the first 100 kb of each gene. To determine elongation rates with greater accuracy, we filtered out genes with a length of less than 100 kb, as short genes can be fully transcribed in less than 45 min or even 30 min. This first approach of estimating Pol II speeds is characterized by high accuracy, but is limited to genes longer than 100 kb. The data in Fig. 1e are based on this approach.
The direct comparison of the 4sU data to the approach using read-coverage slopes in introns required a large set of genes for which Pol II speed could be measured using both assays. To maximize this gene set, we devised a second alternative approach for deriving speed from 4sU-labelling data that is applicable to shorter genes. For this second approach, we measured the front position of the polymerase in the same way as before (using the 99th percentile) but across the whole gene. For genes 30-100 kb long, we calculated the elongation rate from the difference in the front positions of the polymerase at 15 min and 30 min and divided this distance by 15 min to obtain speed measures per minute. For genes more than 100 kb long, we calculated the elongation rate from the difference in the positions of the polymerase at 30 min and 45 min divided by 15 min. This second speed measure is less accurate than the first measure, because it uses only two time points per gene; however, it enables estimating speed for genes shorter than 100 kb. The data in Fig. 1d are based on this second measure. Note that both measures confirmed the increase in average Pol II elongation speed from proliferating to senescent IMR90 cells.

S-methionine or 35 S-cysteine incorporation to measure translation rates in Drosophila
Ex vivo incorporation of radio-labelled amino acids in fly heads was performed as previously described 82 . In brief, 25 heads of each young (10 days) and old (50 days) wDah control and RpII215 C4 -mutant animals were dissected in replicates of five and collected in DMEM (#41965-047, Gibco) without supplements, at room temperature. For labelling, DMEM was replaced with methionine and cysteine-free DMEM (#21-013-24, Gibco), supplemented with 35 S-labelled methionine and cysteine (#NEG772, Perkin-Elmer). Samples were incubated for 60 min at room temperature on a shaking platform, then washed with ice-cold PBS and lysed in RIPA buffer (150 mM sodium chloride, 1.0% NP-40, 0.5% sodium deoxycholate, 0.1% SDS and 50 mM Tris, pH 8.0) using a pestle gun (VWR). Lysates were centrifuged at 13,000 rpm at 4 °C for 10 min and protein was precipitated by adding 1 volume of 20% TCA, incubating for 15 min on ice and centrifuging at 13,000 rpm at 4 °C for 15 min. The pellet was washed twice in acetone and resuspended in 200 µl of 4 M guanine-HCl. Of the sample, 100 µl was added to 10 ml scintillation fluid (Ultima Gold, Perkin-Elmer) and counted for 5 min per sample in a scintillation counter (Perkin-Elmer). Protein determination was done in duplicates (25 µl each) per sample using the Pierce BCA assay kit (Thermo Fisher Scientific). Scintillation counts were normalized to total protein content.

Mismatch detection
Mismatch detection was performed using the tool rnaseqmut (https:// github.com/davidliwei/rnaseqmut), which detects mutations from the NM tag of BAM files. To avoid detection of RNA editing or DNA damage-based events, we only considered genomic positions with only one mismatch detected (that is, occurring in only one single read). Reads with indels were excluded and only mismatches with a distance of more than four from the beginning and the end of the read were considered. A coverage-level filter was applied so that only bases covered by at least 100 reads were kept. A substantial number of mismatches may result from technical sequencing errors. However, as young and old samples were always handled together in the same batch, we can exclude that consistent differences in the number of mismatches are due to technical biases. The fraction of RNA editing events is generally relatively low and not expected to globally increase with age 83 .

MNase-seq sample preparation
Mononucleosomal DNA from proliferating and senescent IMR90 cells (from two donors) were prepared and sequenced on an Illumina HiSeq4000 platform as previously described 84 . For fly heads, a MNase digestion assay was performed using the EZ nucleosomal DNA prep kit, as per the manufacturer's guidelines (Zymo Research). In brief, 25 snap-frozen heads were lysed in nuclei prep buffer and incubated on ice for 5 min. Cuticle fragments were then removed via centrifugation (at 50g for 30 s). Nuclei were pelleted (for 500g at 5 min) and washed twice in digestion buffer and resuspended in 100 µl of digestion buffer. Nucleosome footprints were then digested using 0.05 U of MNase (Zymo Research). Samples were taken at 0, 2, 3 and 5 min or 10 min for prolonged digestion, and immediately stopped in MN stop buffer (Zymo Research). Samples were isolated using Zymo Spin IIC columns. Nucleosome footprints (1:10 dilution) were visualized by Tapestation using High-sensitivity D1000 ScreenTape (Agilent).

MNase-seq analysis
We used nucleR 85 with default parameters to calculate peak sharpness as a combination of peak width and peak height. Peak width was quantified as the standard deviation around the peak centre, and peak height was quantified as the number of reads covering each peak 85 and the distance between peak summits. Intron and exon annotations were downloaded from UCSC table utilities 74 and filtered as described in definition of intronic regions. Nucleosome density (Fig. 5a) is defined as the number of nucleosome peaks found within an exon or an intron divided by the length of the exon or intron.

Western blotting
Western blots were carried out on protein extracts of individual dissected tissues. Proteins were quantified using BCA (Pierce). Equal amounts were loaded on Any-KD pre-stained SDS-PAGE gels (Bio-Rad) and blotted according to standard protocols. Antibody dilutions varied depending on the antibody and are listed here: histone H3 (1:1,000) and HP1 (DSHB) (1:500). Appropriate secondary antibodies conjugated to horseradish peroxidase were used at a dilution of 1:10,000.