Transcriptional memory of dFOXO activation in youth curtails later-life mortality through chromatin remodelling and Xbp1

Abstract

A transient, homeostatic transcriptional response can result in transcriptional memory, programming subsequent transcriptional outputs. Transcriptional memory has great but unappreciated potential to alter animal ageing as animals encounter a multitude of diverse stimuli throughout their lifespan. Here we show that activating an evolutionarily conserved, longevity-promoting transcription factor, dFOXO, solely in early adulthood of female fruit flies is sufficient to improve their subsequent health and survival in mid- and late life. This youth-restricted dFOXO activation causes persistent changes to chromatin landscape in the fat body and requires chromatin remodellers such as the SWI/SNF and ISWI complexes to program health and longevity. Chromatin remodelling is accompanied by a long-lasting transcriptional programme that is distinct from that observed during acute dFOXO activation and includes induction of Xbp1. We show that this later-life induction of Xbp1 is sufficient to curtail later-life mortality. Our study demonstrates that transcriptional memory can profoundly alter how animals age.

Introduction

Transcription is the first and key step in gene expression, the process linking a genotype to a phenotype. Orchestrated by transcription factors (TFs)¹, transcriptional programmes allow an animal to develop as well as to respond to its environment as an adult^2,3. Indeed, transcriptional control has been studied within two such paradigms⁴: the programmatic paradigm of development and differentiation whereby permanent changes in gene expression determine cellular fates; and the homeostatic paradigm where cells maintain function by transiently reshaping gene expression. Similarly, TFs can be seen as either determining cell identity or allowing environmental responsiveness. However, these two paradigms cannot account for certain phenomena, such as transcriptional memory, where a homeostatic, transient transcriptional event produces a lasting, developmental-like impact on subsequent gene expression^5,6. The mechanisms underpinning such transcriptional memory have attracted attention and are being intensely examined in cell systems, e.g. the response to galactose in the budding yeast⁷ and the interferon response in mammalian cells in culture^6,8. However, a broader understanding of the occurrence of transcriptional memory, its mechanisms and relevance for adult animal physiology is lacking.

Ageing is an intrinsic process that occurs in most animals⁹ and results in increased disease susceptibility and reduced likelihood of survival with time^10,11. In humans, old-age health is shaped by numerous environmental variables throughout our life course¹², and similar observations have been made in animal models^13–18. However, the mechanisms underlying these long-term effects remain unclear. The involvement of transcriptional memory in this context has not been tested. Long-term effects of transcriptional memory have the potential to be especially important in ageing as animals encounter a variety of stimuli throughout their lifespan.

Forkhead Box O (FOXO) TFs maintain metabolic homeostasis in response to nutritional cues³. They are inhibited by the insulin/IGF signalling pathway and their activation promotes longevity in yeast, worms, flies, and likely humans^3,19–21. In flies, this activity is tissue-specific, with adult-onset over-expression of dfoxo in fruit fly fat body (equivalent to mammalian liver and adipose tissue) and midgut (hereafter the gut, equivalent to the mammalian small intestine) promoting longevity; it is also sexually dimorphic extending female lifespan robustly, without a substantial effect on male longevity^22,23. This fly model provides a highly tractable system with which to study transcriptional memory in the context of animal ageing, as it allows tissue-specific dfoxo induction to be switched on and off. Using this system, we set out to study transcriptional memory in the context of longevity in Drosophila females.

Results

dfoxo induction solely in early adulthood extends lifespan

We wanted to activate dFOXO in young adults and examine the effects of this activation in later life. Drosophila females are youthful in the first three weeks of adulthood; they are reproductively active, maintain essentially normal physiological function and are unlikely to die. Middle-aged flies, 20 to 60 days old in our healthy outbred background, lay substantially fewer eggs²⁴, experience functional impairments, e.g. loss of neuromuscular performance²⁵, but not high mortality. Older flies display substantial pathology, e.g. of the gut²⁶, and high mortality²⁷.

We transiently overexpressed dfoxo in the fat body and gut of adult females from day 2 to 23 of adulthood (hereafter referred to as dfoxo-switch, Fig. 1a) employing an inducible driver, S106²⁸ and feeding the inducer, RU⁴⁸⁶. This driver did not show any substantial induction in the absence of RU⁴⁸⁶ (“leakiness”, as reported with other GeneSwitch drivers^29–31) and has been thoroughly characterised by us (Extended data Fig. 1a-c) and others^28,32. Western blotting confirmed that dFOXO was increased in both organs during the induction with levels returning to normal within a week on food without RU⁴⁸⁶ (Fig. 1b, c, Extended data Fig. 1d; levels in gut/fat body as expected for this driver, Extended data Fig. 1a). Interestingly, even though dFOXO levels did not persist, induction of dfoxo in young adults consistently extended subsequent lifespan, in line with previous reports (Fig. 1d, ref.^16,28). The effect was not observed in S106 or UAS-dfoxo-alone controls, indicating that it is not an RU⁴⁸⁶-feeding artefact (Extended data Fig. 2a and b). We observed the expected sexual dimorphism, with dfoxo-switch not extending male lifespan (Extended data Fig. 2c and d). Furthermore, dfoxo induction in the gut alone was insufficient for longevity (Extended data Fig. 2e), indicating that the fat body plays an essential role.

Fig. 1. Transient expression of dfoxo in early adulthood extends subsequent lifespan.

a Experimental setup: 2-day-old S106>dfoxo females were placed on food containing RU⁴⁸⁶ to induce dfoxo expression until day 23 when they were placed back on food without RU⁴⁸⁶ (dfoxo-switch flies, red arrow). Their sisters were not fed RU⁴⁸⁶ at any time (grey arrow). b Western-blot quantifications of gut or fat body dFOXO during induction (day 7) and one week after (day 30). Boxplot – quantiles; whiskers – extremes; overlay – individual data points. N = 4 biologically independent samples; day 7: effect of RU⁴⁸⁶ p = 0.035, tissue p = 0.0018, RU⁴⁸⁶-by-tissue interaction p = 0.33; day 30: RU⁴⁸⁶ p = 0.169, tissue p = 0.048, interaction p = 0.13; linear model (LM). C Representative western blots against dFOXO and Actin. * non-specific binding by dFOXO antibody in fat body samples. FB – fat body, G – gut. Same-day samples on the same membrane, sequentially probed with anti-dFOXO and anti-Actin. Image source data – Extended data Fig. 1d. d dfoxo-switch lifespan. Control n = 139 dead/3 censored flies, dfoxo-switch n = 141 dead/4 censored flies, p = 2 × 10^–10, log-rank test. e Height climbed by dfoxo-switch compared to control females, from day 23, pooled from three independent trials. Individual trials – Extended data Fig. 2g. N – individual flies; boxplots – quantiles; whiskers – extremes; overlay – individual data points. Effect of dfoxo-switch p = 0.5648, age p < 10^–4, age-by-dfoxo-switch interaction p = 0.0118, mixed-effects LM. f Hazard ratios (HRs) – points, and 95% confidence intervals (CI) – whiskers, from a mixed-effects Cox Proportional Hazards (CPH) model on the combined events (985 dead, 9 censored) from four independent trials. Individual trials – Extended data Fig. 2h. HRs < 1 indicate switched flies exhibit lower risk of death compared to uninduced controls. Detailed statistical analyses for b and f are shown in Supplementary Data.

Can this long-term memory be established by any pro-longevity TF? Anterior open (Aop) is a transcriptional repressor whose chronic activation extends female lifespan from the same tissues as dFOXO’s^32,33. However, transiently inducing the activated form of aop (aop^ACT) in young adult females did not extend their subsequent lifespan (Extended data Fig. 2f), indicating that the ability to trigger a persistent effect is not a general characteristic of pro-longevity TFs.

Induction of dfoxo in the gut and fat body was previously shown to delay the age-related decline in neuromuscular performance³⁴, as measured by the fly’s climbing ability (negative geotaxis assay)³⁵. Similarly to its effects on survival, dfoxo-switch was sufficient to delay the age-related decline in climbing ability (detected as significant age-by-dfoxo-switch interaction, Extended data Fig. 2g). Combining data from three independent replicates allowed us to confirm that this beneficial effect persisted after day 23, when dfoxo is already switched off (Fig. 1e, Supplementary Data). Hence, early-adulthood induction of dfoxo improves subsequent health.

Finally, to explicitly determine how long the effect of dfoxo-switch persists, we combined demographic data from four independent experiments, recording 985 fly deaths, in a mixed effects Cox Proportional Hazards (CPH) model. By estimating the Hazard Ratios (HR), we found that transient dfoxo induction in young adults significantly reduced the fly’s risk of death in all subsequent periods by almost 50%. (Fig. 1f, Extended data Fig. 2h, Supplementary Data). Hence, our data show that induction of dfoxo in young adults generates a beneficial memory effect that persists in old age, influencing mortality half a lifetime later.

Benefits of dfoxo-switch require chromatin remodellers

Chromatin alterations are the prime candidate mechanisms for this lasting effect of dfoxo-switch as they can be triggered by TFs, persist over time, and are implicated in transcriptional memory^36,37. To assess in vivo chromatin architecture, we employed the Assay for Transposase-Accessible Chromatin combined with next generation sequencing (ATAC-Seq) on dissected fat bodies and guts of dfoxo-switched females a week after foxo induction had ceased. As controls, we used their uninduced sisters, housed in parallel. We observed 3447 ATAC-Seq peaks in the fat body and 2906 in the gut samples (Supplementary Data). Their chromosomal distribution, sizes, and transcriptional start site (TSS) distance are shown in Extended data Fig. 3a-c. Dimensionality reduction, using t-distributed stochastic neighbour embedding (t-SNE), easily distinguished dfoxo-switched flies from controls based on the intensity of the ATAC-Seq peaks from the fat body but not from the gut samples (Fig. 2a). Indeed, DESeq2 analysis identified 81 regions that were differentially accessible to the transposase in the fat body after dfoxo-switch but none in the gut at 10% false discovery rate (FDR) (Supplementary Data; examples of genomic regions with different characteristics and confirmed by qPCR are shown in Extended data Fig. 3d and e). 37% of these 81 peaks were 1kb from TSS and 72% were more accessible after dfoxo-switch, indicating opening of chromatin at promoter-proximal regions. Hence, dFOXO imprints persistent changes on chromatin landscape specifically in the adult fly fat body.

Fig. 2. dfoxo-switch induces persistent changes in chromatin structure and requires chromatin remodellers for longevity.

a tSNE plots generated from the intensities of all detected ATAC peaks in foxo-switch and control fat bodies and guts, after variance stabilizing transformation (VST). b Venn diagram showing overlap between fat body ATAC peaks with significantly altered accessibility after dfoxo-switch (81 peaks) and previously described dFOXO-bound ChIP peaks (1361 peaks, p = 0.008, one-sided permutation test). c Δdfoxo S106>dfoxo switch lifespans. Control n = 139 dead/4 censored flies, dfoxo-switch n = 136 dead/9 censored flies, p = 1.95 × 10^-19, log-rank test. d Proportion of histone modification ChIP datasets deposited in ChIP-atlas, classified as underpinning gene activation, repression, or other, for the datasets where the ChIP peaks significantly overlap (10% FDR) our ATAC peaks that show differential accessibility caused by dfoxo-switch (right bar; 89 marking activation, 22 repression and 8 other) or those for which this overlap is not significant (left bar; 702 marking activation, 545 repression and 98 other), p = 5.893 × 10^-6, Pearson's Chi–squared test. See also Supplementary Data. e HRs – points, and 95% CIs – whiskers, indicate the relative risk of death in switched flies compared to their uninduced sisters after day 23 and were determined for dfoxo-switch when dfoxo is induced in the presence of RNAi constructs targeting the indicated genes (right) or the same RNAi constructs transiently expressed on their own (left). HRs < 1 indicate a lower risk of death compared to uninduced controls, HRs > 1, the opposite. P value for HRs being significantly different from 1, CPH models. Grey vertical area highlights the results for mor, osa and iswi. Individual lifespans and demographic details - Extended data Fig. 4 and Supplementary Data.

Interestingly, we found that 4 out of 81 genomic regions where chromatin accessibility was altered by dfoxo-switch overlapped regions that were previously characterised as bound by dFOXO in the fat body/gut using Chromatin Immunoprecipitation (ChIP) (Fig. 2b, ref.³²). This relatively small overlap was statistically significant and prompted us to examine if dFOXO induction in early adulthood was sensitive to the levels of endogenous dFOXO. We found that dfoxo-switch performed in flies lacking endogenous dfoxo could extend their subsequent lifespan (Fig. 2c), indicating that dFOXO is unlikely to programme longevity by facilitating its own, subsequent access to chromatin.

To further characterise the genomic sites associated with these 81 regions of dfoxo-switch-responsive chromatin, we examined the 5176 ChIP datasets publicly available in ChIP-Atlas³⁸. Focusing on histone modifications classified as either marking active or repressive chromatin (or “other”, Supplementary Data), we found that active marks tended to be enriched within our differentially accessible ATAC peak regions (Fig. 2d). Such active histone marks are often associated with a number of ATP-dependent chromatin remodellers that have vital and flexible roles in modulating chromatin structure, including formation of transcriptional memory in yeast and mediating the longevity effects of TF such as DAF-16, the worm FOXO^36,39,40.

We tested the involvement of proteins that are constituents of 10 of the 11 Drosophila ATP-dependent chromatin remodelling complexes (noted in Fig. 2e) in the physiological memory established by dFOXO, using validated RNAi constructs. We observed two different types of behaviour (Fig. 2e; all lifespans are presented in Extended data Fig. 4, data and analyses in Supplementary Data). Downregulation of tip-60, Mi-2 and domino during the first three weeks of adult life significantly increased subsequent mortality (after 23 days). Hence, the chromatin landscape set up in early adulthood by their chromatin remodelling and histone acetylation activity⁴¹ can impact fly physiology at older ages. However, the detrimental effects of tip60 and domino could be countered by co-expression of dfoxo, suggesting that dFOXO can extend lifespan independently from tip60 and domino levels with the caveat that tip60 and domino knockdowns may have been incomplete even though ubiquitous, constitutive expression of these RNAi constructs with the daughterlessGal4 driver resulted in 100% pre-adult lethality, in concordance with previously published work⁴². The detrimental effect of mi-2 was not fully counteracted by dfoxo, indicating it may be involved in its long-term effects.

On the other hand, RNAi against moira, osa, and iswi induced solely in young adults did not influence their subsequent lifespan while it did block the beneficial effects of dfoxo, demonstrating that they are required to form the physiological memory of dFOXO. Indeed, inducing either mor or iswi RNAi was sufficient to block the beneficial effects of dfoxo-switch on age-related decline in neuromuscular performance, whereas each RNAi alone did not reduce climbing performance (Extended data Fig. 5, Supplementary Data).

Additionally, we tested the involvement of etl1 and nejire (nej), as they also have ATP-dependent chromatin remodelling functions⁴³ and nej (p300/CBP) interacts with FOXO in different contexts^44,45. We found that both silencing etl1 and nej was able to block dFOXO’s ability to programme lifespan, similarly to mor, osa and iswi. Note that RNAis against mCherry, luciferase and driver alone controls did not extend lifespan when induced on their own and did not block the lifespan extension produced by dfoxo-switch. Moreover, reducing the expression of HP1, a protein involved in transcriptional regulation and heterochromatin formation⁴⁶, did not have a significant effect on lifespan on its own, and did not alter the lifespan effect of dfoxo-switch.

Altogether, we identified specific members of chromatin remodelling complexes, namely moira, osa and iswi, as required for dFOXO to cause its beneficial, long-term effect on longevity and healthspan. These data reinforce the results of our ATAC-Seq analysis and indicate that transient activation of dfoxo in young adult females generates persistent changes in chromatin landscape that foster improved health and survival in mid- and late life.

A transcriptional programme is triggered by dfoxo-switch

The changes in chromatin landscape observed after dfoxo-switch are likely to underpin a long-term transcriptional programme. To investigate this, we profiled the transcriptomes of isolated fat bodies and guts from dfoxo-switched females a week after dfoxo induction had ceased, using their uninduced sisters as controls. Genes that were detected as responding to RU⁴⁸⁶ in driver-alone females (two in the fat body and 10 in the gut, Supplementary Data) were removed to avoid artifacts of RU⁴⁸⁶ feeding. 461 genes were differentially expressed after dfoxo-switch in the fat body (10% FDR, Fig. 3a and b, Supplementary Data), and only 87 in the gut. We confirmed by qPCR the expected fat body induction for three genes tested (HDAC6, Pfk and Pepck1; Extended data Fig. 6a). Interestingly, at least some of the transcriptional changes could still be detected in whole flies at week 7 of adulthood (Extended data Fig. 6b), indicating a long-lasting effect.

Fig. 3. A unique transcriptional programme is triggered in the fat body by dfoxo- switch.

a Volcano plots showing the effect of dfoxo-switch on transcripts in the fat body and the gut with differentially expressed genes (FDR 10%) shown in red. b Overlaps of differentially expressed genes between the effects of dfoxo-switch and acute dfoxo induction in the fat body and gut. Differential expression in dfoxo-acute set is based on a meta-analysis of two previously described datasets that profiled the transcriptome during dFOXO induction (essentially day 7). Overlap p-values from one-sided hypergeometric test. c Heatmaps of dfoxo-switch gene – dfoxo-switch peak pairs assigned with BETA. The left shows ATAC-seq and the right RNA-seq signal intensities (after VST and scaling to the control condition for each peak/gene). Each column is a biologically independent sample. Genes down-regulated in RNA-Seq are presented at the bottom.

We assessed if the gene expression programme imposed after dfoxo-switch (after dfoxo induction has ceased) is related to transcriptional changes observed during acute dfoxo induction. We performed a meta-analysis of two published transcriptomic studies^32,33, each performed in the same genetic background and with the same experimental approaches and conditions as ours, to define a list of genes differentially expressed during acute dfoxo induction in the adult fat body/gut (hereafter referred to as dfoxo-acute) and compared it to our dfoxo-switch gene list. In the fat body, only 10 genes were differentially expressed during both acute dfoxo induction and a week after dfoxo-switch (Fig. 3b, Extended data Fig. 6c and d, Supplementary Data). For these, the expression changes tended to be in the opposite direction between acute dfoxo induction and dfoxo-switch (Extended data Fig. 6e). We confirmed by qPCR that the levels of several transcripts differentially expressed after dfoxo-switch were not responsive to acute dfoxo induction (Extended data Fig. 6f). Overall, the persistent changes triggered by dfoxo-switch in the fat body did not appear to be carried over from acute dfoxo induction in this organ. On the other hand, we observed a significant overlap between acute and persistent transcriptional changes in the gut (30 genes, Fig. 3b, Extended data Fig. 6c, d), indicating some, albeit limited continuation of the acute transcriptional programme in this organ.

Interestingly, 98% of differentially expressed genes in the fat body were upregulated (Fig. 3a). This is consistent with the detected increase in chromatin accessibility in this organ, and the involvement of chromatin remodelling complexes linked to transcriptional activation⁴⁷. To further elucidate the links between differentially open chromatin regions and differentially regulated genes, we used Binding and Expression Target Analysis (BETA⁴⁸) to integrate the two datasets (ATAC-Seq and RNA-Seq) and infer genes whose expression changed as the result of alterations in chromatin accessibility. We found that differentially accessible chromatin regions significantly explained upregulation of transcript levels (Extended data Fig. 6g) and identified 190 chromatin region-gene pairs, including 159 genes for which expression changes can be explained by altered chromatin accessibility in neighbouring genomic regions in the fat body; the majority of the genes were upregulated after the switch (Fig. 3c).

Our transcriptomic data are in line with the ATAC-Seq findings, both demonstrating that the effects of dfoxo-switch are different between the gut and the fat body. Indeed, transient induction of dFOXO can set up a distinct transcriptional programme in the fat body, likely mediated by changes to chromatin landscape in that organ, while it only leaves a small, residual of the original programme in the gut. For this reason, we focused our further investigation on the fat body.

dfoxo-switch flies exhibit a distinct metabolic profile

To explore the physiological consequences of the long-term transcriptional programme in the fat body, we conducted Gene Ontology (GO) enrichment analysis on the entire set of genes differentially expressed in this organ after dfoxo-switch. We observed a strong overrepresentation of genes involved in metabolic pathways, particularly those of glucose metabolism (Fig. 4a, Extended data Fig. 7a). The overrepresented GO terms observed in the dfoxo-switch were different to those in dfoxo-acute (Extended data Fig. 7 and ref.^32,33). As metabolic dysregulation is a consequence of ageing¹⁰ and as TFs and chromatin structure play important roles in metabolic homeostasis³⁷, we interrogated the metabolic profiles of dfoxo-switch flies, using liquid chromatography–mass spectrometry (LC/MS) analysis on whole-fly extracts. We detected 96 putative metabolites with levels significantly affected by dfoxo-switch, nine of which were identified using internal standards (Fig. 4b, Supplementary Data). Several could be mapped to the metabolic pathways that were transcriptionally regulated by dfoxo-switch in the fat body (Fig. 4c). We observed a significant increase of pyruvate and decrease of glycerol-3-phosphate in our dfoxo-switch flies compared to the controls, showing a concordance between transcriptional and metabolite changes in glucose metabolism, indicating that many of the metabolic changes, e.g. those in glucose metabolism, are likely occurring in the fat body. However, metabolic alterations in other tissues may also have been detected in our metabolomics analysis as it was performed on whole flies: e.g. the observed, high increase in acetylcholine may be coming from the brain.

Fig. 4. dfoxo-switch flies exhibit a distinct metabolic profile.

a. Top 5 GO terms and KEGG pathways for all the genes differentially expressed after dfoxo-switch in the fat body. See Supplementary Data. b Relative changes in metabolites comparing females after dfoxo-switch with their controls. Identified metabolites are labelled and marked in red. See Supplementary Data. c Metabolic map highlighting the enzymes with significant changes in mRNA levels after dfoxo-switch (red lines) and significantly altered, identified metabolites (grey and red dots). The pathways in which both differentially expressed genes and altered metabolites are involved are annotated. d Pfk-reverse switch lifespan. S106>Pfk females were fed food containing RU⁴⁸⁶ chronically or from day 23 (reverse switch). Control n = 127 dead/10 censored flies, Pfk-reverse switch n = 119 dead/3 censored flies, p = 0.03 vs control, log-rank test, Pfk-chronic n= 127 dead/10 censored flies p = 0.007 vs control, log-rank test. e Pepck1-reverse switch lifespan. S106>Pepck1 females were fed food containing RU⁴⁸⁶ chronically or from day 23 (reverse switch). Control n = 141 dead/5 censored flies, Pepck1-reverse switch n = 115 dead/5 censored flies, p = 0.97 vs control, log-rank test, Pepck1-chronic n= 91 dead/4 censored flies p = 0.46 vs control, log-rank test.

This metabolic reprogramming mediated by dfoxo-switch may underly the long-term effects on longevity. Indeed, reduced expression of carbohydrate metabolism enzymes is a cause of ageing in Drosophila males and the over-expression of the genes encoding these enzymes can extend male lifespan⁴⁹. We tested if transcriptionally upregulating the enzymes catalysing the committed and rate-limiting step in glycolysis, Phosphofructokinase (Pfk), or gluconeogenesis, Phosphoenolpyruvate carboxykinase 1 (Pepck1), is enough to affect female longevity. Both genes were found to be upregulated in the fat body after dfoxo-switch, but not during acute dfoxo induction (Extended data Fig. 6a and f, Supplementary Data). To mimic the effects of dfoxo-switch, we induced their expression from day 23 of adulthood (reverse-switch). Both this late-onset and chronic overexpression of Pfk or Pepck1 produced either a modest or no effect on lifespan (Fig. 4d, e). Hence, our data are consistent with induction of dfoxo in young adults reprogramming metabolism in later life. However, this may not be sufficient to explain fully the lifespan extension observed.

Induction of Xbp1 accounts for longevity after dfoxo-switch

We sought to further understand the nature of the transcriptional programme triggered by dFOXO activation in youth. 17 TFs were differentially expressed in the fat body after dfoxo-switch. To examine if any one of them may drive transcriptional changes after the switch, we sought DNA sequence motifs over-represented in the promoters of all the genes differentially expressed in the fat body after dfoxo-switch (Supplementary Data). The top three enriched motifs contained the ACGT sequence (Fig. 5a), the core sequence bound by the human XBP1. Interestingly, our transcriptomic data indicated that the fly Xbp1 was upregulated after dfoxo-switch, highlighting a dFOXO–Xbp1 relay as a potentially important component of the lifespan programming by dfoxo-switch.

Fig. 5. Xbp1 activation accounts for longevity resulting from dfoxo-switch.

a The three top-ranked motifs identified as enriched within the promoters of genes differentially expressed after dfoxo-switch in the fat body. Underlined – ACGT core sequence bound by Xbp1; NES – Normalised Enrichment Score. b qPCR quantifications of Xbp1^s and Xbp1^u transcripts in dfoxo-switched fat bodies at day 30. N – biologically independent samples; boxplots – quantiles; whiskers – extremes; overlay – individual data points. Effect of dfoxo-switch p < 7×10^-4, effects of transcript or dfoxo-switch-by-transcript interaction p > 0.05, mixed effects LM. xc Overlap of differentially expressed genes between dfoxo-switch in the fat body and Xbp1 mutant larvae. Overlap p-value from one-sided hypergeometric test. d HRs - points, and 95% CI – whiskers, from CHP models showing the relative risk of death during tunicamycin feeding initiated after a week of recovery from dfoxo-switch or dfoxo-switch co-induced with RNAi against mor or iswi. HR < 1 indicates a lower risk of death compared to uninduced controls. P value from CPH models. See Extended data Fig. 7g-j. e qPCR quantifications of Xbp1^s and Xbp1^u in fat bodies a week after dfoxo-switch or dfoxo-switch performed together with induction of mor or iswi RNAi. N – biologically independent samples; boxplots – quantiles; whiskers – extremes; overlay – individual data points. Effect of dfoxo-switch p = 0.035, dfoxo + iswi^RNAi-switch p>0.05, dfoxo + mor^RNAi-switch p < 4×10^-4; effects of transcript or switch-by-transcript interaction p > 0.05; mixed effects LMs. f Lifespans of S106>Xbp1^RA females fed food containing RU⁴⁸⁶ chronically or from day 23 (reverse switch). Control n = 105 dead/1 censored flies; Xbp1^RA-reverse switch n = 106 dead/1 censored flies p = 0.021 vs control, log-rank test; Xbp1^RA-chronic n = 138 dead/2 censored flies p = 0.0265 vs control, log-rank test. g Lifespans of S106>Xbp1^S females. Control n = 145 dead/3 censored flies; Xbp1^S-reverse switch n = 143 dead/0 censored flies, p = 6.03 × 10^-5 vs control, log-rank test; Xbp1^S-chronic n = 154 dead/0 censored flies p = 0.0001 vs control, log-rank test.

Xbp1 is an evolutionarily conserved TF that acts in one of the three branches of the unfolded protein response⁵⁰. Xbp1 activity is regulated post-transcriptionally through unconventional splicing of its mRNA⁵¹. The transmembrane protein encoded by Inositol-requiring enzyme 1 (Ire1 in Drosophila) catalyses the unconventional splicing of the Xbp1 mRNA in response to cues, e.g., accumulation of unfolded proteins in the endoplasmatic reticulum (ER) or metabolic dysregulation. The spliced Xbp1 mRNA encodes a highly active TF that induces the expression of a plethora of genes to alleviate proteostatic stress and regulate metabolism^51–56.

We further investigated the role of Xbp1 in the long-term benefits of dfoxo-switch. Our RNA-Seq analysis couldn’t distinguish between different Xbp1 isoforms. Using qPCR, we found that both the spliced and unspliced version of Xbp1 mRNA (Xbp1^S and Xbp1^U respectively) were indeed increased after dfoxo-switch in the fat body (Fig. 5b), indicating regulation of this TF at the level of transcription rather than splicing, which still results in increased levels of Xbp1^S mRNA. Consistent with activation of Xbp1 after the switch, we found a significant overlap in the genes whose expression depends on Xbp1 in fly larvae⁵⁷ and those installed by dfoxo-switch in adult fat body (Fig. 5c). We could identify at least three genes, hsc70-4, kay and eip75b, for which the ectopic induction of Xbp1^S mimicked the effect of dfoxo-switch on their expression (Extended data Fig.8a and b). To check for physiological relevance of Xbp1^S upregulation, we examined the tolerance of our dfoxo-switched flies to orally administered tunicamycin, which induces proteostatic stress by inhibiting glycosylation of proteins in the ER⁵⁸. We found a small but significant increase in tunicamycin resistance after dfoxo-switch [shown as reduced risk of death (HR) in Fig. 5d; survival in Extended data Fig. 8c and d, see Supplementary Data], consistent with activation of Xbp1, whereas we found no difference in starvation resistance (Extended data Fig. 8e and f) indicating a specific increase in proteostatic stress tolerance.

How does dFOXO regulate Xbp1 expression? Xbp1 was not upregulated during acute dFOXO induction, dFOXO binding has not been documented near the Xbp1 locus in the fat body and we did not see significant changes in chromatin accessibility at this locus following dfoxo-switch (ref.^32,33, Extended data Fig. 6f, Extended data Fig. 3e), making it overall unlikely that Xbp1 is a direct transcriptional target of dFOXO. Rather, Xbp1 appears indirectly induced by dfoxo-switch. To test weather this indirect induction requires the same chromatin remodellers that are required for the longevity effect of the dfoxo-switch, we knocked down mor or iswi during the switch: this prevented the upregulation of Xbp1 (Fig. 5e). Similarly, tunicamycin tolerance after dfoxo-switch was dependent on mor and iswi: knockdown of either during dfoxo induction in the first three weeks of adulthood blocked any beneficial effect of dfoxo-switch on tunicamycin tolerance in week 4 (Figure 5d; note mor^RNAi alone reduced tunicamycin tolerance, Extended data Fig. 8g-j). Hence, like longevity, Xbp1 induction and the resulting proteostatic-stress tolerance after dfoxo-switch are also dependent on chromatin remodellers. This prompted us to directly query the role of Xbp1 in longevity.

Xbp1 has been intensely studied in the context of proteostatic stress and has been linked to metabolism and animal longevity more recently^59–61. Therefore, we tested whether the fly orthologue also promotes longevity when expressed in a time-restricted manner by inducing either the full, splicable Xbp1 mRNA (Xbp1^RA) or specifically the spliced Xbp1^S, ⁵⁷. To mimic the effects of dfoxo-switch, we induced expression only from day 23 of adulthood (reverse-switch). Both this late-onset and chronic overexpression of Xbp1^RA or Xbp1^S were sufficient to extend lifespan, recapitulating the longevity of dfoxo-switched flies (Fig. 5f and g). Hence, our data are consistent with induction of dfoxo in young adults resulting in increased, later expression of Xbp1, which improves survival of these flies in mid- and late life.

dfoxo-switch may counteract transcriptional ageing

Chromatin disorganisation and gene expression dysregulation occur during ageing in multiple species⁶². We were interested to see if dfoxo-switch-responsive transcripts show age-related deregulation. To define the ageing fat body transcriptome, we analysed transcriptomic changes occurring in female fat bodies from day 10 to day 50 using published data⁶³. Interestingly, we found that the genes differentially expressed after dfoxo-switch were significantly enriched for those affected by ageing (p=8.9×10^-12, one-sided hypergeometric test). For most, the direction of change induced by ageing is counteracted by dfoxo-switch (Fig. 6a). The transcriptional dysregulation that occurs with age is correlated to the transcriptional profile ensuing from Xbp1 loss-of-function (Fig. 6b), indicating a decline in the expression of Xbp1-target genes with age that is remedied by Xbp1 induction after dfoxo-switch.

Fig. 6. dfoxo-switch counteracts age-related transcriptional dysregulation.

a Relationship between log2Fold Change (FC) in gene expression caused by ageing and dfoxo-switch in Drosophila fat body, for genes differentially regulated by both dfoxo-switch and age. β = -8.23, p < 2.2 x10^-16, LM. b Relationship between FC in gene expression caused by ageing and xbp1 null mutant (xbp1^-/-), for genes differentially regulated by both xbp1^-/- and age. β = -8.64, p = 0.00589, LM. Points – individual genes; line – line of best fit, shaded – 95% CI for the line. c Overlaps of mouse orthologues of genes differentially expressed after dfoxo-switch in flies and genes differentially expressed with age in the mouse. Grey – mouse genes that show age-related expression changes, red – mouse orthologues of fly genes that are differentially expressed after dfoxo-swich in the fat body. P-values from one-sided hypergeometric tests.

As similar age-related changes can occur in different species^64,65, we identified mouse orthologues of the genes differentially expressed in the fat body after the dfoxo-switch. We examined if this set of genes is enriched for genes susceptible to age-related dysregulation, based on a recently published mouse ageing transcriptome study⁶⁵ and focusing on the organs equivalent to the fly fat body: mouse liver and adipose tissue. The expression of these orthologues was more likely to be susceptible to ageing generally (Fig. 6c), albeit without a predictable directionality (Extended data Fig. 9). Hence, the genes whose expression maintains a memory of past dfoxo induction tend to be dysregulated with age. In the fly fat body, the long-term programme initiated by dFOXO appears to counteract age-related dysregulation.

Discussion

We employed a tissue-specific, inducible system of in vivo dfoxo induction in the fruit fly to show that dFOXO activation in youth can trigger transcriptional memory to curtail later life mortality. FOXO TFs are at crossroads of multiple pathways signalling changes in external environment and the internal milieu²¹. Hence, alteration in FOXO activity may underly the risks and benefits of early life experiences, such as nutrition or exercise, that are increasingly documented to programme health and survival later in life^{5,11,16,17,66}. Understanding the mechanisms and relevance of transcriptional programming to animal physiology will have a profound impact on strategies ensuring human health throughout the life course.

Our study highlights the importance of chromatin architecture set in youth for later ageing. Firstly, we demonstrated that specific chromatin remodellers, those linked to histone acetylation (mi2, dom and tip60), are crucial in early adulthood to determine subsequent longevity in a wild-type female. Indeed, histone modifications, including histone acetylation, are recognised as important regulators of animal longevity^41,67–71. Our findings support and add to this as we found that even a short-time downregulation of these proteins in early life can dramatically curtail lifespan, regardless of their levels later in life. Secondly, we identified specific chromatin remodelling complexes required by dFOXO to extend lifespan, indicating that the activity of e.g. SWI/SNF and ISWI in youth promotes later longevity.

FOXOs are canonically considered as homeostatic TFs, allowing transient responses to a changing environment. The SWI/SNF complex is known to be essential for the worm FOXO orthologue, DAF-16, to extend lifespan⁴⁰ and for transcriptional memory of galactose exposure in yeast⁷. However, the link between the roles of SWI/SNF in transcriptional memory and lifespan had not been made. There is increasing evidence that FOXO TFs make long-lasting alterations to chromatin landscape, both via direct recruitment of chromatin organisers and indirect changes in their expression^16,72–74. Our study indicates that these combined features of FOXOs, homeostatic responsiveness and the ability to organise chromatin, result in their ability to orchestrate transcriptional memory with long-term consequences for animal physiology. This aspect of FOXO function is likely to be relevant to mammals.

FOXO TFs are well known regulators of glucose metabolism and metabolic gene expression⁷⁵. We did not find that manipulating the expression of carbohydrate metabolic enzymes could recapitulate the effects of dFOXO. Rather, our study implicates a dFOXO-to-Xbp1 transcriptional relay as an important component of longevity programming by dFOXO activation in youth: Xbp1 acts after dfoxo induction has ceased to curtail later-life mortality, most likely by counteracting the age-related loss of proteostasis, as documented in other contexts^59,60. Xbp1 transcriptional induction appears to be an indirect consequence of prior dFOXO activation in the fat body, likely as a secondary effect of primary transcriptional or metabolic changes. Interestingly, complex links between XBP1, metabolism, proteostasis and ageing are still emerging^76–79. Particularly, new roles of Xbp1 in lipid metabolism have been reported in Drosophila. Zhao and colleagues demonstrated that the Ire1/Xbp1 axis mobilises lipids in response to chronic starvation, and this effect is dependent on dFOXO inactivation⁷⁶. It will be interesting to understand in more detail how Xbp1 activation is impacted by dFOXO activity, chromatin structure and transcriptional and metabolic reprogramming.

Age-related changes in chromatin organisation are a key hallmark of ageing¹⁰. They are pervasive across species, cell types and timescales and are accompanied by transcriptional dysregulation³⁷. While this dysfunction in epigenetic and transcriptional control of gene expression has often been speculated as causal in ageing, direct evidence has only recently started emerging^71,80. Our work adds to this body of evidence that preserving or restoring a youthful epigenetic and transcriptional programme by a relatively short intervention promotes health and longevity, highlighting that such resetting can occur even before the visible onset of ageing.

Methods

Fly husbandry

The Dahomey wild-type stock was obtained in 1970 in what is today Benin and has been maintained in large population cages on a 12L:12D cycle at 25°C. The white Dahomey (wDah) and vermillion Dahomey (vDah) stocks were derived by incorporation of the w¹¹¹⁸ or v¹ mutation, respectively, into the Dahomey background by extensive backcrossing. wDah population used is Wolbachia-free and females from this background were used in all crosses to generate experimental flies. Mutants and transgenes were backcrossed into wDah (or vDah for BDSC TRiP lines) population for at least six generations, except for etl1, mi2, tip60, dom, nej and HP1 RNAi lines, UAS-Pepck1, UAS-Xbp1^RA and UAS-Xbp1^S Stock maintaince and experimental conditions: 25°C, a 12h:12h light/dark cycle, constant 60% humidity and standard sugar/yeast/agar (SYA) medium unless otherwise stated⁸¹. The chromatin complexes chosen for the epistasis experiments shown in Fig. 2 were selected under the FlyBase category Chromatin Remodelling Complexes (ATP-dependent), curated by FlyBase curators⁸². All fly stocks are listed in Supplementary Data.

Lifespan assays

Lifespan assays were performed as described in ref.⁸³. In brief, crosses were set up in cages containing grape juice agar and live yeast, flies were allowed to mate and then embryos were collected for < 22-hr, washed in PBS and seeded into bottles at ∼20 μl per bottle to achieve standard density^81,84. Females, allowed to mate for 48h, were used in experiments, unless otherwise noted. Flies were subsequently lightly anaesthetized with CO₂ and split into 15 per vial. Experimental vials contained either food with RU (RU+) at 200μM RU⁴⁸⁶ (Sigma, #M8046) or control food (RU-, containing the equivalent volume of EtOH vehicle). Vials were kept in DrosoFlippers (drosoflipper.com) and flies were transferred to fresh food three times a week, when deaths/censors were recorded. For the switch group, RU+ food was supplied from day 2 to day 23 and the flies were switched to control food for the rest of the experiment. The reverse switch group were supplied with RU+ food from day 23. Data were collected using Excel. Details of statistical analyses and number of flies per condition (n) are provided in figure legends or Supplementary Data. Log-rank tests of survivorship and Cox proportional hazards (CPH) analysis to obtain Hazard Rations (HR) were performed using R (R Core Team). Experimental trial was used as a random effect when required. GraphPad Prism 6 (GraphPad, La Jolla, CA) was used for graphic representation of survival curves. Lifespan data are provided in Supplementary Data.

Negative geotaxis (climbing) assays

Climbing assays were performed as described previously⁸⁵. In brief, at indicated times flies were transferred to empty vials placed in DrosoFlippers to allow climbing of 2 vial heights. The same cohort was continuously assayed. After acclimatising for 30 min, flies were gently tipped to the bottom of the vial and video recorded for 20 s. Video stills from the same time point (15 s; when young wild-type flies nearly reach maximum height) were analysed in Fiji⁸⁶ and fly coordinates exported to R (R Core Team). If the height could not be determined, the fly was not used in the analysis. Top and bottom 5% of data per condition/timepoint were excluded to protect from outliers. Data were analysed with a mixed effects LM (using flipper as random effect) and the two-sided F-test p values obtained with the anova() function in R were reported. Climbing data are provided in Supplementary Data.

Tunicamycin and starvation stress assays

Tunicamycin (2 mg l^-1 in DMSO, Cell Signaling) was added to sugar and agar food. In control treatments, equivalent volumes of the vehicle alone were added. For starvation assays, flies were placed in vials containing 1% agarose only. In both tunicamycin and starvation assays, 30 day-old, dfoxo-switch or control flies (1 week after recovery) were used. The vials were check for dead flies once or twice a day until no living flies remained. All statistical analyses were performed in R (R Core Team). Survival data are provided in Supplementary Data.

Cloning and generation of UAS-pepck1 fly line

Pepck1 coding sequence (CDS) was PCR amplified from wDah cDNA to include CACC preceding the ATG, was transferred to pENTR™/D-TOPO® vector using the pENTR™/D-TOPO® Cloning Kit and transformed into NEB® Stabl Competent E. coli (c30401). The correct clone was confirmed by DNA sequencing at Source Biosciences. Gateway LR reactions were conducted between the pENTR-pepck entry clone and pGW.AttB destination vector and transformed into library efficiency competent cells (Invitrogen #C404003). The UAS-pepck construct was confirmed by restriction enzyme digestion and sequencing. The construct were injected into embryos and integrated through the phiC31 integrase-mediated transgenesis⁸⁷ at the Department of Genetics Fly Facility (University of Cambridge) (https://www.flyfacility.gen.cam.ac.uk/Services/Microinjectionservice). phiC31 expressing stock nos-int; attP2 (Stock 13-18) was provided by the Cambridge fly facility.

Immunostaining

Fat bodies and guts were dissected from 1 week-old S106>nL8-GFP females kept for 5 days on RU+ or RU- food in ice-cold PBS and fixed with 4% paraformaldehyde for 20 min at room temperature. Tissues were washed 3 times in PBS + 0.1% Triton and blocked with 5% Bovine Serum Albumin for 2h. Incubations with the primary antibody (anti-GFP, 1:1000, 488 conjugate, Invitrogen) were performed overnight at 4°C. After mounting in Vectashield (Vector Laboratories), the samples were imaged using a Zeiss LSM AxioObserver confocal microscope (Zeiss, Oberkochen, Germany) and processed with Zeiss ZEN software and Adobe Illustrator. Details about the antibodies used in this study are provided in the Supplementary Data.

Western blotting

Midguts and fat bodies (associated with the abdominal cuticle) were dissected, transferred to 12.5% trichloroacetic acid, homogenized with glass beads (#G8772) and spun for 15 min at 14,000 rpm at 4°C. Pellets were washed twice with 1M Tris and the resulting pellet was resuspended in 50μl of LDS sample buffer (50% LDS sample buffer, 100 mM DTT, in nuclease-free water). Protein samples were then separated in 8% Poly-Acrylamide gels (Acrylamide/Bis-Acrylamide solution, Sigma A7168) following manufacturer’s instructions, and transferred to a nitrocellulose membrane. Membranes were blocked, incubated at 4°C overnight with primary antibodies (anti-dFOXO 1:5000, anti-Actin 1:1000) in 5% skimmed milk (Millipore #70166), washed and probed with secondary antibodies (1:10000) for 2h at room temperature. Details about the antibodies used in this study are provided in the Supplementary Data. Anti-dFOXO and anti-Actin were applied sequentially to the same membrane. Densitometric analysis of blot images was carried out using Fiji software⁸⁶.

RNA-Sequencing

Midguts and fat bodies (associated with the abdominal cuticle) were dissected in ice-cold PBS. Dissected tissues were placed directly into ice-cold Trizol (ThermoFisher Scientific #15596026). Five experimental replicates were obtained per condition, each containing twelve fat bodies or guts. RNA was extracted by Trizol-chloroform extraction, quantified on a NanoDrop 2000c spectrophotometer, and integrity was assessed on an Agilent Bioanalyzer. Poly(A) RNA was pulled down using NextFlex Poly(A) beads (PerkinElmer NOVA-512981). RNA fragments were given unique molecular identifiers and libraries were prepared for sequencing using NextFlex Rapid Directional qRNAseq v2 reagents, (barcode sets C and D, PerkinElmer NOVA-5130-14 and NOVA-5130-15) and 16 cycles of PCR. Individual and pooled library quality, size and molarity were assessed on an Agilent Bioanalyzer and quantified with a Qubit spectrophotometer. Libraries were pooled at equal molarity in NextFlex resuspension buffer. Due to poor RNA or library yield/quality, three samples were not sequenced. Sequencing was performed by the UCL Cancer Institute, using an Illumina NextSeq 500 instrument, with a single-end 75bp sequencing length.

ATAC-Sequencing

The ATAC-Seq protocol was adapted from ref.⁸⁸. Fat bodies and guts were dissected as for RNA-Seq. Three experimental replicates were obtained per condition, each containing forty fat bodies or guts (approximately 50k nuclei were employed for each replicate). After dissection, midguts were placed in 1ml of HB buffer (15mM Tris-HCl pH 7.4, 15mM NaCl, 1M KCl, 0.2mM EDTA pH 8, 0.2mM EGTA pH 7, with Roche complete mini protease inhibitor (#11836170001), in nuclease-free water). Mid guts were homogenised with a Dounce homogeniser on ice and filtered with a 20μm filter. Nuclei were spun at 3500g, 4°C for 5 min and washed with HB buffer twice. Then, nuclei were resuspended in 50μl of PBS. For the fat bodies, they were placed after dissection in 50μl of lysis buffer (10mM Tris-HCl ph 7.4, 10mM NaCl, 3mM gCl2, 0.1% IGEPAL CA-630). They were mixed by pipetting vigorously 10 times, and the supernatant was transferred to a new tube. Nuclei were spun at 3500g, 4°C for 5 min, and resuspended with lysis buffer. Then, nuclei were resuspended in 200μl of PBS. Nuclear DNA was tagmented by resuspending in Illumina buffer TD (from Illumina Nextera kit, kind gift from Richard Jenner and Maria Vila De Mucha, UCL Cancer Institute) with 1.6μg of Tn5 produced as described in ref.⁸⁹ to a total volume of 50μl. Reactions were incubated at 36°C for 30 minutes, pausing every 10 minutes for gentle manual agitation. DNA was eluted immediately using the Qiagen minelute PCR purification kit, and the whole eluate was amplified by PCR (10μl DNA eluate, 10μl H₂O, 2.5μl 25μM forward primer, 2.5μl 25 μM reverse primer (primer sequences are provided in Supplementary Data), 25 μl NEBNext Ultra II Q5 Master Mix (M0544). After an initial extension step (5 minutes at 72°C, 30 seconds at 98°C), tagmented DNA was amplified by 5 initial cycles of PCR (10s at 98°C, 30s at 63°C, 1min at 72°C). To determine the number of subsequent PCR cycles required to reach exponential phase per library, 5μl of the PCR reaction were amplified by qPCR (with 4.41μl nuclease-free H₂O, 0.25μl forward primer, 0.25μl reverse primer, 0.09μl 100x SYBR green, 5μl NEBNext Ultra II Q5 Master Mix; in a PCR program comprising 30s at 98°C followed by 20x cycles of 10 s at 98°C, 30 s at 63°C, 1 min at 72°C). Each initial PCR reaction was then cycled for the number of additional cycles required indicated by qPCR. DNA was eluted and size-selected using AMPure XP beads, quantified on a Qubit spectrophotometer, and library size and integrity were checked on an Agilent Bioanalyser. Individual library molarity was calculated from the Bioanalyser size estimate and Qubit density measurement, and libraries were pooled at equal molarity in NextFlex resuspension buffer.

qPCR

RNA samples were converted to cDNAs using SuperScript II Reverse Transcriptase (ThermoFisher Scientific, #18064014) and Oligo-dT. qPCR was performed on an Applied Biosystems QuantStudio 6 Flex real-time PCR instrument with Fast SYBR Green PCR Master Mix (Applied Biosciences #4385612), with primers supplied by Thermo Fisher (all primer sequences used in this study are shown in Supplementary Data), relative to a standard curve comprising a pool of all samples and the instrument’s standard PCR cycle. Actin was used for normalisation in expression analysis unless its quantity was suspected of changing with one of the experimental conditions, in which case tubulin or eIF1A were used instead. Mixed effects LM (dissection batch as random effect) was used in expression qPCR analysis, where each model was sequentially reduced removing all the insignificant terms and the two-sided F-test p values obtained with the anova() function reported.

ATAC-seq and RNA-seq data processing and analysis

The following workflow was applied for all the raw sequencing data [raw data have been deposited in Gene Expression Omnibus GEO (GSE183542)]. Read quality was assessed using FastQC. ATAC reads were trimmed using Trimmomatic version 0.33 to improve alignment rates. Both RNA-Seq and ATAC-Seq reads were aligned to the Drosophila melanogaster genome (r6.19) using HiSAT2, version 2.1.0. For RNA-Seq, alignments were counted with featureCounts over exons per gene. Preliminary principal component analysis of dfoxo-switch RNA-seq showed that one midgut and one fat body of S106>dfoxo +RU⁴⁸⁶ sample, originating from the same flies, clustered away from all other samples, and were removed from subsequent analyses. One further dfoxo-switch gut sample had low gene counts assigned (<10⁶) and was also removed. ATAC-Seq peaks were called with MACS2 version 2.1.2 for each tissue/treatment. The peaks were combined per tissue and per-peak counts for each sample were generated using featureCounts. Differential gene expression and differential ATAC signal intensities were assessed with DESeq2 and the ihw packages in R (R core team). The effect of RU⁴⁸⁶ feeding in the first three weeks followed by one week recovery was assessed in S106>dfoxo and S106 flies, separately. Gene Ontology (GO) term analysis was performed by topGO package in R⁹⁰, and KEGG pathway analysis was performed using DAVID⁹¹. The circos plot of peak distribution was generated by circlize in R⁹². The bar plots showing peaks distance related to TSS were generated by ChIPseeker in R⁹³. Other plots were generated using ggplot2 in R. Data associated with ATAC-Seq and RNA-Seq analyses, including analyses results, are given in Supplementary Data.

BETA analysis

BETA analysis was carried out in BETA 1.0.7⁴⁸ with default parameters. To explore the potential chromatin region-gene interactions comprehensively, differentially open ATAC-Seq peaks with a p-value < 0.05 (244 peaks) were regarded as binding data input. Expression data input include the dfoxo-switch fat body RNA-seq gene list, FDR 10% as threshold was applied to specify differentially expressed genes. The Drosophila melanogaster genome R6 (r6.19) was used for genome annotation.

RNA-seq and ChIP-seq analysis from public datasets

Biological information of the following RNA-seq datasets were obtained from GEO, and raw data (fastq files) were downloaded from Sequence Read Archive (SRA) employing SRA Toolkit (version 2.11). Datasets include Drosophila female fat body gene expression in different ages (GSE130158), gene expression of Xbp1 mutant flies at larval stage 2 (GSE99676). Raw gene counts data from mouse gene expression datasets obtained from different ages and tissues were obtained from the Tabula Muris Senis project’s website (https://twc-stanford.shinyapps.io/maca/). 5176 processed peak files of the Publicly available Drosophila ChIP-seq datasets following the uniform processing protocol were obtained from ChIP-Atlas³⁸. The annotations of each peak file were also obtained from ChIP-Altas.

Changes in gene expression with age analysis

To assess transcriptomic impacts of age, gene expression was modelled as a function of age as a continuous covariate in a linear model, fitted by DESeq2. Daily gene expression change was calculated from a previously published Drosophila female fat body gene expression dataset⁶³, and mouse weekly gene expression change was calculated from the dataset from⁶⁵, where the DESeq2 linear model was fitted in every combination of sexes and tissues. Comparison of gene expression list from different studies was performed within their DESeq2 results. Hypergeometric tests were used to quantify overlap significance between differential expressed gene lists. Linear regressions of Log2 fold change were used to estimate gene expression correlation.

Comparisons to published ChIP datasets and acute dFOXO transcriptome meta-analysis

To compute overlap between dfoxo-switch ATAC-peaks and publicly available Drosophila ChIP peaks, permutation tests (n = 10000), which generated random peaks based on given length and numbers, were performed in regioneR package⁹⁴ and FDR10% was used as the threshold for further analyses. Meta-analysis of the induced dfoxo gene expression was conducted using the metafor package in R⁹⁵ where random effect model with maximum likelihood (ML) estimators was used.

Motif analysis

TF binding motifs enriched in differentialyl expressed dfoxo-switch gene list (FDR 10%) were calculated using the R package RcisTarget⁹⁶. Motif-ranking dataset used the “dm6-5kb-upstream-full-tx-11species”. Motif-annotation dataset used the motifAnnotations_dmel embedded in the package. Functions were run using default settings.

Metabolite extraction

Five 30-day-old female flies (whole) were collected per replicate for metabolite extraction. 7 technical replicates were used for the RU-condition and 8 for RU switch condition. They were briefly anesthetised in CO₂ and snap-frozen in liquid nitrogen prior to metabolite extraction. Frozen flies were suspended in 200μL of Chloroform/Methanol (analytical grade)/Water (1:3:1 ratio) at 4°C containing glass beads and were homogenized at 4°C. Samples were then centrifuged for 3 minutes at 13,000g at 4°C and 180μL of supernatant was subtracted and stored at -80°C for further analysis by LC–MS. A pooled sample was generated by combining 20μl of each sample, to be used as a quality control sample in the LC–MS procedure.

LC/MS

LC/MS was performed by Glasgow Polyomics, as described before⁹⁷. Hydrophilic interaction liquid chromatography (HILIC) used a Dionex UltiMate 3000 RSLC system (Thermo Fisher Scientific, Hemel Hempstead, UK) with a ZIC-pHILIC column (150 mm × 4.6 mm, 5 μm column, Merck Sequant) maintained at 30°C. A linear gradient (20 mM ammonium carbonate in water, A and acetonitrile, B) was used to elute the samples over 24 min at a flow rate of 0.3 ml/min as follows: min 0: 20% A, 80% B, min 15: 80% A, 20% B, min 15: 95% A, 5% B, min 17: 95% A, 5% B, min 17 20% A, 80% B, min 26 20% A, 80% B. Injection volume was 10μl. The samples were kept at 5°C before injection. For the MS analysis, a Thermo Orbitrap QExactive (Thermo Fisher Scientific) was operated in polarity switching mode. MS settings were: Resolution 70,000, AGC 1e6. m/z range 70–1050, Sheath gas 40, Auxiliary gas 5 Sweep gas 1, Probe temperature 150°C, Capillary temperature 320°C. For positive mode ionisation: source voltage +3.8 kV, S-Lens RF Level 30.00, SLens Voltage -25.00 (V), Skimmer Voltage -15.00 (V), Inject Flatapole Offset -8.00 (V), Bent Flatapole DC -6.00 (V). For negative mode ionisation: source voltage-3.8 kV.

Metabolomics data analysis

Raw data was uploaded by Glasgow Polyomics to PiMP (http://polyomics.mvls.gla.ac.uk). Peaks Data used for subsequent analysis was generated by PIMP. Metabolite identification was performed in MetaboAnalyst 5.0⁹⁸ using the following parameters: Missing value estimation by K-nearest neighbour, filtering by standard deviation, sample normalization by sum, data transformation by log (generalized logarithm transformation or glog) and data scaling by autoscaling (mean-centered and divided by the standard deviation of each variable). Significant peaks were identified by t-test in MetaboAnalyst, and p-values were adjusted by Benjamini-Hochberg method (FDR) using R (R core team). Peaks with an FDR < 0.1 were assigned as significant. Peaks were assigned to their corresponding metabolites regarding their mass and retention time (RT). Metabolite annotation was assigned to putative metabolites to the signal based on matching their mass and RT with database or library entries. Metabolite identification was performed by direct comparison of the properties of an authentic standard ran in parallel. Metabolomics raw data, relevant metadata and protocols are available in the MetaboLights repository (ref.⁹⁹, study identifier MTBLS3251). Data and results tables from metabolomic analysis are given in Supplementary Data.

Statistics and reproducibility

No statistical method was used to predetermine sample size. Sample size selection was informed by and is consistent with those employed in previous research in the field^16,20,23,100. The investigators were not blinded to allocation during experiments and outcome assessment as one or few researcher(s) tended to perform the entire experiment and blinding was not practical. Treatment group allocation of experimental conditions was randomised in such a way as to avoid any potential confounding effects. Few data were excluded: 1) due to technical failure 2) two RNA-Seq samples based on principal component analysis (detailed above), 3) trimming of 5% top/bottom for climbing data to avoid outlier impact (detailed above), 4) censors from lifespan experiments due to accidental killing or escape (numbers reported). Statistical tests for each experiment are mentioned in their corresponding figure captions with additional detail provided in Methods. Adjustments for multiple testing were used for ‘omics data (RNA-Seq, ATAC-Seq, metabolomics) as reported. Normal distribution of residuals for LMs was confirmed by visual inspection; no formal testing of assumptions underlying the statistical tests was performed. All statistical analyses were performed in R or RStudio (R Core Team), versions 3.6.3 – 4.1.0.

Extended Data

Extended Data Fig. 1. Experimental setup – expression.

a Expression pattern of UAS-n8-GFP, driven with the S106 driver in 7 day-old female guts and fat bodies. All images were taken at exactly the same laser conditions under the confocal microscope, to allow the comparison of GFP levels between the two tissues. Scale bars are 50μm. b representative example of a S106 > n8-GFP fat body, induced or not with RU486, under different imaging conditions to (a), allowing a better view of GFP induction in fat body cells. Scale bars are 50μm. c Quantification of dfoxo mRNA levels in S106 alone and S106 UAS-dfoxo both uninduced. Boxplots – quantiles; whiskers – extremes; overlay – individual data points. N = 4, 3 (left to right), p = 1, unpaired two-sided t-test. d Uncropped images of the western-blot membranes shown in Fig.1c.

Extended Data Fig. 2. dfoxo-switch – additional lifespans, and climbing ability measurements.

a Driver-alone (S106) RU⁴⁸⁶-switch control lifespan. Control n = 150 dead/1 censored fly, switch n = 148 dead/0 censored flies, p = 0.69, log-rank test. b UAS-dfoxo alone RU⁴⁸⁶-switch control lifespan. Control n = 135 dead/2 censored flies, switch n = 136 dead/12 censored flies, p = 0.86, log-rank test. c Male driver-alone (S106) RU⁴⁸⁶- switch control lifespan. Control n = 126 dead/12 censored flies, RU⁴⁸⁶-switch n = 143 dead/6 censored flies, p = 0.49, log-rank test. d Male dfoxo-switch lifespan. Control n = 154 dead/6 censored flies, dfoxo-switch n = 142 dead/8 censored flies, p = 0.72, log-rank test. e TiGS>dfoxo-switch lifespan. Control n = 128 dead/3 censored flies, dfoxo-switch n = 141 dead/4 censored flies, p = 0.85, log-rank test. f S106>aop^Act-switch lifespan. Control n = 145 dead/7 censored flies, aop^Act-switch n = 147 dead/2 censored flies, p = 0.00274, log-rank test. g Experimental trials of the negative geotaxis assays of dfoxo-switch and control that were combined for the analysis presented in Fig. 1e. N = individual flies, boxplots – quantiles; whiskers – extremes; overlay – individual data points. Experiment 1: effect of dfoxo-switch p = 0.008, age p < 10^–4, age-by-dfoxo-switch interaction p = 0.0008, mixed-effects LM. Experiment 2: effect of dfoxo-switch p > 0.05, age p < 10^–4, age-by-dfoxo-switch interaction p = 0.033, mixed-effects LM. Experiment 3: effect of dfoxo-switch p > 0.05, age p < 10^–4, age-by-dfoxo-switch interaction p = 0.07, mixed-effects LM. h Experimental trials of S106>dfoxo-switch lifespans used for the analyses presented in Fig 1f. Experimental trial 1: control n = 61 dead/0 censored flies, dfoxo-switch n = 76 dead/0 censored flies, p = 0.009, log-rank test. Experimental trial 2: shown in Fig. 1d. Experimental trial 3: control n = 145 dead/1 censored fly, dfoxo-switch n = 145 dead/1 censored fly, p = 1.72189 x 10^-5, log-rank test. Experimental trial 4: control n = 125 dead/0 censored flies, dfoxo-switch n = 145 dead/2 censored flies, p = 6.05 x 10^-9, log-rank test.

Extended Data Fig. 3. ATAC-Seq – additional information.

a Schematic distribution of the ATAC peaks on the four main Drosophila chromosomes. Additional peaks were detected on contigs and are not shown. b Violin plots showing the size of all ATAC peaks detected in the gut and fat body, and in the significantly differentially accessible peaks in the fat body. c Distribution of the distance of ATAC peaks from a transcriptional start site (TSS). Proportion of peaks in each distance category are presented from 5’ to 3’ relative to the TSS. d ATAC-qPCR of the levels of sequences near the MED1 locus where an ATAC-Seq peak opened by dfoxo-switch was detected, near Prosap where a peak unaltered by dfoxo-switch was detected, and Sox21b region which contained no peak in ATAC-Seq. Boxplots – quantiles; whiskers – extremes; overlay – individual data points. N = 3 biologically independent samples; effect of dfoxo-switch: MED1 p=0.00024, Prosap p=0.59, Sox21b p=0.76, pairwise comparisons with two-sided unpaired t-tests with pooled SD. e ATAC-qPCR signal intensity for regions near the Xbp1 locus within a peak detected by ATAC-Seq (3’end of the gene) or two regions within the promoter of Xbp1. The levels were normalized to Prosap. Boxplots – quantiles; whiskers – extremes; overlay – individual data points. N = 3 biologically independent samples, effect of region p = 0.0011, effect of dfoxo-switch p > 0.05, mixed effects LM.

Extended Data Fig. 4. dfoxo-switch dependence on chromatin remodelers – lifespan.

Lifespan curves of the switch in S106>dfoxo & RNAi (a) or S106>RNAi (b) with indicated RNAi lines. These were used to generate the analysis shown in Fig. 2e. P values are obtained comparing control vs RU⁴⁸⁶-switch conditions (after day 23, log-rank test). Detailed statistical analyses including number of flies per experiment are shown in Supplementary Data.

Extended Data Fig. 5. dfoxo-switch dependence on chromatin remodelers - climbing ability.

a Negative geotaxis assay of dfoxo-switch + mor RNAi at all ages, combining two independent trials. Effect of the switch p = 0.6, age p < 10^–4, age-by-switch interaction p = 0.36, mixed-effects LM. b Negative geotaxis assay of dfoxo-switch + iswi RNAi at all ages. Effect of the switch p = 0.12, age p < 10^–4, age-by-switch interaction p = 0.14, mixed-effects LM. c Negative geotaxis assay of mor RNAi switch at all ages. Effect of the switch p = 0.02, age p < 10^–4, age-by-switch interaction p = 0.006, mixed-effects LM. d Negative geotaxis assays of iswi RNAi switch at all ages. Effect of the switch p = 0.62, age p < 10^–4, age-by-switch interaction p = 0.49, mixed-effects LM. N – individual flies, boxplots – quantiles; whiskers – extremes; overlay – individual data points. The negative geotaxis assays of dfoxo-switch that were performed at the same time are shown in Extended Data Fig. 2g, experiments 2 and 3.

Extended Data Fig. 6. RNA-Seq – additional information.

a qPCR quantification of transcripts detected as differentially expressed in our RNA-Seq data (HDAC6, Pfk, Pepck1) in fat bodies after dfoxo-switch. Effect of dfoxo p = 0.0055, transcript p = 0.0386, and dfoxo-switch-by-transcript interactions p = 0.032, mixed effects LM. b qPCR quantifications of dfoxo-switch targets (MED1, HDAC6, Xbp1^s, Pfrx) at week 7 in dfoxo-switch females. Effect of dfoxo-switch p = 0.0133, effect of transcript or dfoxo-switch-by-transcript interaction p > 0.05, mixed effects LM. N – biologically independent samples; boxplots – quantiles; whiskers – extremes; overlay – individual data points. c and d Overlaps of sets of differentially expressed genes between dfoxo-switch (red circles) and dfoxo-acute (grey circles) in the fat body and gut employing previously published gene lists. Overlap p-values from one-sided hypergeometric test. e Bar plot comparing the log₂ fold change of the transcripts in common between the sets of genes differentially regulated by dfoxo-acute (meta analysis) and by dfoxo-switch. f qPCR quantifications of transcripts during acute induction of dfoxo in the fat body. HDAC6, Pfk, Xbp1^s, Xbp1^u were examined as they are all differentially expressed after dfoxo-switch in the fat body (RNA-Seq analysis and qPCR confirmation shown elsewhere). N – biologically independent samples; boxplots – quantiles; whiskers – extremes; overlay – individual data points. Effects of RU⁴⁸⁶, transcript or RU⁴⁸⁶-by-transcript interaction p > 0.05, mixed effects LM. g Activating/Repressing Function prediction in BETA applying one-tailed Kolmogorov-Smirnov test. Differentially accessible ATAC peaks explained transcriptional activation (p=2.97x10^-46) and repression (p=0.028) _after dfoxo-switch.

Extended Data Fig. 7. Additional GO terms enrichment analysis.

Top 5 GO terms and KEGG pathways for genes differentially expressed (DE) a exclusively after dfoxo-switch and b exclusively during dfoxo-acute induction in the fat body. Note that no significant GO enrichment was observed in the set of genes that are differentially expressed in both dfoxo-switch and dfoxo-acute.

Extended Data Fig. 8. Involvement of Xbp1 in the effects of dfoxo-switch – additional information.

a qPCR quantification of transcripts whose levels are increased after dfoxo-switch in fat bodies (Hsc70-4, Eip75b, kay) in female S106>Xbp1^s fat bodies, with or without RU⁴⁸⁶ induction. Effect of RU⁴⁸⁶ p = 0.0088, effect of transcript or RU⁴⁸⁶-by-transcript interaction p > 0.05, mixed effects LM. b qPCR quantifications of the same transcripts in fat bodies after dfoxo-switch. Effect of RU⁴⁸⁶ p = 0.0001, effect of transcript or RU⁴⁸⁶- by-transcript interaction p > 0.05, mixed effects LM. N – biologically independent samples; boxplots – quantiles; whiskers – extremes; overlay – individual data points. c Survival of dfoxo-switch flies challenged with tunicamycin after 1 week of recovery (day 30; control n = 142 dead/0 censored, dfoxo-switch n = 129 dead/0 censored, p = 0.001, log-rank test). d Same for driver-alone (control n = 114 dead/0 censored, switch n = 131 dead/0 censored, p = 0.21, log-rank test). e Starvation assay of 30-day-old dfoxo-switch flies (control n = 142 dead/0 censored, dfoxo-switch n = 151 dead/0 censored, p = 0.67, log-rank test). f Same for driver-alone (control n = 149 dead/0 censored, switch n = 149 dead/0 censored, p = 0.07, log-rank test). g-j. Survival in the presence of tunicamycin a week after the switch in: S106>mor^RNAi (control n = 141 dead/1 censored, switch n = 154 dead/0 censored, p < 6x10^-7, log-rank test), S106>iswi^RNAi (control n = 146 dead/0 censored, switch n = 145 dead/0 censored, p = 0.60, log-rank test), S106>dfoxo mor^RNAi (control n = 140 dead/0 censored, switch n = 139 dead/0 censored, p < 8x10^-7, log-rank test), S106>dfoxo iswi^RNAi (control n = 153 dead/0 censored, switch n = 148 dead/0 censored, p = 0.079 log-rank test).

Extended Data Fig. 9. Age-related expression changes in the mouse – additional information.

Relationship between the expression changes triggered by dfoxo-switch in the fly fat body and the expression changes caused by ageing of their mouse orthologues (FDR 10%) in the functionally equivalent organs in the mouse. Points – genes; lines with shading – line of best fit and 95% CI; grey – those that are not significantly changed with age, red – those that are significantly changed with age. None of the organs show significant correlation between age-related change and dfoxo-switch change (p > 0.05, LM).

Data availability statement

Raw RNA-Seq and ATAC-Seq data are available from the Gene Expression Omnibus GEO (accession number GSE183542). Metabolomics data are available from the MetaboLights repository (study identifier MTBLS3251). All other data are available as Supplementary Data provided with the manuscript or can be made available by the corresponding author on reasonable request.

Publicly available data used in the study were:

• 1)Drosophila female fat body gene expression in different ages (GSE130158) and gene expression of Xbp1 mutant flies at larval stage 2 (GSE99676) both obtained from https://www.ncbi.nlm.nih.gov/geo/
• 2)Raw gene counts data from mouse gene expression datasets obtained from different ages and tissues were obtained from the Tabula Muris Senis project’s website https://twc-stanford.shinyapps.io/maca/
• 3)5176 processed peak files of the Publicly available Drosophila ChIP-seq datasets following the uniform processing protocolas well as the annotations of each peak file were obtained from https://chip-atlas.org/
• 4)Fly genome release and annotation were obtained from https://flybase.org/