Decreased spliceosome fidelity and egl-8 intron retention inhibit mTORC1 signaling to promote longevity

Changes in splicing fidelity are associated with loss of homeostasis and aging, yet only a handful of splicing factors have been shown to be causally required to promote longevity, and the underlying mechanisms and downstream targets in these paradigms remain elusive. Surprisingly, we found a hypomorphic mutation within ribonucleoprotein RNP-6/poly(U)-binding factor 60 kDa (PUF60), a spliceosome component promoting weak 3′-splice site recognition, which causes aberrant splicing, elevates stress responses and enhances longevity in Caenorhabditis elegans. Through genetic suppressor screens, we identify a gain-of-function mutation within rbm-39, an RNP-6-interacting splicing factor, which increases nuclear speckle formation, alleviates splicing defects and curtails longevity caused by rnp-6 mutation. By leveraging the splicing changes induced by RNP-6/RBM-39 activities, we uncover intron retention in egl-8/phospholipase C β4 (PLCB4) as a key splicing target prolonging life. Genetic and biochemical evidence show that neuronal RNP-6/EGL-8 downregulates mammalian target of rapamycin complex 1 (mTORC1) signaling to control organismal lifespan. In mammalian cells, PUF60 downregulation also potently and specifically inhibits mTORC1 signaling. Altogether, our results reveal that splicing fidelity modulates lifespan through mTOR signaling.

cellular function of rnp-6, we tagged endogenous rnp-6 with green fluorescent protein (GFP) using clustered regularly interspaced short palindromic repeats (CRISPR)-Cas9. Consistent with its role as an essential splicing factor, rnp-6(wt) was ubiquitously expressed in all examined tissues and mainly localized in the nucleus (Extended Data Fig. 1c). GFP-tagged endogenous rnp-6(G281D) showed a similar expression pattern, but was present at significantly lower levels ( Fig. 1c and Extended Data Fig. 1d), and further validated by western blotting (Extended Data Fig. 1e), suggesting a reduction of function. RNA-sequencing (RNA-seq) analysis also showed that rnp-6(G281D) caused changes in mRNA processing and transcription similar to, but not as extensive as, rnp-6i, including alternative splicing, intron retention and circular RNA formation (Fig. 1d,e, Extended Data Fig. 1f-i and Supplementary Tables 1 and 3), as well as differential gene expression (Extended Data Fig. 1j and Supplementary Table 4), confirming that rnp-6(G281D) represents a reduction-of-function mutation. Notably, ~80% of the differentially expressed genes (DEGs) (1,142 of 1,366 genes in rnp-6(G281D) and 3,730 of 4,707 genes in rnp-6i) were upregulated (Supplementary Table 4). Gene ontology (GO) analysis of the DEGs showed that stress response was among the most enriched physiological categories in both rnp-6(G281D) and rnp-6i (Extended Data Fig. 1k and Supplementary Table 5), revealing that impaired spliceosome function triggers cellular stress responses.
Next, we asked whether rnp-6i mimicked rnp-6(G281D) longevity. To bypass developmental defects, we performed rnp-6i during adulthood. Whereas rnp-6i from day 1 adults (coinciding with the onset of reproduction) decreased the lifespan (Extended Data Fig. 1l), rnp-6i initiated from day 4 adults (coinciding near the end of reproduction) onward significantly extended it (Fig. 1f). This happens in either the presence or the absence of fluoro-2′deoxyuridine-5′-phosphate (FUDR; an inhibitor of DNA synthesis commonly used in C. elegans lifespan experiments to prevent eggs from hatching), suggesting that lifespan extension is independent of progeny production (Extended Data Fig. 1m). These results reveal that, like several other essential genes 14,15 , rnp-6 shows antagonistic pleiotropy; knockdown can be detrimental early in life but beneficial later and imply that the fine-tuning of rnp-6 activity is critical for longevity.
In addition, we found that the rnp-6(E161K) intragenic mutation was also a potent suppressor of rnp-6(G281D), which fully restored all measured phenotypes to wild-type levels (Extended Data Fig. 4c-e). It also significantly suppressed mRNA processing as well as transcriptional changes (Extended Data Fig. 4f-k and Supplementary Tables 10-13), confirming that, like rbm-39 mutation, restoration of splicing correlates with reversal of phenotype.
Egl-8 intron retention contributes to rnp-6(G281D) longevity. To decipher the downstream mechanisms by which the RNP-6/RBM-39 complex regulates longevity, we focused on splicing events. In particular, intron retention is an important but not well-understood mechanism of gene expression regulation 23 . It is mostly associated with downregulation of gene expression via nonsense-mediated decay 24 and has recently emerged as an important splicing feature in both normal aging and longevity interventions 4,25,26 . To reveal functionally relevant targets for the RNP-6/RBM-39 complex, we focused on intron retention induced by rnp-6(G281D) and restored by rbm-39(S294L). We narrowed down the list of candidates to 44 events by cross-referencing with the rnp-6(E161K) revertant and manual curation in the genome browser (Supplementary Table 14). These 44 events correspond to 42 genes and, notably, all showed increased intron retention in rnp-6(G281D). We performed RNAi  knockdown to assess the impact on wild-type lifespan, reasoning that both RNAi and intron retention should result in partial loss of function. Of those genes tested, we found one candidate, egl-8, with a knockdown that yielded significant lifespan extension in wildtype, but did not further extend rnp-6(G281D) longevity (Fig. 3a). We further confirmed this genetic interaction with an egl-8(n488) null allele (Fig. 3b). However, the egl-8 null did not recapitulate rnp-6(G281D) cold tolerance (Extended Data Fig. 4l), suggesting separatable mechanisms for cold tolerance and longevity. Egl-8 encodes an ortholog of human PLCB4. It plays vital physiological roles in neurotransmission 27,28 , lifespan and infection response in C. elegans 29,30 , although the underlying molecular mechanisms are not well understood. RNA-seq data showed that rnp-6(G281D) altered the retention of several introns within egl-8 ( Fig. 3c and Extended Data Fig. 5a) without decreasing the total egl-8 mRNA level (Extended Data Fig. 5b). Importantly, both rbm-39(S294L) and rnp-6(E161K) suppressed intron 8 retention (Extended Data Fig. 5a,c). Reverse transcription (RT)-PCR experiments also validated these results (Fig. 3d,e and Extended Data Fig. 5e-g). Intron 8 harbors a weak noncanonical splice acceptor site (Extended Data Fig. 6a) and its retention introduces a premature stop codon in the transcript (Extended Data Fig. 6b), which could either result in mRNA degradation by nonsense-mediated mRNA decay or give rise to a nonfunctional truncated protein. To examine its expression, we tagged endogenous EGL-8 with mNeonGreen at the N terminus. The mNeonGreen-tagged EGL-8 was mainly detected in head neurons as well as intestinal adherens junctions (Extended Data Fig. 6c), in agreement with previous immunofluorescence staining results 27 . As expected, the expression levels of EGL-8 in neurons and the nerve ring were significantly lower in rnp-6(G281D) compared with that in wild-type controls, and rbm-39(S294L) efficiently restored EGL-8 expression levels back to wild-type ( Fig. 3f,g), consistent with a restoration of intron removal. Notably, neural EGL-8 expression levels were similarly reduced by approximately 20% in both rnp-6(G281D) and egl-8i (Extended Data Fig. 6d), in line with their similar impact on longevity. Furthermore, neuronal expression of rnp-6 + or the fully spliced egl-8 + complementary DNA suppressed rnp-6(G281D) longevity (Fig. 3h,i). To examine egl-8 intron retention in neurons, we designed primers to detect neuronal egl-8 transcripts based on published tissue-specific RNA-seq data 31 (Extended Data Fig. 6e,f). RT-PCR results showed that rnp-6(G281D) caused more pronounced changes in neuronal egl-8 intron retention and rbm-39(S294L) completely restored the defects (Extended Data Fig. 6g,h). These findings are consistent with the idea that rnp-6(G281D) promotes longevity via intron retention of egl-8 within the nervous system. In addition, intestinal expression of rnp-6 also partially rescued the rnp-6(G281D) lifespan (Extended Data Fig. 6i), indicating that the gut also contributes to rnp-6-mediated longevity.
To further test whether egl-8 intron retention is sufficient for lifespan extension, we generated two T-A cis-acting mutations that weakened the intron 8 3′-splice acceptor site, using the intron consensus sequence as reference (Extended Data Figs. 6a and 7a). Although these nucleotide substitutions were not as potent as rnp-6(G281D) in disrupting splicing, they nevertheless caused a modest but significant increase in intron 8 retention (Extended Data Fig. 7b,c) and extended lifespan (Fig. 3j). Taken together, these data indicate that egl-8 intron retention contributes to lifespan extension.
PUF60 regulates mammalian mTORC1 signaling. Last, we sought to understand whether the functional interaction of RNP-6 and mTORC1/RAGA-1 was evolutionarily conserved. To this end, we knocked down PUF60 by small interfering (si)RNA in HEK293FT cells, which show several characteristics that are normally observed in immature neurons 41 and have been extensively used in studies on the mechanisms of amino acid sensing by mTORC1 (ref. 32 ). As rnp-6 regulates mTORC1 upstream of raga-1 (Fig. 4g,h), we surmised that PUF60 might affect amino acid signaling to mTORC1. In accord with this view, PUF60 knockdown decreased mTORC1 reactivation on amino acid re-supplementation, assayed by the phosphorylation of its direct substrates S6K (ribosomal protein S6 kinase β1) and TFEB (Fig. 5a), without influencing mTORC2 activity (assayed by Akt phosphorylation) (Extended Data Fig. 9). Accordingly, we consistently observed decreased RAPTOR (regulatory-associated protein of mTOR) protein levels on PUF60 knockdown, whereas the levels of the respective mTORC2 core component, RICTOR, or of mTOR itself, were largely unaffected ( Fig. 5a and Extended Data Fig. 9). In line with the C. elegans results, and further supporting decreased mTORC1 activity, PUF60 knockdown enhanced the nuclear localization of transcription factor E3 (TFE3) (Fig. 5b,c). As amino acid sufficiency controls mTORC1 localization to lysosomes via promoting RAPTOR binding to the lysosomal Rag GTPase dimers, we then hypothesized that PUF60 may regulate mTORC1 activity by influencing its subcellular localization. Indeed, knocking down PUF60 caused a significant drop in the colocalization of mTOR with the lysosomal marker LAMP2 (lysosome-associated membrane glycoprotein 2) in cells re-supplemented with amino acids (Fig. 5d,e). These findings reveal that PUF60 acts as a specific and integral part of the mTORC1 signaling pathway, influencing the amino acid-induced activation of mTORC1 at the lysosomal surface. Whether the mechanistic details of mTORC1 regulation by PUF60 in mammalian cells are the same as those in worms remains to be seen in future studies.

Discussion
Messenger RNA splicing is a fundamental cellular process which has recently emerged as important to organismal aging. Although specific splicing factors and splicing events have been shown to be associated with the aging process 42,43 , the underlying molecular mechanisms remain largely unknown. Our studies provide direct evidence that a genetic mutation within a core spliceosome component promotes longevity partially through intron retention in C. elegans, highlighting such events in the fidelity of information processing and stress response.
Rnp-6 encodes an essential splicing factor and the null mutation causes lethality in C. elegans. However, rnp-6(G281D) shows no overt defects under standard growth conditions at 20 °C. Our RNAi, rescue and RNA-seq experiments indicate that rnp-6(G281D) is a unique change or selective loss-of-function allele, which is similar to, yet distinct from, canonical functional depletion by knockdown.
Notably, rnp-6(G281D) mutants are long-lived. In comparison, rnp-6i during development and from D1 of adulthood onward impair essential functions and shorten lifespan, whereas rnp-6i from D4 of adulthood promotes longevity. In contrast to rnp-6 RNAi, mTOR RNAi extends life from D1 onward 44 . These findings imply that the activities of splicing factors are fine-tuned, and suggest an antagonistic pleiotropic trade-off that is beneficial early on but detrimental later in life, similar to other essential genes of autophagy 14,15 and deubiquitination 45 in C. elegans. Whether rnp-6(G281D) and D4 on rnp-6i work by the same precise mechanism is currently unknown. Splicing factors form extraordinarily large and highly dynamic macromolecular assemblies to catalyze splicing. Similar to transcription factors, there are both positive and inhibitory regulators of this process 1 . From our study, we identified a lesion in splicing factor RBM-39(Ser294Leu), which forms nuclear speckles, alleviates RNP-6(Gly281Asp) defects and reverses longevity phenotypes, giving us mechanistic insight into critical targets of splicing. It is interesting that in mammalian cells RBM39, PUF60 and the large subunit of the U2AF complex, U2AF65, share a similar domain architecture and work in close proximity to regulate 3′-splice site assembly. Notably, all three proteins interact with the U2 snRNP subunit, SF3b155, through their UHM domains, and U2AF65 and RBM39 have been shown to do so cooperatively 46 . Conceivably, RBM-39(Ser294Leu) ameliorates RNP-6(Gly281Asp) by altering interactions with other splicing factors within nuclear speckles. Whether RBM39 and PUF60 bind to SF3b155 in a cooperative or competitive manner remains to be seen, but their close proximity suggests possible modes of functional interaction and highlights the importance of this specific complex in stress signaling and lifespan control. Intron retention is a major mechanism of gene expression regulation 23 and has recently been shown as a common feature in agingrelated splicing changes 25 . Intron retention is also associated with longevity interventions, such as diet restriction, in both C. elegans and mice 4 . Several lines of functional evidence provide direct causal evidence that egl-8 intron retention in neurons contributes to organismal longevity. First, rnp-6(G281D) mutation causes egl-8 intron retention, reduced neural EGL-8 protein levels and longevity, in a manner restored by rbm-39(S294L). Furthermore, rnp-6(G281D) induces a level of egl-8 downregulation comparable to that of egl-8 RNAi, which similarly extends life. Second, cis-acting mutations that weaken the egl-8 intron 8 3′-splice site are sufficient to promote both intron retention and longevity, despite a smaller magnitude effect to rnp-6(G281D) itself. Third, neuronal expression of fulllength egl-8 cDNA is sufficient to suppress rnp-6(G281D) longevity, although gut-specific egl-8 expression reportedly has little impact on lifespan 29,30 . By contrast, expression of rnp-6 in gut can rescue lifespan (Extended Data Fig. 6i), suggesting that egl-8-independent mechanisms contribute to rnp-6(G281D) longevity as well. In sum, we conclude that rnp-6(G281D) longevity arises at least in part from egl-8 intron retention, although other activities probably contribute.
We would like to point out that the observed ratio of intron 8 retention as measured by RNA-seq was relatively small, ranging from 1.5% in wild-type to 4% in rnp-6(G281D) (Extended Data Fig.  5a), raising some question about its significance. However, because of technical limitations, we obtained RNA from whole worms and thus the relevant isoforms in functionally important tissues are underestimated. Notably, egl-8 mRNA has a highly complex, alternative splicing pattern and is expressed in multiple tissues. Despite this, the egl-8 cDNA containing all exons when expressed in neurons can regulate lifespan throughout the body (Fig. 5h and previous reports 29,30 ). In contrast to RNA-seq, the RT-PCR experiments showed greater levels of intron retention ( Fig. 3e and Extended Data Fig. 6h). This apparent enrichment reflects the use of primers annealing to exon 9, which is present in the full-length cDNA used to rescue lifespan, but not in all isoforms (Extended Data Fig. 6e). In addition, it is possible that processes downstream of intron retention, such as reduced translation, may also impact expression 47,48 . Notably, it is not uncommon for relatively small changes in gene expression to influence phenotype in other settings 49,50 .
Our study reveals that RNP-6 and PUF60 share an evolutionarily conserved role in regulating mTORC1 signaling, but it is not yet clear if this regulation also happens via PLCB4 in mammalian cells. This will require additional follow-up studies and the outcome could depend on cell type or signaling context. The question arises as to whether RNP-6/PUF60 regulates intron retention in a normal physiological context of cellular signaling. In accord with this idea, we also found that egl-8 intron retention is induced under conditions of food deprivation in wild-type animals within the adult reproductive diapause 51,52 , which is reversed by re-feeding (Extended Data Fig. 10a,b), resembling the amino acid starvation/ re-supplementation protocols used in cell culture assays to study the reactivation of mTORC1. Conceivably, external environmental cues or internal physiological signals may trigger intron retention events to modulate the host response. An interesting question is whether the impact of rnp-6/rbm-39 on egl-8 and mTORC1 signaling represents a specific signaling pathway or a broader stress response in which egl-8 serves as a sentinel for aberrant splicing.
In summary, our results suggest a model whereby the RNP-6/ RBM-39 spliceosomal complex impacts splicing fidelity to regulate mTOR signaling and longevity (Extended Data Fig. 10c). Our study implicating components of the splicing machinery working upstream of mTOR signaling may provide new approaches to manipulate this pathway in aging, metabolism and disease. Precise targeting of PUF60, and perhaps RBM39, could be used to downregulate mTORC1 signaling to confer health benefits similar to rapamycin and other rapalogs 53,54 . Conversely, as many spliceosomopathies that reduce spliceosomal function trigger growth defects 8,55-58 , it may be possible to treat these diseases with mTOR modulators.

Plasmid construction and transgenesis.
For rnp-6 rescue plasmid, rnp-6 promoter (3,135 bp) was amplified from the N2 genome and inserted into a pDC4 vector-generated rnp-6p::FLAG::HA::GFP::::unc-54 3′-UTR construct. Then, rnp-6b cDNA was amplified from N2 cDNA and cloned into this plasmid to generate a rnp-6p::FLAG::HA::GFP::rnp-6b cDNA::unc-54 3′-UTR rescue plasmid. Sitedirected mutagenesis (Q5 Site-Directed Mutagenesis Kit, New England Biolabs (NEB)) was performed to incorporate Gly281Asp point mutation to generate rnp-6p::FLAG::HA::GFP::rnp-6b(G281D) cDNA::unc-54 3′-UTR plasmid. To generate neuronal rescue plasmid, rnp-6 promoter was replaced by neuronalspecific promoter rgef-1 (2,670 bp). The unc-17p::gfp::egl-8 plasmid is a kind gift from S. Nurrish (Harvard Medical School). The microinjection experiments were performed according to a standard protocol 61 : 10 ng μl −1 of plasmid of interest together with a 5-ng μl −1 of co-injection marker plasmid (myo-3p::mCherry) were injected into the gonads of young adult-stage worms. Positive offspring were singled to maintain stable lines. PCR primers related to these plasmids are available in Supplementary Table 15. EMS mutagenesis screen and mapping. The cold resistance-longevity screen that identified rnp-6(G281D) had been performed previously 13 . The heat-sensitivity suppressor screen was done on an rnp-6(G281D) mutant background. Briefly, ~1,000 L4 larvae worms (P0) were exposed to 0.15% ethyl methane sulfonate (EMS, Sigma-Aldrich) in M9 buffer for 4 h at room temperature. After overnight recovery, young adult P0 animals were transferred to new plates for egg laying at 20 °C. After 3 d of growing, adult F1 worms were bleached and eggs were seeded on nematode growth medium (NGM) plates and incubated at 25 °C. After 3 d, F2 worms that reached adulthood were singled and maintained at 20 °C. Falsepositive hits were excluded by re-testing heat sensitivity in F4/F5 generations. The rnp-6(G281D) animals were used as a negative control in all heat-sensitivity assays. Hawaiian SNP mapping and whole-genome sequence were used to map the causative mutation 62 . The rnp-6(G281D) mutation was first introduced to Hawaiian CB4856 by 6× outcrossing. Then, the EMS mutants were crossed with Hawaiian males that carry the rnp-6(G281D) mutation. Eggs of the F1 generation worms were grown at 25 °C and adult F2 were singled after 3 d. The heat-resistant strains were then pooled together, and genomic DNA was purified using Gentra Puregene Kit (QIAGEN). The pooled DNA was sequenced on an Illumina HiSeq platform (paired-end 150 nt). An MiModD pipeline (https://celegans.biologie.uni-freiburg. de/?page_id=917) was used to identify the mutations. The WS220/ce10 C. elegans assembly was used as reference genome for annotation. The causative mutations were confirmed by either CRISPR-Cas9 or multiple outcross. Dh1183 and dh1187 were identified as rbm-39(S294L) and rnp-6(E161K), respectively. The causative genes for the other 11 mutants remain unclear.
Stress tolerance assays. For cold tolerance assays, worms were synchronized and grown on standard NGM plates. When the worms reached the young adult stage, they were transferred to a 2 °C incubator for 24 h. The worms were recovered at room temperature for 4 h and the number of alive and dead worms were scored. The cold survival ratio was measured as the ratio of the number of live worms to the number of total worms. At least 60 worms from each genotype were used in the assay for each biological replicate. Three independent repeats were performed. For the heat-sensitivity assay, synchronized eggs from different genotypes were grown on standard NGM plates at 20 °C or 25 °C. After 3 d, images of the worms were captured. Body length or body area was measured by ImageJ software. At least 15 worms from each genotype were used for each biological replicate. Three independent repeats were performed except in Extended Data Fig. 2a,c in which two repeats were performed.

Protein alignments and phylogenetic analysis.
Homologs of RNP-6 and RBM-39 were identified from Wormbase (https://wormbase.org). A T-Coffee algorithm 63 was used to align RNP-6, RBM-39 and their homologs from different species. Phylogenetic analysis of RBM-39 and its homologs was done with Phylogeny.fr 64
Lifespan. All lifespans were performed at 20 °C unless otherwise noted. Worms were allowed to grow to the young adult stage on standard NGM plates with OP50. For each genotype, ~150 young adults were transferred to NGM plates with OP50 supplemented with 10 µM FUDR. For lifespan experiments with egl-8(n488), a final concentration of 50 µM FUDR was added to NGM plates. Survival was monitored every other day. Worms that did not respond to gentle touch by a worm pick were scored as dead and were removed from the plates. Animals that crawled off the plate or had ruptured vulva phenotypes were censored. All lifespan experiments were blinded and performed at least 3× independently unless otherwise noted. Graphpad Prism (9.0.0) was used to plot survival curves. Survival curves were compared and P values were calculated using the log(rank) (Mantel-Cox) analysis method. Complete lifespan data are available in Supplementary Table 16.

Infection assay. Staphylococcus aureus (strain MW2) was grown in Tryptic Soy
Broth medium at 37 °C with gentle shaking overnight. Then, 100 μl of the bacterial culture was seeded and spread all over the surface of the trypticase soy agar plate with 10 μg ml −1 of nalidixic acid. The plates were allowed to grow overnight at 37 °C. On the next day, the plates were left at room temperature for at least 6 h before the infection experiments. Around 25 synchronized young adult worms were transferred to the plates. Three technical replicate plates were set up for each condition. Worms were treated with 100 μM FUDR from L4 stage to prevent internal hatching during experiments. The plates were then incubated at 25 °C to initiate the infection experiment. Scoring was performed every day. Worms were scored as dead if the animals did not respond to gentle touch by a worm pick. Worms that crawled off the plates or had ruptured vulva phenotypes were censored from the analysis. All C. elegans killing assays were performed 3× independently unless otherwise stated. At least 60 worms per genotype were included at the start of the assay for each replicate. Genotypes were blinded for all C. elegans infection survival experiments to eliminate any investigator-induced bias. Results of each biological replicate of infection survival experiments can be found in Supplementary Table 17.

RNAi in C. elegans.
RNAi experiments were performed as previously described 13 . E. coli HT115 and E. coli OP50(xu363) bacterial strains were used in the present study. The HT115 bacteria were from the Vidal or Ahringer library. The OP50(xu363) competent bacteria were transformed with double-stranded RNA expression plasmids, which were extracted from the respective HT115 bacterial strains. The RNAi bacteria were grown in lysogeny broth medium supplemented with 100 µg ml −1 of ampicillin at 37 °C overnight with gentle shaking. The culture was spread on RNAi plates, which are NGM plates containing 100 µg ml −1 of ampicillin and 0.4 mM isopropyl β-d-1-thiogalactopyranoside. RNAi-expressing bacteria were allowed to grow on the plates at room temperature for 2 d. RNAi was initiated by letting the animals feed on the desired RNAi bacteria. RNAi experiments related to Fig. 1d-f and Extended Data Fig. 2g,h were done with OP50(xu363) bacteria. For RNAi lifespan experiments related to Fig. 3a,m,n, worms were grown on HT115 RNAi bacteria from the egg until day 1 adulthood and then transferred to NGM plates seeded with OP50.
RNA extraction and cDNA synthesis. C. elegans was lysed with QIAzol Lysis Reagent. RNA was extracted using chloroform extraction. The samples were then purified using RNeasy Mini Kit (QIAGEN). Purity and concentration of the RNA samples were assessed using a NanoDrop 2000c (peqLab). The cDNA synthesis was performed using an iScript cDNA synthesis kit (BioRad). Standard manufacturers' protocols were followed for all mentioned commercial kits.

RNA-seq and bioinformatic analysis.
Total RNA, 1 µg, was used per sample for library preparation. The protocol of Illumina Tru-Seq-stranded RiboZero was used for RNA preparation. After purification and validation (2200 TapeStation, Agilent Technologies), libraries were pooled for quantification using the KAPA Library Quantification kit (Peqlab) and the 7900HT Sequence Detection System (Applied Biosystems). The libraries were then sequenced with the Illumina HiSeq4000 sequencing system using paired-end 2× 100-bp sequencing protocol. For data analysis, the Wormbase genome (WBcel235_89) was used for alignment of the reads. This was performed with Hisat v.2.0.4 (ref. 66 ). DEGs between different samples were identified using Stringtie (v.1.3.0) 67 , followed by Cufflinks (v.2.2) 68 . The enrichment visualization was performed with WormCat 2.0 (ref. 69 ). P values were calculated from Fisher's exact tests and adjusted with Bonferroni's multiple hypothesis test. P < 0.05 was defined as significant. An SAJR pipeline 70 was used for splicing analysis. Significant splicing changes were defined as those with P < 0.05 after adjusting for multiple testing using the Benjamini-Hochberg correction. For intron retention analysis, Bedtools coverage (v.2.29.0) was used to count intron and total gene expression. IBB (v.20.06; R v.4.0.3) 71 was used to calculate differential intron expression. DCC/CircTest (v.0.1.0) pipeline 72 was performed to quantify circular RNA expression. Weblogo 73 was used to generate intron 3′-splice site consensus sequence logos (Extended Data Fig. 6a). A total of 9,484 3′-splice site sequences from SAJR analysis were used. Row z-score heat maps (Fig. 2k-m) were generated by using the iHeatmap function from Flaski (v.2.0.0) (https://doi. org/10.5281/zenodo.5254193). Adjusted P value/q value <0.05 is considered to be significant for SAJR and DEG analysis; P < 0.001 is considered to be significant for intron retention analysis.
Co-immunoprecipitation. Worms expressing HA::RNP-6, RBM-39::mKate2 or both were harvested and proteins were extracted using the following standard protocol 74 . A solubilization buffer containing 0.5% NP40, 150 mM NaCl and 50 mM Tris, pH 7.4 supplemented with cOmplete Protease Inhibitor (Roche) and PhosSTOP (Roche) was used for immunoprecipitation. Flag immunoprecipitation was performed using Dynabeads Protein G (Thermo Fisher Scientific) and FLAG M2 mouse monoclonal antibody (Sigma-Aldrich), following the manufacturers' protocols. Proteins were eluted from the beads by boiling with Laemmli buffer.
Worm imaging. Analysis of worm reporters GFP::RNP-6, RBM-39::mKate2, mNeonGreen::EGL-8 and HLH-30::mNeonGreen was performed on a Zeiss Axioplan2 microscope (Axio Vision SE64, Rel.4.9.1) with a Zeiss AxioCam 506 CCD camera. Analysis of worm size was performed on a Leica stereo microscope (Leica M165 FC, LAS X) with Leica DFC3000G CCD. Fiji software (v.2.0.0/1.52p) 75 was used for quantifying fluorescence intensity. For mNeonGreen::EGL-8 images, the head neuron region was selected for quantification. For HLH-30::mNeonGreen images, the nuclei of hypodermal cells were selected for quantification. For GFP::RNP-6 images, the whole worm was selected for quantification. To reduce bias, individual worms were randomly picked under a dissection microscope and imaged. At least 20 worms per genotype were picked for imaging and all the experiments were done 3× independently. Data from a representative experiment are shown in the figures for all the imaging panels. HEK293FT cells (siRNA transfections). Transient knockdowns were performed using a pool of four small interfering GENOME siRNAs (Horizon Discoveries) against PUF60, whereas an RLuc duplex siRNA that targets the Rotylenchulus reniformis luciferase gene (Horizon Discoveries) was used as control. In brief, HEK293FT cells were seeded in 12-well plates at 20% confluence and the following day transfected with 20 nM of the siRNA pool using Lipofectamine RNAiMAX (Thermo Fisher Scientific) according to the manufacturer's instructions. Cells were harvested or fixed 72 h post-transfection and knockdown efficiency was verified by western blotting.

Immunofluorescence and confocal microscopy in HEK293FT cells.
Ιmmunofluorescence/confocal microscopy experiments and quantification of colocalization were performed as previously described 60 . In brief, cells were seeded on fibronectin-coated coverslips and treated as indicated in each experiment. After treatment, cells were fixed for 10 min at room temperature with 4% paraformaldehyde in PBS. Samples were washed/permeabilized with PBT solution (1× PBS-T), and blocked with BBT solution (1× PBS-T, 0.1% BSA). Staining was performed with the indicated primary antibodies in BBT (1:200 dilution) for 2 h at room temperature for mTOR and LAMP2 staining or overnight at 4 °C for TFE3 staining. Next, samples were washed 4× with BBT (15 min each), followed by incubation with appropriate highly cross-adsorbed, secondary fluorescent antibodies (1:200 in BBT) for 1 h at room temperature (Supplementary Table 18). Finally, nuclei were stained with DAPI and cells mounted on slides using Fluoromount-G (Invitrogen, catalog no. 00-4958-02). Images from single-channel captures are shown in grayscale. For the merged images, Alexa 488 is shown in green, TRITC in red and DAPI in blue. Images were captured using a ×40 objective lens on an SP8 Leica confocal microscope (Leica Application Suite X 3.5.7.23225). To quantify colocalization of mTOR signal with the lysosomal marker LAMP2, the Fiji software (v.2.1.0/1.53c) 75 was used to define regions of interest corresponding to individual cells, excluding the nucleus. Fifty individual cells from five independent fields per condition were selected for the analysis. The Coloc2 plugin was used to calculate the Manders' colocalization coefficient (MCC), using automatic Costes' thresholding 76,77 . The MCC yields the fraction of the mTOR signal that overlaps with the LAMP2 signal. Subcellular localization of TFE3 was analyzed by scoring cells based on the signal distribution of TFE3, as shown in the example images in Fig. 5c. Signal was scored as nuclear (more TFE3 signal in the nucleus) or cytoplasmic (similar TFE3 signal between nucleus and cytoplasm). Cells from five independent fields, containing approximately 70 individual cells, were scored per condition for each experiment. Data from a representative experiment out of three independent replicates are shown in the figures for all confocal microscopy panels.
Statistics and reproducibility. In all figure legends, 'n' denotes the number of independent experiments performed, whereas 'N' indicates the total number of animals analyzed in each condition. All statistical analyses were performed in GraphPad Prism (v.9.0.0 (86)). Asterisks denote corresponding statistical significance: * P < 0.05; ** P < 0.01; *** P < 0.001; **** P < 0.0001. Data distribution was assumed to be normal but this was not formally tested. No statistical method was used to predetermine sample size but our sample sizes are similar to those reported in our previous publications 13,52,78,79 . No data were excluded from the analyses. At least three independent experiments for each assay were performed to verify the reproducibility of the findings (if there were two independent experiments, this was also noted in the figure legend). For the worm experiments, sample preparations and data collection were randomized. For lifespan experiments, all the genotypes and RNAi treatments were blinded. For cold tolerance, developmental rate, body area, infection, western blotting and imaging experiments, the genotypes were not blinded before assay because mutant worms have obvious phenotypes that revealed the sample identity (body size and developmental rate). However, worms were randomly picked and assigned to the different treatment conditions and the different conditions were assessed in random order. For RNA-seq experiments, the genotypes were not blinded before collecting samples. Once the RNA samples were ready, they were processed by staff at the Cologne Centre for Genomics in a blinded manner. For mammalian cell studies, no blinding was included in the data collection or analysis, because the method of quantification over multiple replicates and individual cells (for microscopy experiments) ensures unbiased processing.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All RNA-seq datasets generated and analyzed in the present study are available in the Gene Expression Omnibus datasets with the accession no. PRJNA757629. For protein alignment and phylogenetic analysis, all sequences are accessible through the UniProt database (https://www.uniprot.org) with the UniProt ID no. All other data are available as Source data files or from the corresponding author upon reasonable request.