Mondo complexes regulate TFEB via TOR inhibition to promote longevity in response to gonadal signals

Germline removal provokes longevity in several species and shifts resources towards survival and repair. Several Caenorhabditis elegans transcription factors regulate longevity arising from germline removal; yet, how they work together is unknown. Here we identify a Myc-like HLH transcription factor network comprised of Mondo/Max-like complex (MML-1/MXL-2) to be required for longevity induced by germline removal, as well as by reduced TOR, insulin/IGF signalling and mitochondrial function. Germline removal increases MML-1 nuclear accumulation and activity. Surprisingly, MML-1 regulates nuclear localization and activity of HLH-30/TFEB, a convergent regulator of autophagy, lysosome biogenesis and longevity, by downregulating TOR signalling via LARS-1/leucyl-transfer RNA synthase. HLH-30 also upregulates MML-1 upon germline removal. Mammalian MondoA/B and TFEB show similar mutual regulation. MML-1/MXL-2 and HLH-30 transcriptomes show both shared and preferential outputs including MDL-1/MAD-like HLH factor required for longevity. These studies reveal how an extensive interdependent HLH transcription factor network distributes responsibility and mutually enforces states geared towards reproduction or survival.

Of particular interest is the gonadal longevity pathway. Removal of germline precursors results in a prodigious 60% increase in lifespan 9 , which depends on bile acid-like steroids emanating from the somatic gonad [10][11][12] . Virtually, every known longevity factor including the steroid receptor DAF-12/FXR, DAF-16/FOXO, HSF-1, NHR-80/HNF4, PHA-4/FOXA and NHR-49/PPARa function in this pathway 4,9,[13][14][15] but how these factors cooperate in transcriptional networks is poorly understood. Dissecting such circuitry may reveal how regulatory functions are distributed and converge at key nodes within networks to establish signalling states and help pinpoint processes critical for longevity.
Although the major longevity pathways behave independently by genetic criteria, a number of convergent processes have emerged in common. This includes the process of autophagy, the intracellular engulfment of protein aggregates, defective organelles and internal membranes, with delivery to the lysosome for turnover. Autophagy is necessary for longevity in many pathways and, in some cases, enhanced autophagy is sufficient to extend life [16][17][18][19][20] . HLH-30/TFEB transcription factor, recently shown to promote autophagy and lysosomal biogenesis 21,22 , also works in various longevity pathways, revealing a key regulator of core longevity mechanisms 19 . However, the regulatory cascades that govern these core longevity mechanisms remain unexplored.
To uncover new regulators of the gonadal longevity pathway, we performed systematic RNA interference (RNAi) screens and discovered a key role for Myc superfamily members MML-1 (Myc/Mondo-like) and its partner MXL-2 (Max, Max-like) in mediating gonadal longevity, as well as longevity in other conserved pathways. Our studies suggest that MML-1 and MXL-2 work in transcriptional cascades that reduce TOR activity, which in turn stimulates autophagy and HLH-30/TFEB activity in the nucleus. Together, MML-1/MXL-2 and HLH-30 cooperate to extend life but also regulate distinct gene sets important for longevity. These studies illuminate a core regulatory network broadly affecting animal lifespan.
Results mml-1/mxl-2 are required for multiple longevity pathways. To identify new mediators of gonadal longevity, we performed RNAi screens encompassing ca. 600 transcription factors for suppressors of longevity in germline-less glp-1 mutants (Fig. 1a). We obtained previously known regulators such as daf-16, daf-12 and hlh-30, as well as several novel regulators including mml-1 and mxl-2 as potent candidates. MML-1 and MXL-2 belong to the Myc super family of basic helix-loop-helix (bHLH-Zip) E-box factors: MML-1 (Myc and Mondo-like) is homologous to MondoA/ChREBP, whereas MXL-2 is homologous to Max-like and works together with MML-1 in an activation complex 23 (Fig. 1b). In mammals, MondoA/Max-like and ChREBP (MondoB)/Max-like complexes respond to glucose and regulate glycolysis and lipogenesis 24,25 .
Demographic analysis confirmed that mml-1 and mxl-2 deletions abolished lifespan extension in glp-1 mutants and in animals whose germline precursors were removed by laser microsurgery, while only modestly shortening wild-type lifespan (Fig. 1c,d). We also examined interactions with other longevity pathways and found that mml-1 and mxl-2 were largely required for daf-2/InsR longevity (Fig. 1e), similar to previous reports 26 . In addition, mml-1 was partially required for isp-1 longevity but had little effect on longevity on cco-1 RNAi, whereas mxl-2 had little specific effect on either (Fig. 1f,g). Consistent with a role in longevity regulation, mml-1 overexpression in wild type sufficed to extend lifespan in three of six experiments (Supplementary Fig. 1a and Supplementary Table 1) but did not further extend glp-1 longevity (Supplementary Fig. 1b). These findings indicate that mml-1 and mxl-2 work within several longevity pathways and have overlapping but distinct functions.
Signals from the reproductive system regulate mml-1. A rescuing MML-1::GFP translational reporter (Supplementary Fig. 1c) was widely expressed and found in the cytoplasm and nuclei of the intestine, neurons, muscle, hypodermis, excretory cell and other tissues ( Supplementary Fig. 2a) 23 . A closer examination of the subcellular localization revealed that MML-1::GFP also co-localized with the mitochondria, comparable to mammalian MondoA (Fig. 2a) 27 . A rescuing MXL-2::GFP translational reporter construct was also widely expressed and overlapped with MML-1 in most tissues ( Fig.2a and Supplementary Figs 1d and 2b), but was distributed smoothly in the cytoplasm and nucleus, with only sporadic mitochondrial localization. Interestingly, we noticed that both mml-1 messenger RNA levels and MML-1::GFP nuclear accumulation were increased in glp-1 mutants (Fig. 2b-d). In contrast, MXL-2::GFP did not change on germline removal, although mxl-2 transcripts were increased ca. 1.5-fold ( Supplementary Fig. 2c-e). Thus, mml-1 is visibly regulated in response to germline signalling.
To gain further insight into activities of these transcription factors, we compared transcriptomes of wild type, glp-1, mml-1, mxl-2, hlh-30 and double mutants with glp-1 for differentially expressed genes (DEGs) measured by RNA-seq. We used a false discovery rate of 5% when considering DEGs. Hierarchical clustering of gene expression values revealed that biological replicates clustered well together, demonstrating the sample quality ( Supplementary Fig. 6a).  Table 4). For example, 827 genes (382 downregulated and 445 upregulated) were commonly regulated by mml-1 and mxl-2 in the glp-1 background. Each transcription factor also regulated large gene sets independent of the other, in particular mxl-2 deletion, which resulted in hundreds of genes being regulated differently from mml-1.
Gene Ontology (GO) term analysis of glp-1mml-1 versus glp-1 revealed an enrichment of genes implicated in oxidation/reduction, ageing, lipid modification, carbohydrate catabolic process, mitochondrial transport and amine biosynthetic pathway among others ( Fig. 6b and Supplementary Data 1). Kyoto Encyclopedia of Genes and Genomes (KEGG) process analysis also revealed enrichment in lysosomal function and fatty acid desaturation (Supplementary Data 2). mxl-2-dependent DEG in the glp-1 background included similar GO terms as glp-1mml-1 but also included GO terms specific to mxl-2 such as oocyte development, b-oxidation, regulation of translation and cell division among others (Supplementary Fig. 6c and Supplementary Data 1).

Discussion
Numerous transcriptional factors mediate the outputs of the major longevity pathways 1 ; yet, how they form coherent transcriptional regulatory networks is not understood. The gonadal longevity pathway is a rich multi-layered regulatory network, reflecting the intense evolutionary pressure to propagate an intact germline in the face of changing environments, nutrient conditions and physiologic milieus. In this work, we unravel a core regulatory node governing C. elegans gonadal longevity. We have found that the HLH Mondo/Max-like complex MML-1/MXL2 is required to promote longevity of germline-less animals. This complex also promotes longevity triggered by reduced TOR and insulin/IGF signalling, reduced mitochondrial respiration in some measure, as well as reportedly by dietary restriction 26 . Thus, MML-1/MXL-2 is among a handful of select factors working at the convergence of major longevity pathways. Most interestingly, we have discovered that the MML-1/MXL-2 complex dramatically stimulates nuclear localization and activity of the TFEB homologue HLH-30, a convergent regulator of autophagy, lysosome biogenesis, fat metabolism and longevity 19,28,29 , via TOR signalling. In germline-less animals, MML-1/MXL-2 transcriptionally repress the leucyl-tRNA synthase LARS-1, a positive effector of the TOR signalling pathway 34,35 , which results in HLH-30 nuclear localization and activity (see Model; Fig. 8f). HLH-30 itself promotes upregulation of MML-1 in the germline-less glp-1 background. Our findings expand the functional outputs of MML-1/MXL-2 complex as key regulators of autophagy, TOR and TFEB signalling. Importantly, we found that aspects of these regulatory relationships are evolutionarily conserved: mammalian homologues MondoA and ChREBP stimulate TFEB nuclear localization on amino acid starvation in cell culture. Thus, MML-1/Mondo and HLH-30/TFEB form a tight regulatory ensemble linked through TOR signalling, as well as presumably through E-box elements commonly bound by HLH transcription factors.
bHLH transcription factors regulate proliferation, growth, development and metabolism, and form an extensive interlocking network through heterodimerization and similar DNA-binding sites. The C. elegans HLH transcription factors have been assembled into an incipient network and Myc/Mondo complexes bind to similar E-box elements (CACGTG) as HLH-30/ TFEB 22,23,40 . Consistent with interlinked regulatory cascades, our transcriptome data indicate that MML-1, MXL-2 and HLH-30 substantially overlap in target gene expression, but also have preferential outputs. MML-1 and MXL-2 are well known to work in a complex 23,40 and accordingly show considerable overlap in their transcriptomes, tissue distribution and their mutant phenotypes. Surprisingly, more genes were upregulated than downregulated in the mutants, suggesting that this 'activation complex' may well have potent repressive activity. MXL-2 shows broader transcriptomic changes than MML-1, indicating participation in other complexes. Supporting the idea of independence, mml-1 and mxl-2 mutants exhibit somewhat different patterns of genetic epistasis, for example, mml-1 is required for isp-1 longevity, whereas mxl-2 is not. Perhaps this difference reflects our observed association of MML-1 with the mitochondria and the enrichment of mitochondria GO terms from mml-1 transcriptome analysis. MML-1 and HLH-30 also show significant transcriptional overlap but preferentially regulate distinct gene sets. Overlappingregulated processes include amino acid, fat, amine, nucleoside and xenobiotic metabolism, oxidation/reduction and others, possibly pinpointing processes important for longevity (Fig. 6). ARTICLE methionine restriction and H 2 S metabolism, which affect life span 41 . Another common striking output is tts-1, a long non-coding RNA that represses protein synthesis and affects multiple longevity pathways [36][37][38] . As mml-1 and hlh-30 represent convergent regulators of longevity, further functional dissection of their transcriptomes should illuminate core processes for lifespan extension. ModEncode data indicate that MML-1 and HLH-30 often reside at the same promoters 23,40,42 . Currently, it is not known whether the two factors bind independently or cooperatively, act in series as part of transcriptional cascades or work together in a complex. In addition, their temporal dynamics and whether they function in a tissue-or stage-specific manner remain to be elucidated. Future work to identify transcriptional complexes and their binding sites in the genome (chromatin immunoprecipitation sequencing) in the context of ageing should clarify these issues.
Although HLH-30 overexpression modestly extends lifespan 19 , our data indicate that HLH-30 activity is not sufficient to promote longevity in the gonadal pathway. Although knockdown of TOR signalling partially bypassed the mml-1/mxl-2 requirement for HLH-30 nuclear localization and activity in the glp-1 background, it was, surprisingly, unable to drive autophagy and longevity. This observation indicates that these factors act within partially overlapping networks, rather than a strict linear pathway, and that mml-1/mxl-2 leverages key processes for longevity. Accordingly, mml-1 preferential transcriptional targets fat-5, swt-1 and mdl-1 also facilitated longevity. Previously, Goudeau et al. 13 showed that fat-5 deletion does not affect glp-1 longevity, which is contradictory to our results ( Fig. 7b and  supplementary Fig. 7b). Although the exact cause for this difference is unknown, Goudeau et al. 13 used bacterial strain HT115, whereas we used OP50, suggesting that the food source may be responsible. mml-1 and hlh-30 also regulated the process of autophagy in common but preferentially affected distinct components: mml-1 mainly affected atg-2 and lgg-2, whereas hlh-30 affected atg-1 and lgg-1. Conceivably, other mml-1/mxl-2enriched processes such as glycerol or mitochondrial metabolism also contribute.
Our data suggest that MML-1 represents an ancestral MondoA/ ChREBP transcription factor conserved at many levels, which can inform mammalian metabolism and ageing. First, MML-1 and Mondo share common protein-domain structure including nuclear localization and glucose regulatory domains at the amino terminus and the bHLH-Zip domains at the carboxy terminus 23 . Second, mml-1 target genes are remarkably similar to those regulated by mammalian counterparts including genes involved in sugar transport (fgt-1), fatty acid desaturation (fat-5), pyruvate kinase (pyk-1), sterol regulatory binding protein (sbp-1) and glycerol-3phosphate dehydrogenase (gpdh-1) 24,27,43 . Whether C. elegans MML-1 senses glucose metabolites and has an impact on glycolysis and lipogenesis remain to be tested. Conversely, novel genes and processes regulated by C. elegans MML-1/MXL-2 (for example, autophagy, lysosomal and mitochondrial function) are plausible targets to further test in the mammalian system. Third, our work here reveals an evolutionarily conserved interrelationship between MONDO and TFEB. In response to germline removal, MML-1/MXL-2 promotes HLH-30 nuclear localization and activation; conversely, HLH-30 also upregulates MML-1 in germline-less animals. Similarly, MondoA and ChREBP facilitate TFEB nuclear localization and gene expression in response to amino acid starvation in cell culture, and TFEB knockdown reduced MondoA and ChREBP expression, revealing that the homologous transcription factors show mutual regulation. MondoA and ChREBP furthermore regulated TFEB gene targets, revealing a novel role in lysosomal function but, unexpectedly, affected distinct genes, perhaps reflecting their functional or temporal diversification. Other transcription factors working within the gonadal pathway show dependence: DAF-12/steroid receptor signalling facilitates DAF-16/FOXO nuclear localization 44,45 and MXL-2 stabilizes MML-1 (ref. 23). Speculatively, such positive reinforcement could stabilize a fragile interdependent network architecture to maintain coherent states geared towards reproduction or survival.
Fourth, MML-1/MXL-2 and MondoA/ChREBP function are intimately linked to TOR signalling. On the one hand, our data suggest that MML-1 suppresses TOR in germline-less animals; mml-1 loss-of-function hyperactivates TOR through upregulation of lars-1 and perhaps other factors, resulting in increased S6K activity, decreased autophagy and impaired nuclear localization of HLH-30. On the other hand, mml-1 loss often suppressed transcriptional changes and longevity arising from TOR knockdown. Thus, MML-1 may act both upstream, to transcriptionally repress TOR signalling, and downstream, to mediate the output of reduced TOR signalling. By inference, MML-1 may work in a feedback loop to limit TOR signalling. Transcription factors genetically regulated by C. elegans TOR are limited to DAF-16/ FOXO, SKN-1/NFE2, PHA-4/FOXA and HLH-30/TFEB 19,46,47 , but mechanistic links still remain obscure. Thus, the identification of MML-1/MXL-2 as potent suppressors of TOR-induced autophagy, longevity and gene expression reveals these transcription factors as critical outputs of this nutrient-sensing pathway.
Although MondoA and ChREBP are best known as glucose sensors that signal replete states, our studies suggest a potential role under nutrient limitation. Accordingly, Ayer and colleagues 48 have recently shown that mammalian MondoA suppresses mammalian TOR (mTOR) through upregulation of the TXNIP (thioredoxin-interacting protein), a regulator of glucose uptake and cellular redox signalling. Conversely, mTOR binds and sequesters MondoA activity in response to reactive oxygen species (ROS) and mTOR knockdown stimulates specific MondoA gene targets. MondoB/ChREBP also binds to mTOR. Whether MML-1/MondoA/ChREBP are direct targets of TOR phosphorylation is currently unknown. It is also unclear how germline removal triggers MML-1 activation: is it through glucose or related metabolites, TOR, ROS signalling or other signals? In this regard, it is intriguing that both MML-1 and MondoA are not only nuclear localized but are also present in the mitochondria, a site of ROS production 27 . Preliminary data also suggest that a fraction of MML-1 resides at the lysosome (S.N. and A.A., unpublished observation), a critical site of TOR action and TFEB localization 31,33 .
Finally, our studies highlight a convergent core role for Myc/Mondo complexes in the regulation of animal longevity. Previous work implicated mml-1/mxl-2 in IIS and dietary restriction pathways 26 ; our studies greatly expand this role to longevity induced by HLH-30, germline loss, decreased TOR signalling and, in part, mitochondrial longevity. In recent times, Sedivy and colleagues 49 showed that Myc þ / À heterozygote mice exhibit healthier lipid metabolism, increased physical activity, higher metabolic rate, less age-related pathologies and enhanced longevity. Molecular correlates include reduced IGF, increased AMPK, decreased AKT, mTOR and S6K activity, and protein synthesis. Evidently, various members of the Myc/Mondo HLH transcription factor family function to regulate metazoan lifespan through major growth factor and energyand nutrient-sensing pathways. It is striking that among target genes regulated by MML-1/MXL-2 is yet another member of the Myc subfamily, MDL-1, encoding the Mad-like homologue, which we found to be required for germline-less longevity, and whose overexpression is sufficient to extend life. We therefore suggest that Myc/Mondo HLH transcription factors comprise an extensive cascade regulating metabolism and longevity. Disentangling their distributed and shared responsibilities, as well as cross-regulation, and understanding how these interactions specifically impact metabolism and ageing will be fascinating to explore.

Methods
C. elegans growth conditions. Nematodes were cultured using standard techniques at 20°C on nematode growth medium agar plates with the Escherichia coli strain OP50, unless otherwise noted. Strains with glp-1(e2141) background were maintained at 15 coli°C and grown at 25°C to induce germline-deficient phenotype. All strains used in this study are listed in Supplementary Table 2.
Lifespan assays. Synchronized eggs were obtained by 4-6 colih egg lay. When they reached day 1 adults, lifespan experiments were set up at a density of 15-20 animals per plate and carried out at 20°C. These animals were transferred to new plates every other day. Survivorships of worms were also counted every other day. Death was scored as the absence of any movement after stimulation by a platinum wire. For lifespan experiments using isp-1(qm150) strains, egg lays were conducted 3 days before that of other strains, to obtain day 1 adults at the same time. Worms undergoing internal hatching, bursting vulva or crawling off the plates were censored. For all experiments, strains and/or conditions were blinded. For the ablation of germline precursor cells, newly hatched L1 worms were anaesthetized in 0.5% phenoxypropanol in M9 buffer. Z2/Z3 germline precursor cells were ablated by laser microsurgery (Micropoint) and successful ablations were confirmed under the dissecting microscope at day1 adult stage. The controls (mock ablated) were also anaesthetized in parallel with experimental animals and included for lifespan assay. For lifespan experiments using the glp-1 strain, all strains were raised at 25°C for 2 days, to induce the glp-1 phenotype after the egg lay at 15°C. Subsequently, lifespan experiments were resumed at 20°C. In some cases, we observed a relatively long median lifespan of N2 under these conditions, presumably due to hormesis effects of the temperature shift. For all lifespan experiments, strains and/or conditions were blinded. Statistical analyses were performed with the Mantel-Cox log rank method in Excel (Microsoft).
RNA interference. RNAi was conducted by feeding HT115 (DE3) bacteria transformed with L4440 vector that produces double-stranded RNA against the targeted gene. Synchronized eggs obtained by egg lay were put on corresponding RNAi plates containing isopropyl-b-D-thiogalactoside and ampicillin. RNAi clones were available from the Ahringer or Vidal RNAi library. let-363/TOR RNAi clone was a gift from Dr Hansen (Sanford-Burnham Medical Research Institute). Empty vector (L4440) or Luciferase (L4440::Luc) RNAi were used as non-targeting control. glp-1(e2141) suppressor screening was conducted using transcription factor RNAi library 7 . glp-1(e2141) worms were synchronized by tight egg lay and treated by control RNAi and RNAi targeting transcription factors. The survivorship at day 25 was used as an estimate of longevity. RNAi suppressor screening was replicated twice and only the candidates that show reproducible results were taken into consideration for further analysis. For all genes that are targeted by RNAi in this study, we indicated 'i' after the gene name.
RNA extraction and qRT-PCR. Worms and cell samples were harvested in TRIzol (Invitrogen). Total RNA was extracted using RNeasy or miRNAeasy kit (QIAGEN). Complementary DNA was generated using iScript (Bio-Rad). Quantitative reverse transcriptase-PCR (qRT-PCR) was performed with Power SYBER Green (Applied Biosystems) on a ViiA 7 Real-Time PCR System (Applied Biosystems). Four technical replicates were performed in each reaction. ama-1 and/or cdc-42 (for worms), and GAPDH (for cells) were used as an internal control. Primer sequences are listed in Supplementary Table 3.
Mitochondrial staining. Synchronized eggs were incubated on nematode growth medium plates containing 500 nM MitoTracker Deep Red (Molecular Probe) for 2 days. The worms were washed by M9 and mounted on agarose pads. Pictures were taken by Leica confocal microscopy SP5 X. The deconvolution was conducted using Huygens Professional (Scientific Volume Imaging), to examine localization of MML-1::GFP and MXL-2::GFP to mitochondria.
RNA sequencing. For each genotype, more than 300 day 1 adult worms were collected in Trizol in three independent biological replicates. One microgram of of RNA was extracted using miRNAeasy Mini Kit (QIAGEN). polyA þ mRNA was isolated from 500 ng total RNA with NEBNext Poly(A) mRNA Magnetics Isolation Module (New England Biolabs). RNA-seq libraries were prepared with the NEBNext Ultra Directional RNA Library Prep Kit for Illumina (New England Biolabs). Quality and quantity was assessed at all steps by capillary electrophoresis (Agilent Bioanalyser and Agilent Tapestation). Libraries were quantified by fluorometry, immobilized and processed onto a flow cell with a cBot (Illumina) followed by sequencing-by-synthesis with TruSeq v3 chemistry on a HiSeq2500 at the Max Planck Genome Center (Cologne, Germany). Reads were trimmed for adapter and barcodes using the Flexbar version 2.5 software 51 . Alignment of the reads was done using the Tophat version 2.0.13 software against the Wormbase genome (WBcel235_79). The Tuxedo/cufflinks version 2.2.1 software pipeline was used to perform differential gene expression analysis on pairwise comparisons of the different samples. Dispersion was calculated per condition (genotypes) and differential expressed genes (q-valueo0.05) of each pairwise comparison were identified. GO annotation and enrichment analysis was performed using the DAVID bioinformatics resources database.
Cell culture experiments. HeLa (P5-P10) cells (American Type Culture Collection, USA; ATCC-CCL-2) were maintained in DMEM medium (Gibco), supplemented with 10% fetal bovine serum (Gibco). Transfections of MondoA (Santa Cruz Biotechnology), ChREBP (Santa Cruz Biotechnology) and TFEB (GE Dharmacon) siRNA with respective control siRNAs were performed using INTERFERin (Polyplus Transfection) according to the manufacturer's instruction. For TFEB nuclear localization assay, cells were starved in the starvation medium (140 mM NaCl, 1 mM CaCl 2 , 1 mM MgCl 2 , 5 mM glucose, 1% BSA and 20 mM HEPES, pH 7.4) 52 for 4 h (for immunohistochemistry and western blotting) and fixed by 4% paraformaldehyde followed by MeOH fixation. Immunohistochemistry was performed using anti-TFEB antibody (1/500, rabbit or goat, Cell Signaling, #4240 or Abcam ab2636, respectively) followed by anti-rabbit or anti-goat secondary Alexa 488 antibody (1/500, donkey, ThermoFisher Scientific, A-11055) incubation. For each experimental condition, more than 100 cells were scored for their TFEB nuclear localization in a blinded manner and in at least 3 different biological replicates. For each biological replicate, the quantity of MondoA or ChREBP gene knockdown was measured by western blottings. For western blotting, cells were collected in RIPA buffer (Cell Signaling) with cOmplete protease