A Feedforward Loop Governs The Relationship Between Lipid Metabolism And Longevity

The relationship between lipid metabolism and longevity remains unclear. In particular although fat oxidation is essential for weight loss, whether it is remains beneficial when sustained for long periods, and the extent to which it may alter lifespan remains an important unanswered question. Here we develop an experimental handle in the C. elegans model system, that uncovers the mechanisms that connect long-term fat oxidation with longevity. We find that sustained □-oxidation, via activation of the conserved triglyceride lipase ATGL-1, triggers a feedforward transcriptional loop that involves the mito-nuclear transcription factor ATFS-1, and a previously unknown and highly conserved repressor of ATGL-1 called HLH-11/AP4. This feedforward loop orchestrates the dual control of fat oxidation and lifespan protection, shielding the organism from life-shortening mitochondrial stress in the face of continuous fat oxidation. Thus, we uncover one mechanism by which fat oxidation can be sustained for long periods without deleterious effects on longevity.


INTRODUCTION
The relationship between lipid metabolism, adiposity and lifespan remains unclear. Support for this assertion comes from several sources. On the one hand, accumulation of lipid deposits in ectopic tissues is associated with age-associated illnesses including diabetes, hepatic and pancreatic steatoses and other metabolic illnesses that shorten lifespan (Conte et al., 2019a;Schmeisser et al., 2019;Shulman, 2014). Converting the excess lipid deposits to energy via a cascade of metabolic reactions: lipolysis of triglyceride lipids, fatty acid breakdown via betaoxidation and increased electron transport chain activity in the mitochondria, have been proposed to benefit health and longevity (Bonawitz et al., 2007;Conte et al., 2019b;Vatner et al., 2018). In this context, increased metabolic activity and mitochondrial respiration increase lifespan (Speakman et al., 2004). On the other hand, long-term caloric restriction and decreased metabolic rate reduce electron transport chain activity in the mitochondria and increase lifespan (Burkewitz et al., 2015;Durieux et al., 2011;Mattison et al., 2017). This paradoxical observation, first made in rats (McCay et al., 1935) and then extended to several species including C. elegans (Kimura et al., 1997;Lakowski and Hekimi, 1998) and primates (Mattison et al., 2017;Redman and Ravussin, 2011), argues for the opposite: that decreased metabolic and mitochondrial activity increases longevity.
Additional data supporting a non-linear relationship between lipid metabolism, fatty acid oxidation and longevity come from studies of the systemic insulin and TGF-beta signaling pathways, which are master regulators of physiology in C. elegans. Loss of the insulin/insulinlike growth factor (IGF-1) receptor daf-2, or of the transforming growth factor-β (TGFβ) signal daf-7, each result in substantially increased adiposity and body fat stores, but with a concomitant increase in lifespan (Greer et al., 2008;Kimura et al., 1997;Ogg et al., 1997). This non-linearity between adiposity and longevity is echoed in human observational studies that repeatedly suggest either no relationship (Kuk and Ardern, 2009;Reynolds et al., 2005), or an inverse association between adiposity and mortality in late life (Fontaine et al., 2003;Stevens et al., 1998;Zheng and Dirlam, 2016). A resolution to this dichotomy might stem from considering the nature of experimental interventions. The majority of C. elegans studies showing a relationship between reduced mitochondrial function and increased longevity come from manipulations that decrease fat oxidation and metabolic rate (Dillin et al., 2002;Durieux et al., 2011;Lee et al., 2003). However, the converse has not been tested. In other words, the consequences of a sustained increase in fat oxidation on longevity in non-disease states has remained unexplored. Furthermore, fundamental mechanisms that connect lipid metabolism with longevity regulation at an organismal level still remain poorly understood. To this end, we considered it fruitful to conduct an investigation into the effects of sustained fat loss and increased mitochondrial respiration on longevity.
The neuromodulator serotonin (5-hydroxytryptamine; 5-HT) is a major regulator of metabolism, behavior and physiology in many species. In C. elegans, 5-HT is synthesized by the rate-limiting enzyme tryptophan hydroxylase (TPH-1) in only a few head neurons (Sze et al., 2000) but has wide-ranging effects across the organism including lipid metabolism (Srinivasan et al., 2008) behavioral responses to food (Cunningham et al., 2012), reproduction (Tanis et al., 2008), and pathogen avoidance (Zhang et al., 2005). 5-HT is a powerful stimulator of fat loss in the intestine (Noble et al., 2013), wherein the majority of lipids are stored and metabolized (Srinivasan, 2015). 5-HT-elicited fat loss occurs by activating the mitochondrial beta-oxidation pathway (Srinivasan et al., 2008), in which stored triglyceride lipids are oxidized to usable energy (Salway, 1999). Thus, approaches that increase neuronal 5-HT would have the potential to serve as a system to test the effects of sustained fat oxidation on lifespan. A major caveat is that genetic or pharmacological approaches that globally augment or decrease serotonin signaling lack specificity, leading to confounding and counterregulatory behavioral and metabolic effects that become difficult to disentangle. In this regard, an experimental approach FLP-7 transgenic line represents a potent experimental tool for us to address, in a non-disease context, the question of whether neuronally-driven sustained fat oxidation has an effect on longevity.

An experimental handle to generate sustained fat oxidation in the intestine
The FLP-7/NPR-22 tachykinin neuron-to-intestine signaling pathway is triggered by increases in neuronal 5-HT ( Figures 1A-C), as previously published (Palamiuc et al., 2017). We tested several measures to ascertain the extent to which the FLP-7 transgenic line (henceforth flp-7 tg ) recapitulates the effects of neuronal 5-HT signaling on metabolic parameters in the intestine.
First, we observed that flp-7 tg worms have a significant reduction in intestinal fat stores (Figures 1D,E), recapitulating that seen with 5-HT treatment ( Figures 1F, G). The fat reduction elicited by flp-7 tg is dependent on the presence of the intestinal triglyceride lipase ATGL-1 (Figures 1D, E; note that the atgl-1 null mutant is inviable), as was noted for 5-HT itself (Figures 1F, G) (Noble et al., 2013;Palamiuc et al., 2017). In previous work, we uncovered major components of the mitochondrial beta-oxidation pathway that functionally connect ATGL-1 activity to the electron transport chain (ETC) that underlies mitochondrial respiration (Noble et al., 2013;Srinivasan et al., 2008). Accordingly, FLP-7-stimulated fat loss is accompanied by increased basal and maximal respiration that are both abrogated upon RNAi-mediated inactivation of atgl-1 ( Figures 1H, I). Together, Figure 1 shows that increased mitochondrial respiration in flp-7 tg animals results from conversion of stored triglycerides to energy via beta-oxidation and ETC activity. Thus, the flp-7 tg line fully recapitulates the metabolic effects of genetic and pharmacological manipulation of neuronal 5-HT signaling without altering other 5-HT-mediated behaviors (Horvitz et al., 1982;Loer and Kenyon, 1993;Palamiuc et al., 2017;Song and Avery, 2012;Sze et al., 2000;Waggoner et al., 1998) and can serve as a valuable experimental handle to examine the long-term effects of sustained fat oxidation and increased ETC activity.

Intestinal fat oxidation via ATGL-1 induction evokes a mitochondrial stress response
Several lines of evidence have suggested that mild reductions in mitochondrial respiration can lead to a sustained increase in lifespan via decreased ROS production, hormesis, or other mechanisms. We examined flp-7 tg animals to test whether the observed increase in mitochondrial respiration (Figures 1H, I) might evoke the opposite effect on longevity. However, we found that flp-7 tg animals had nearly the same lifespan as wild-type animals, with a nonsignificant difference (p=0.08 by Log-rank Test) in survival probability statistics (Figures 2A, B).
Next, we considered whether the increased mitochondrial respiration (Figures 1H, I) might evoke a systemic stress response. To test this possibility, we examined the effects of 5-HT administration or FLP-7 secretion on a variety of well-established stress reporters in C. elegans including those involved in the cytoplasmic heat shock response (hsp-70 and hsp-16.2), oxidative stress (sod-3), nutrient stress (DAF-16 nuclear localization), ER stress (hsp-4) and mitochondrial stress (hsp-60). In the flp-7 tg animals (Figures 2C-E) and in those treated with 5-HT ( Figure S1), we observed a two-fold induction of an hsp-60::GFP transgene, the canonical reporter for the mitochondrial stress response. Other stress responses were absent. The induced mitochondrial hsp-60-mediated stress response was seen predominantly in the intestine ( Figure 2C), and was wholly dependent on the presence of atgl-1 because RNAimediated inactivation of atgl-1 ameliorated hsp-60 induction as judged by the reporter assay ( Figure 2D) as well as by directly measuring hsp-60 transcripts by qPCR ( Figure 2E). The increased mitochondrial respiration in flp-7 tg also depended on atgl-1-dependent utilization of fat reserves ( Figures 1H, I), suggesting the surprising result that fat oxidation evokes a mitochondrial stress response.

The mitochondrial stress response factor ATFS-1 sustains fat oxidation via ATGL-1
All known mitochondrial stress pathways that induce hsp-60 require the mito-nuclear transcription factor called ATFS-1 (Nargund et al., 2012). Under normal conditions, ATFS-1 is transported into mitochondria and degraded. However, under conditions that induce mitochondrial stress, ATFS-1 is stabilized and translocated to the nucleus where it initiates a range of transcriptional responses that allow adaptation to the stressor (Lin and Haynes, 2016;Nargund et al., 2015). To determine whether the FLP-7-induced mitochondrial stress response required atfs-1, we measured hsp-60 induction in its absence. As shown in Figures 2F and G, hsp-60 was no longer induced by FLP-7 in the atfs-1(tm4525) null mutants. We also noted that the combined loss of atgl-1 and atfs-1 completely suppressed the hsp-60 induction, which recapitulates the loss of each gene alone with no additive or synergistic effects (Figures 2F,G).
This finding was replicated with 5-HT-stimulated hsp-60 induction that was completely suppressed with either atgl-1 or atfs-1 inactivation ( Figure S1). This result suggested the possibility that atgl-1-dependent fat oxidation and atfs-1-dependent mitochondrial stress may function in a linear pathway.
Despite our knowledge about the role of ATFS-1 in mitochondrial biology, it had not previously been associated with fat oxidation per se. To our surprise, we found that both 5-HT and FLP-7-induced fat oxidation were partially suppressed in atfs-1 null mutants ( Figures 3A, B and Figure S2A). The absence of atfs-1 also suppressed the transcriptional induction of atgl-1 by FLP-7 and 5-HT as judged by measuring the atgl-1 reporter in vivo (Figures 3C,D), and by direct measurement of atgl-1 transcripts by qPCR ( Figures 3E and S2B). As in the case of the hsp-60-mediated stress response ( Figures 2D, E), loss of both atfs-1 and atgl-1 also did not lead to a further suppression of FLP-7-induced fat loss ( Figures 3A, B), suggesting again that they function in a linear pathway. However, these data also indicated that the effects of secreted FLP-7 on the intestine are intertwined: on the one hand the mitochondrial stress response requires atgl-1 and fat oxidation (red arrow; Figure 3F), and on the other hand fat oxidation requires the mito-nuclear stress response transcription factor atfs-1 (green arrow; Figure 3F).
Published ChIPseq and microarray studies of the ATFS-1 transcription factor did not suggest a direct induction of atgl-1 by ATFS-1 (Nargund et al., 2015;Nargund et al., 2012). We also did not find the putative cis-binding site of ATFS-1 within a 5 kb region upstream of the atgl-1 transcriptional start site. Yet, ATFS-1 is required for both the induction of atgl-1 expression ( Figures 3C-E), and the ensuing fat loss ( Figures 3A, B). Thus, ATFS-1 likely regulates atgl-1 transcription by an indirect mechanism.
The conserved transcription factor HLH-11 governs fat oxidation via direct control of

ATGL-1
To identify direct transcriptional regulators of atgl-1 in vivo, we began by conducting an RNAibased screen of the ~ 900 transcription factors in C. elegans (Fuxman Bass et al., 2016). We used the Patgl-1::GFP reporter we had previously developed (Noble et al., 2013) and screened for genes that regulate atgl-1 under basal conditions, as well as for those that were essential for 5-HT or FLP-7-stimulated atgl-1 induction. Our top hit was a gene called hlh-11, the sole conserved ortholog of the mammalian transcription factor AP4 (Lee et al., 2009). We obtained and outcrossed a null mutant for hlh-11(ok2944), which when crossed into the Patgl-1::GFP reporter line showed significant in vivo induction of atgl-1 in the intestine ( Figures 4A, B). These results were reinforced by measuring transcript levels of atgl-1 mRNA by qPCR, which were significantly greater in hlh-11 null mutants ( Figure 4C). We also noted that the extent of atgl-1 induction by FLP-7 was matched by the absence of hlh-11 mutants without a further increase in flp-7 tg ;hlh-11 animals ( Figures 4A-C). These data suggested that in wild-type animals, HLH-11 functions as a repressive transcription factor that suppresses atgl-1 under basal conditions. In turn, this predicts that a metabolic phenotype may result from hlh-11 removal.
hlh-11 null mutants showed a dramatic ~70-80% decrease in body fat stores relative to wild-type and phenocopied the FLP-7 transgenic line ( Figures 4D, E). flp-7 tg ;hlh-11 animals also had significant reductions in body fat stores and resembled either single mutant alone. Removal of atgl-1 by RNAi completely suppressed fat oxidation under all conditions ( Figures 4D, E).
Further, the fat loss phenotype of hlh-11 null mutants was accompanied by a significant increase in basal and maximal respiration that was also abrogated in the absence of atgl-1 ( Figures 4F, G). Together, these data show that loss of the transcription factor hlh-11 constitutively increases atgl-1 gene expression, fat oxidation and energy expenditure.
To test whether HLH-11 was also instructive in regulating atgl-1 expression and subsequent fat metabolism, we generated an HLH-11 overexpression line (henceforth HLH-11 ox ) fused to GFP using the endogenous 3 kb promoter (Lee et al., 2009). We observed robust nuclear HLH-11 expression in the intestine ( Figure 5A), suggesting a plausible model for the interaction between HLH-11 and atgl-1. Remarkably, HLH-11 ox animals showed an ~40% increase in body fat stores compared to wild-type animals and showed a near-complete suppression of the fat loss elicited in flp-7 tg animals ( Figures 5B, C). The increase in fat stores was accompanied by a corresponding decrease in maximal respiration, which was not further decreased upon removal of atgl-1 ( Figure 5D). The fat accumulation and metabolic output in HLH-11 ox and flp-7 tg ;HLH-11 ox animals were accompanied by a significant decrease in atgl-1 mRNA, as judged by qPCR ( Figure 5E). Thus, HLH-11 plays an instructive role as a negative regulator of atgl-1 expression and influences fat oxidation and mitochondrial respiration in the intestine.
The cis-binding site of HLH-11 has been precisely mapped to an 8-mer (Lee et al.,

2009), but
has not yet been functionally tested. We identified two HLH-11 cis-binding sites at 389 and 1,149 bp upstream of the atgl-1 transcriptional start site ( Figure 5F). To test for direct binding of the atgl-1 promoter by HLH-11, we conducted chromatin immunoprecipitation experiments followed by qPCR (ChIP-qPCR) of the transgenic line in which HLH-11 was fused to GFP. Relative to wild-type animals, we observed a substantial and significant increase in atgl-1 promoter regions bound to HLH-11 ( Figure 5G), thus the transcription factor HLH-11 binds directly to the atgl-1 promoter. Finally, we generated an atgl-1 reporter line that lacked both hlh-11 cis sites (Δcishlh-11) and observed a ~ 2-fold increase in atgl-1 expression ( Figures 5H, I).
Taken together, the conserved transcription factor HLH-11 is a direct repressor of atgl-1 expression, fat oxidation, and mitochondrial respiration.
The mito-nuclear transcription factor ATFS-1 promotes fat oxidation via HLH-11 regulation Next, we wished to examine how hlh-11 itself is regulated in the context of the FLP-7 pathway.
However, we had originally identified HLH-11 in light of our data suggesting that the mito-nuclear transcription factor ATFS-1 was indirectly required for FLP-7-mediated atgl-1 induction and fat oxidation in the intestine (Figure 3). In seeking a greater understanding of how ATFS-1 may be connected to HLH-11, we serendipitously found that a previous report had listed hlh-11 as one of the direct targets of ATFS-1 (Nargund et al., 2015). To test the possibility of an interaction between ATFS-1 and hlh-11 in the context of FLP-7-mediated fat oxidation and mitochondrial stress, we measured hlh-11 transcripts in the absence of atfs-1. We found that atfs-1 removal blunted FLP-7-mediated hlh-11 repression ( Figure 6C), suggesting that FLP-7mediated hlh-11 repression requires atfs-1. In flp-7 tg animals, inactivation of atgl-1, a condition that would simultaneously block fat oxidation and atfs-1 activation, also blunted this hlh-11 repression, suggesting a connection between fat oxidation, atfs-1 induction, and hlh-11 repression. Loss of both atfs-1 and atgl-1 diminished the FLP-7-dependent suppression of hlh-11 to a similar extent as loss of either gene alone ( Figure 6C). Thus, the repression of hlh-11 by neuronal FLP-7 signaling occurs via ATFS-1, which is stabilized during fat-oxidation-induced mitochondrial stress ( Figure 3). These results suggest the interesting possibility of a feedforward loop ( Figure 6D): repression of hlh-11 by FLP-7 permits atgl-1-dependent fat oxidation, which induces atfs-1 and in turn represses hlh-11 to sustain and augment fat oxidation. Predictions from this model are two-fold. On the one hand, the ATFS-1/HLH-11 interaction should modulate the hsp-60-dependent mitochondrial stress response, and on the other, it should also regulate the atgl-1 transcriptional response ( Figure 6D).

A feedforward loop orchestrates the relationship between fat oxidation, mitochondrial stress and longevity.
To functionally test the feedforward loop model ( Figure 6D), we studied the relationships between HLH-11 and ATFS-1 in the context of FLP-7 signaling. First, we predicted that HLH-11, a transcription factor that suppresses atgl-1 and fat oxidation, would also regulate the mitochondrial stress response, in an atgl-1-dependent manner. Accordingly, we found that overexpression of HLH-11, which represses atgl-1 mRNA ( Figure 5D), showed a significant decrease in hsp-60 mRNA ( Figure 6E). Thus, HLH-11 ox decreases not only fat oxidation and mitochondrial respiration (Figures 5B, C, E), but also the ensuing stress response. In contrast, hlh-11 null mutants, which constitutively stimulate fat oxidation and mitochondrial respiration by modulating atgl-1 transcription (Figure 4), significantly increased hsp-60 mRNA ( Figure 6E).
Effects of HLH-11 overexpression and absence were identical in the presence and absence of FLP-7 secretion ( Figure 6E). Also, the induction of hsp-60 upon loss of hlh-11 was fully suppressed by removal of either atgl-1, atfs-1, or both, suggesting that HLH-11 mediated mitochondrial stress is a direct consequence of fat oxidation ( Figure 6E). Thus, rather than being independent of one another, HLH-11-mediated control of fat oxidation and energy expenditure evoke a stress response as a direct consequence of these mitochondrial functions.
Second, we tested the relationship between ATFS-1 and its transcriptional target hlh-11, with respect to atgl-1 expression (refer to model in Figure 6D). As expected, hlh-11 mutants increased atgl-1 expression with and without increased FLP-7 secretion as judged by atgl-1 reporter expression ( Figure 6F) as well as qPCR ( Figure 6G). Also as predicted, atfs-1 loss alone does not lead to appreciable changes in atgl-1 (Figures 6F, G); this is because atfs-1 is non-functional in wild-type animals (Nargund et al., 2012). In contrast, hlh-11;atfs-1 double mutants resembled hlh-11 mutants alone, thus during increased FLP-7 secretion, atfs-1dependent suppression of atgl-1 induction requires hlh-11 repression ( Figures 6F, G). This result again suggested that rather than being a simple consequence of fat oxidation, the mitochondrial stress response is an integral component of sustained fat loss via atgl-1 transcriptional regulation.
We were curious whether disrupting the HLH-11/ATGL-1/ATFS-1 feedforward loop ( Figure 6D) would shed light on the relationship between sustained fat oxidation and longevity.
To this end, we measured the consequences of these transcriptional changes on physiological parameters. flp-7 tg animals have reduced fat stores because of increased atgl-1-dependent fat oxidation ( Figure 1). In the context of flp-7 tg animals, hlh-11 mutants also show augmented fat loss, whereas atfs-1 removal blocked fat oxidation ( Figures 7A, B). In accordance with the increased atgl-1 transcript levels ( Figures 6F, G), we found that hlh-11;atfs-1 double mutants also resembled hlh-11 single mutants alone in their fat phenotypes, suggesting that the suppression of fat oxidation in atfs-1 mutants is dependent on the presence of hlh-11 ( Figure   S3). Thus, ATFS-1-dependent fat oxidation requires hlh-11 repression of atgl-1 ( Figures 7A, B); HLH-11 acts genetically downstream of ATFS-1. We had already noted that increased fat oxidation via augmented FLP-7 secretion in flp-7 tg animals lead to no appreciable change in lifespan (Figures 2A, B). One explanation for this result is that FLP-7 signaling induces ATFS-1 activation because of mitochondrial fat oxidation. In addition to its role in repressing hlh-11, ATFS-1 has hundreds of additional targets which in combination have been postulated to serve mitochondrial recovery functions (Lin and Haynes, 2016;Nargund et al., 2012). Thus, in the context of flp-7 tg animals, concomitant with the fat oxidation, our data suggest that an ATFS-1dependent mechanism is simultaneously evoked, thus protecting lifespan. In testing this idea, we found that in flp-7 tg animals removal of both hlh-11 and atfs-1 led to a significant decrease in both median and maximal lifespan (p<0.001, Figures 7C, F) that was not seen in hlh-11;atfs-1 mutants ( Figure 7D), hlh-11 ( Figure 7E) or atfs-1 mutants (Tian et al., 2016). We reasoned that in hlh-11;atfs-1 mutants alone (that is, without increased FLP-7 secretion) we observed modest fat loss ( Figure S3) that is not sufficient to shift lifespan in either direction ( Figure 7D). Even though hlh-11 mutants show a substantial increase in fat oxidation (Figures 4 D,E), the presence of atfs-1 ( Figure 6E) prevents lifespan shortening.
Why did loss of hlh-11;atfs-1 in the flp-7 tg animals alone lead to a decrease in longevity?
Our model ( Figure 7G) is consistent with the following interpretation: FLP-7 secretion serves as the neuronal cue to trigger fat oxidation by repressing hlh-11 expression that in turn derepresses atgl-1 (red pathway, Figure 7G). The resulting increase in fat oxidation and mitochondrial respiration generates an ATFS-1-mediated mitochondrial response, as judged by induction of the mitochondrial stress sensor hsp-60. We propose that ATFS-1 induction serves as a second cue from the mitochondria to further repress hlh-11, which in turn augments fat oxidation via ATGL-1 in a feedforward loop (green pathway, Figure 7G). The atgl-1/atfs-1/hlh-11 feedforward loop serves the dual functions of sustaining fat oxidation and protecting lifespan.
Although each influences the other, the hlh-11/atgl-1 arm of the pathway primarily drives fat loss, whereas the atfs-1/hlh-11 arm of the pathway primarily protects lifespan.

DISCUSSION
In this study, we wished to address the question of whether, in a non-disease context, sustained fat oxidation and increased mitochondrial respiration have an effect on lifespan. We initiated these studies because the relationship between fat oxidation and longevity, although of major significance, has remained opaque. We report here that sustained fat oxidation evokes a mitochondrial stress response that functions to simultaneously augment fat oxidation and confer longevity protection (modeled in Figure 7G). We have found that the dual control of fat oxidation and lifespan protection emerges from a feedforward transcriptional loop and shields the organism from life-shortening mitochondrial stress in the face of continuous fat oxidation. Thus, sustained fat oxidation does not shorten lifespan because a mitochondrial response protects both. Under normal or wild-type conditions when fat oxidation levels are not high, this loop remains latent.
In this study, we have uncovered a regulatory pathway that is initiated in the nervous system and functions in the intestine, the major seat of metabolic and longevity regulation. The conserved transcription factor HLH-11 functions as the nexus of this signaling pathway and receives two signals: one non-cell-autonomous via neuronal FLP-7 that transmits sensory signals to initiate a metabolic response, and the other cell-autonomous via the mito-nuclear transcription factor, ATFS-1, that coordinates the mitochondrial stress response. HLH-11 is a direct transcriptional repressor of the triglyceride lipase atgl-1: loss of hlh-11 increases atgl-1dependent fat oxidation, and overexpression of hlh-11 has the opposite effect. Thus, controlling hlh-11 levels can serve as an excellent surrogate for titrating intestinal fat stores in future efforts.
Signaling loops as a network feature of biological systems are not uncommon; three predominant types have been described in the literature: negative feedback loops, positive feedback loops, and feedforward loops (Hornung and Barkai, 2008;Reeves, 2019). Negative feedback loops are commonly found in metabolic pathways such as glycolysis, in which a signal input (glucose) initiates a biochemical cascade that must be eventually be turned off as the pathway reaches capacity (Salway, 1999). Thus, negative feedback loops are sensitive to the external input and serve to diminish output over time. In contrast, positive feedback loops serve to amplify external signals over time, and as the network amplifies, become insensitive to the external stimulus (Abdel-Sater, 2011;Doncic and Skotheim, 2013). Such networks states have typically been described as unstable, and noted in deleterious physiological settings, such as an uncontrolled drop in blood pressure leading to death (Doncic and Skotheim, 2013;Goldstein and Kopin, 2017). Feedforward regulation is a network motif that is distinct from the above states. Network modeling approaches show that feedforward loops remain sensitive to the external input, are stable yet reversible, and are used to balance tradeoffs (Abdel-Sater, 2011;Jesty and Beltrami, 2005;Mitrophanov and Groisman, 2008). Here, we identify the broad molecular features of one such feedforward network state: the hlh-11/atgl-1/atfs-1 signaling loop, which remains sensitive to input from neuronal FLP-7, is stable and reversible, coordinates multiple physiological outputs, and balances the tradeoff between sustained fat oxidation and longevity. Such a feedforward motif may be a general conserved feature in balancing multiple physiological parameters. Funding Acquisition, S.S.

DECLARATION OF INTEREST
The authors declare no competing interests.

Materials Availability
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Supriya Srinivasan (supriya@scripps.edu).

Worm maintenance and strains
Nematodes were cultured as previously described (Brenner, 1974). N2 Bristol was obtained from the Caenorhabditis Genetic Center (CGC) and used as the wild-type reference strain. All mutant and transgenic strains used in the study are listed in the Supplemental Table 1. The hlh-11(ok2944) strain was provided by the C. elegans Gene Knockout Project at OMRF, which is part of the International C. elegans Gene Knockout Consortium. The atfs-1(tm4525) strains were generously provided by Cole Haynes and originally obtained from the National BioResource Project (Tokyo, Japan). Animals were synchronized for experiments by hypochlorite treatment and then hatched L1 larvae were seeded on plates with the appropriate bacteria. All experiments were performed on day 1 adults.

Cloning and transgenic strain construction
The hlh-11 promoter and gene were cloned using standard PCR techniques and Gateway Technology TM (Life Technologies) from N2 lysates. The final plasmid (Phlh-11::hlh-11GFP) encoded an HLH-11GFP fusion. 25 ng/µL of Phlh-11::hlh-11GFP plasmid was injected with 10 ng/µL of Pmyo-2::mCherry as a co-injection marker and 65 ng/µL of an empty vector to maintain a total injection mix concentration of 100 ng/µL. A well-transmitting transgenic line with consistent expression was integrated using the UV psoralen 2400 (Stratagene) and backcrossed 4 times before experimentation. To generate a plasmid of the atgl-1 promoter lacking the two hlh-11 cis-binding sites (Patgl-1 Δcishlh-11 ::GFP), the Q5 ® site-directed mutagenesis kit (New England Biolabs) was used on the Patgl-1::GFP plasmid previously generated (Noble et al., 2013). The primers used are listed in the Supplemental Table 2. 25 ng/µL of the Patgl-1 Δcishlh-11 ::GFP plasmid, 50 ng/ µL of rol-6 co-injection marker, and 25 ng/µL of empty vector were injected into wild-type worms. A transgenic line was selected based on consistency of expression and transmission.

5-HT treatment
5-HT hydrochloride powder (Alfa Aesar) was dissolved in 0.1 M HCl and was added to plates for a final concentration of 5 mM as previously described (Palamiuc et al., 2017).

RNAi
RNAi experiments were conducted as previously described (Kamath and Ahringer, 2003;Palamiuc et al., 2017). Plates were seeded with HT115 bacteria containing vector or the relevant RNAi clone four days prior to seeding larvae.

Oil Red O staining
Oil Red O staining was performed as described and validated (Hussey et al., 2018;Hussey et al., 2017;Noble et al., 2013;Palamiuc et al., 2017;Witham et al., 2016). Briefly, animals were washed off plates with PBS and incubated on ice for 10 min. Animals were fixed as described Wild-type animals were always included as controls within each experiment.

Image acquisition and quantitation
Black and white images of oil Red O stained animals were captured using a 10X objective on a Zeiss Axio Imager microscope. Images were quantified using ImageJ software (NIH) as previously described (Noble et al., 2013). All reported results were consistent across biological replicates. Fluorescent images of reporters for FLP-7 secretion were captured using a 20X objective on a Zeiss Axio Imager microscope. The first pair of coelomocytes was imaged. mCherry fluorescence intensity in one of the two imaged coelomocytes was quantified and normalized to the area of the coelomocyte GFP as previously described and validated (Palamiuc et al., 2017). For fluorescence imaging of gene expression reporter lines (animals with integrated Patgl-1::GFP, Phsp-60::GFP, or Phlh-11::hlh-11GFP transgenes), an equal number of animals were chosen blindly and lined up side by side. Images were take using a 10X or 20X objective on a Nikon Eclipse 90i microscope. Fluorescence intensity for all chosen animals was quantified for each condition and normalized to area of the animals excluding the head as indicated in the figure legend. Images were quantified using ImageJ software (NIH).

Oxygen consumption
Oxygen consumption rates (OCR) was measured using the Seahorse XFe96 Analyzer (Agilent) as previously described (Hussey et al., 2018). Briefly, adult animals were washed with M9 buffer and approximately 10 animals per well were placed into a 96-well plate. 5 measurements were taken for baseline, then at 37 min FCCP (50 µM) was injected to measure maximal OCR. Lastly, sodium azide (40 mM) was injected at 62 min to measure residual OCR. Afterwards, the number of worms per well was counted, and OCR values were normalized to number of worms per well. Basal OCR was calculated by averaging all measurements prior to FCCP (50 µM) addition, and maximal OCR was calculated by averaging the first two measurements after FCCP injection. In all genotypes tested, we did not observe any changes in worm size, growth or developmental stage.

Chromatin immunoprecipitation-qPCR
Chromatin immunoprecipitation was done as previously described (Mukhopadhyay et al., 2008). 50,000 synchronized D1 adult animals were washed with PBS three times and fixed with 1.1% formaldehyde for 15 min. Animals were partially lysed using a Dounce homogenizer. Fixative was quenched with 2.5 M glycine for 20 min. Animals were then washed and incubated in HEPES lyses buffer (50mM HEPES-KOH pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.1% (wt/vol) sodium deoxycholate, 1% (vol/vol) Triton X-100, 0.1% (wt/vol) SDS, 1 mM PMSF, and diluted protease inhibitor cocktail). Animal suspensions were sonicated using the Sonic Dismembrator Model 100 (Fischer Scientific) for 10 sec and then place on ice for 2 min, which was repeated eight times. An aliquot of lysate was kept for input DNA analysis. 3 mg of protein was subjected to a pre-clearing step by incubating samples with prewashed salmon sperm DNA/protein-A agarose beads (Millipore Sigma) for 1 hour. The supernatant was then incubated with GFP-Trap® coupled to magnetic agarose beads (Bulldog Bio) overnight at 4°C. As a negative control, wild-type animals, which lack GFP expression, were used. Bead-GFP-Trap-DNA complex was washed three times. Then precipitated DNA was eluted from the beads, and the cross-link was reverse overnight at 65°C. Precipitated chromatin and the input samples were treated with proteinase K, and the DNA was purified using the PCR Purification kit (Qiagen).
Purified DNA was then subjected to quantitative PCR using three sets of primers targeting the promoter region of atgl-1, and act-1 primers were used as a negative control. 3 technical replicates were used per group, and the entire ChIP-qPCR procedure was performed three times for a total of 3 biological replicates. All primer sequences are provided in the Supplemental Table 2. qPCR Total RNA was isolated from D1 adult animals using TRIzol TM (Invitrogen) and purified as previously described (Palamiuc et al., 2017). cDNA was made using iScript TM Reverse Transcription Supermix for RT-qPCR kit (BioRad) following the manufacturer's instructions.
SsoAdvanced TM Universal SYBR Green Supermix (BioRad) was used for performing qPCR according to the manufacturer's instructions. Data was normalized to act-1 mRNA. Primers sequences are listed in the Supplemental Table 2. Fold change was calculated following the Livak method (Livak and Schmittgen, 2001). Each group had at least 3 biological replicates in which worms were harvested and RNA was isolated from each biological replicate separately.
For each qPCR plate, at least 2 technical replicates were included per sample.

Lifespans
All lifespan experiments were performed at 20°C (Keith et al., 2014). Approximately one hundred L4 larvae per group were transferred to NGM plates seeded with OP50, which was recorded as day 0. Animals were transferred to new plates every other day until egg laying stopped. Surviving and dead animals were counted every other day until all animals were dead.
Animals were considered dead when they did not respond to a gentle stimulation with a platinum wire. Bagging, exploding, and contaminated animals were excluded from analysis.

STATISTICAL ANALYSES
Each assay was powered for sample size and statistical test based on the following: (i) pilot studies to assess the strength of the phenotype; (ii) minimum number of animals needed to detect significant differences (p<0.05). For each type of assay, we used the same number of animals for each genotype or condition so as not to over-power or under-power comparisons.
The sample size, statistical method and significance for each experiment is listed in the corresponding figure legend. All actual p-values are given in Supplemental Table 3. Wild-type animals were included as controls for every experiment. Error bars represent standard error of the mean (sem). Student's t-test, one-way ANOVA, Log-Rank Test, and two-way ANOVA were used as indicated in the figure legends. Bonferroni's correction for multiple comparisons was used for all ANOVAs.   GFP intensity was quantified and normalized to the area of each animal, fed vector or atgl-1 RNAi, expressed relative to wild-type vector ± sem. (n=30). ns, not significant. *, p<0.05 by twoway ANOVA. E. hsp-60 mRNA was measured via qPCR in the groups indicated. act-1 mRNA was used as a control. Data are presented as fold change relative to wild-type vector ± sem.
(n=4-6). ns, not significant. *, p<0.05 by two-way ANOVA. F. The fluorescence intensity of hsp-60 expression was quantified in the conditions indicated in the figure panel. Data are presented as a percent of wild-type vector ± sem. (n=30). ns, not significant. *, p<0.05 by two-way ANOVA.
G. qPCR of hsp-60 mRNA. Data are presented as fold change relative to wild-type vector ± sem. (n=4-6). ns, not significant. *, p<0.05 by two-way ANOVA. See also Figure S1. expression is quantified and presented as percent of wild-type vector ± sem. (n=29-30). ns, not significant. *, p<0.05 by two-way ANOVA. E. qPCR of atgl-1 mRNA. act-1 mRNA was used as a control. Data are presented for the indicated genotypes as fold change relative to wild-type ± sem. (n=4). ns, not significant. *, p<0.05 by one-way ANOVA. F. As indicated by the data, model depicting reciprocal regulatory relationship between the fat-burning enzyme ATGL-1 that triggers the hsp-60 mitochondrial stress response (red arrow), and the stress sensor ATFS-1 that is required for fat oxidation (green arrow). See also Figure S2.  Quantification of lipid droplets in the depicted conditions, presented as percent of wild-types ± sem. (n=18-20). ns, not significant. D. Maximal OCR of wild-type, HLH-11 ox , and flp-7 tg ;HLH-11 ox animals on vector or atgl-1 RNAi. Data are presented as pmol/min/worm ± sem. (n=15 wells each containing approximately 10 worms). ns, not significant. *, p<0.05 vs wild-type and $, p<0.05 vs vector RNAi by one-way ANOVA. E. qPCR of atgl-1 mRNA in wild-type, HLH-11 ox , and flp-7 tg ;HLH-11 ox . act-1 mRNA was used as a control. Data presented as fold change relative to wildtype ± sem. (n=4-6). ns, not significant. *, p<0.05 by one-way ANOVA. F. Schematic of the promoter region of atgl-1. There are two hlh-11 cis-sites (grey): one is 389 bp upstream of the atgl-1 transcription start site and the second binding site is another 752 bp upstream. For ChIP-qPCR, 3 primer sets were designed. Set #1 (light grey) flank the distal HLH-11 binding site, set #2 (dark grey) targets the region between the binding sites, and set #3 (black) flanks the proximal binding site relative to the transcriptional start site. G. Wild-type animals and animals bearing the Phlh-11::hlh-11GFP transgene were subjected to ChIP-qPCR. act-1 mRNA was used as a control. Data are presented as fold change relative to wild-type. ChIP-qPCR was performed using 3 technical replicates, and the experiment was repeated three times. nd, not detected. *, p<0.  act-1 mRNA was used as a control. Data are presented as fold change relative to wild-type vector ± sem. (n=4-6). ns, not significant. *, p<0.05 by two-way ANOVA. D. Model for interaction between HLH-11, ATGL-1 and ATFS-1. In wild-type (left panel), HLH-11 is constitutively on, which represses ATGL-1 (red arrow) and keeps fat oxidation low. In this state, mitochondrial stress is not activated, hsp-60 levels are low and ATFS-1 is not induced, keeping HLH-11 levels high. The FLP-7 neuronal signal (right panel) represses HLH-11, which de-represses ATGL-1, triggering fat oxidation. In turn, this generates an ATFS-1-dependent mitochondrial stress response as observed by the hsp-60 induction (blue arrow). ATFS-1 represses HLH-11 (green arrow), providing a feedforward cue to match fat oxidation with mitochondrial capacity. E. qPCR of hsp-60 mRNA in groups indicated in the figure panel. Data are presented as fold change relative to wild-type vector ± sem. (n=3-6). ns, not significant. *, p<0.05 vs wild-type on vector RNAi and $, p<0.05 vs vector RNAi by two-way ANOVA. F. The fluorescence intensity of atgl-1 expression was quantified and presented as percent of wild-type vector ± sem; groups as indicated in the figure panel. (n=26-30). * p<0.05 vs wild-type on vector and $ p<0.05 vs vector by two-way ANOVA. G. qPCR of atgl-1 mRNA in the genotypes indicated on the right. Data presented as fold change relative to wild-type ± sem. (n=3-6). *, p<0.05 vs wild-type by one-way ANOVA.