Molecular mechanisms of dietary restriction promoting health and longevity

Abstract

Dietary restriction with adequate nutrition is the gold-standard for delaying aging and extending healthspan and lifespan in diverse species, including rodents and non-human primates. In this Review, we discuss the effects of dietary restriction in these mammalian model organisms, and discuss accumulating data that suggests that dietary restriction results in many of the same physiological, metabolic and molecular changes responsible for the prevention of multiple age-associated diseases in humans. We further discuss how different forms of fasting, protein restriction and specific reductions in essential amino acids such as methionine and the branched-chain amino acids selectively impact AKT, FOXO, mTOR, nicotinamide adenine dinucleotide (NAD⁺), AMP-activated protein kinase (AMPK) and fibroblast growth factor 21 (FGF21), which are key components of some of the most important nutrient-sensing geroprotective signaling pathways that promote healthy longevity.

Introduction

Dietary restriction (DR) without malnutrition remains the most robust non-genetic intervention to date that can maximize lifespan and healthspan in rodents. It also extends life expectancy and safeguards against obesity, cancer, neurodegeneration, frailty, and a range of cardiometabolic conditions in rhesus monkeys; and in humans, it promotes adaptations that protect against these pathologies (Figure 1). This phenomenon was first discovered in 1917 by Osborne and was followed by hundreds of DR studies that expanded on its disease preventative and pro-longevity efficacy, founding the ‘biology of aging’ field¹.

Figure 1 |. The Hallmarks of dietary restriction.

This schematic enumerates the proven biological adaptations induced by dietary restriction (DR) that have a protective effect against ageing-related pathologies and diseases across rodents, non-human primates, and humans. These protective effects include the prevention of obesity and diabetes, cardiovascular disease, cancer, kidney disease, autoimmune and inflammatory conditions and cancer, leading to increased healthspan and lifespan. It is not yet clear what combination of transcriptional, epigenetic, proteomic, metabolomic, and microbiota changes drive such benefits of DR on healthspan and lifespan. Relevant references can be found in Supplementary information 1.

The mechanisms by which DR promotes health and longevity remained obscure until the 1980s, when the first single gene mutations targeting key nutrient-sensing pathways, which drastically extended the lifespans of nematode worms, were discovered². Since then, several more mutations that affect longevity have been found in invertebrates and their roles in the regulation of healthspan and lifespan has been confirmed in knockout and transgenic mouse models (Table 1). Understanding the mechanisms that underpin the beneficial effects of these mutations is essential to translating these findings to the clinic, through mechanism-based therapeutic interventions.

Table 1:

Genetic mouse models of extended longevity

Genetic Intervention	Genetic background	Sex	Lifespan extension	Phenotype	Ref
Inhibition of growth hormone signaling
Snell dwarf mice (dw/dw)	Snell dwarf mice (C3H/HeJ × DW/J)F1 background	Male and female	Average lifespan increased by 42%	Reduced body size, impaired growth hormone defects. Homozygous loss of function mutation in Pit-1	285,286
Ames dwarf mice (df/df)	Heterogeneous genetic background.	Male and female	Average lifespan increased by 64% (F) and 49% (M)	Reduced body size, impaired growth hormone defects Homozygous loss of function mutation in Prop-1	37,287–289
Little mice (lit/lit)	C57BL/6J	Male and female	Average lifespan increased by 23% (M) and 25% (F)	Slow growth excessive fat. Mutation in GH-releasing hormone (GHRH) receptor gene. Very low GH. Low fat diet for obesity prevention.
GH Receptor Knockout (Laron Dwarf) Mice GHR/BP −/−	129Ola and BalbC	Male and female	Average lifespan increased by 55% (M) and 38% (F)	Slow growth and reduced body weight after birth. Disruption of the GH receptor⧸GH-binding protein (GHR⧸GHBP). Redced body size. Significantly lower IGF-1 levels	39,290,291
AC5 KO	129/SvJ × C57BL/6	Male and female	Median lifespan increased by ~30%	Resistant to cardiac stress. Lower BW and GH.	292
Transgenic overexpression of FGF21	C57BL/6J	Male and female	Median lifespan increased by 30% (M) and 39% (F)	Blunts GH/IGF-1 signaling pathway in liver.	114
Inhibition of mTOR signaling
mTOR(Δ/Δ) mTOR expressed at 25% of WT	129S1 and C57BL/6Ncr	Male and female	Median lifespan increased by 22% (M) and 19% (F)	Reduced mTORC1 and mTORC2 activity. Smaller than WT.	64
mtor+/− mlst8+/−	C57BL6/129S5	Male and female	Median lifespan increased by n.s. (M) and 14.4% (F)	Decreased mTORC1 activity. Normal glucose tolerance and insulin sensitivity	59
S6K1−/−	C57BL/6	Male and female	Median lifespan increased by n.s. (M) and 19% (F)	Reduced fat mass and increased food intake (F)	63
Transgenic overexpression of human TSC1		Male and female	Median lifespan increased by n.s. (M) and 12.3% (F)		293
αMUPA mice	NIH FVB/N inbred mouse line	Female	Median lifespan increased by 16%	Consume 20% less food and exhibit 20% reduced body weight. Overproduce urokinase-type plasminogen activator (uPA) in the brain.	294
Inhibition of iInsulin/IGF-1 signaling
Igf1r+/− mice	129/Sv genetic background	Female	Average lifespan increased by 16% (M, n.s.) and 33% (F)	Heterozygous IGF-1 receptor knockouts. Greater resistance to oxidative stress	47
FIRKO mice	C57BL/6J × FVB/NJ	Male and female	Average lifespan increased by 18%	Fat-specific insulin receptor knockout. Reduced fat mass.	49
Irs1−/− mice	C57BL/6J × FVB/NJ	Male and female	Median lifespan increased by 16% (M) and 32% (F)	Delayed age‐sensitive markers in female Irs1−/− mice	51,52
Irs2+/− mice	C57BL/6J	Male and female	Median lifespan increased by 17%	Insulin resistant	50
Brain-specific Irs2+/− mice	C57BL/6J	Male and female	Median lifespan increased by18%	Insulin resistant, increased metabolic flexibility	50
Brain-specific Irs2−/− mice	C57BL/6J	Male and female	Median lifespan increased by 14%	Insulin resistant, increased metabolic flexibility	50
Irs2−/− mice	C57BL/6J × FVB/NJ	Male and female	Median lifespan decreased by 86% (M) and 23% (F)	Significantly shortens lifespan	51
Akt1+/– mice	C57BL/6	Male and female	Average lifespan increased by 8% (M) and 15% (F)	Decreased TOR signaling and suppressed mitochondrial activity	295
Altered sirtuin function or expression
Whole body Sirt1 overexpression	C57BL6/CBA	Male and female	No significant change in lifespan	No change in longevity, protected from development of age-associated diseases	145
Brain specific Sirt1-overexpressing (BRASTO) transgenic mice	C57BL/6J	Male and female	Median lifespan increased by9% (M) and 16% (F)	Enhanced neural activity	146
Sirt3−/−	C57BL6/J × 129Sv	Male	Median lifespan decreased by 19%	Shortened lifespan and severe cardiac damage	149
Sirt6-transgenic mice	C57BL/6J and BALB/cOlaHsd	Male and female	Average lifespan increased by 15.7% (M) and n.s. (F)	Reduced serum IGF-1 (M)	153
ATRAP‐KO	C57BL/6	?	Median lifespan decreased by 18%	Age‐related pathological changes in the kidney correlated with decreased expression of the prosurvival gene, Sirtuin1.	296
Adipose-specific overexpression of NAMPT	C57BL/6J	Male and female	Median lifespan increased by n.s. (M) and 13.4% (F)	Increased wheel running, and better sleep quality, glucose tolerance, pancreatic beta cell function, and cognitive function with aging	134

n.s.: Not significant; ?: Not stated in the original research.

In this review, we examine the research landscape of different forms of DR in rodents, non-human primates, and humans, focusing our attention on the metabolic and molecular adaptations that result in improved health. We also highlight new emerging scientific trends on the role of meal frequency and timing, and macronutrient composition, as potential mediators of some of the anti-aging and disease preventative effects of DR in rodents, non-human primates, and humans. As much of the mechanistic work regarding the role of specific pathways in the response to DR and macronutrients has been done in invertebrates, we will reference findings in other organisms as needed. However, the reader should remain aware that the role of these evolutionarily conserved pathways may not be precisely the same in mammals, and additional research may be needed to clarify if these pathways play the same role in DR, macronutrient restriction, or meal timing as they do in these model organisms. Understanding the molecular basis of DR and other dietary regimens may be crucial to developing dietary interventions or pharmaceuticals that can mimic some of the benefits of these regimens in the genetically heterogeneous human population.

DR and lifespan

In rodents, DR involves reducing ad libitum food intake by a given proportion (customarily 10–50%) without malnutrition³. Food restriction has a non-linear dose-dependent impact on rodent lifespan that is strain and sex specific^1,4,5. Lifespan increases to a maximum as food intake is reduced, but then rapidly declines when the restriction becomes excessive. Elegant experiments have also shown that the degree of DR that maximizes longevity and the amplitude of this response depend on genetic and, most likely, epigenetic factors. Indeed, even among genetically identical rodents, the same degree of food restriction leads to highly variable effects on lifespan⁶. Because different mice grow (in utero, postnatally, and during puberty) and reproduce at different rates, it is essential to perfectly match metabolism, growth, and fecundity to the optimal intake of food and nutrients, to avoid starvation or overfeeding, while maximizing health and longevity.

Understandably, there has been significant interest into whether DR can slow aging in humans. Dozens of studies, conducted by many independent groups over the last century, have found that the benefits of DR are observed in diverse species, including yeast, worms, flies and rodents. Studies in rhesus macaques show that the benefits of DR can also be observed in non-human primates. In a study conducted at the University of Wisconsin (UW), the hazard ratio for survival was 1.86, indicating that at any time point monkeys that followed a 30% DR regime had almost half the rate of mortality of ad libitum fed controls⁷. Consistently, ageing-associated methylation drift in the UW DR monkeys was markedly delayed and their estimated biological age was 7 years younger than their chronological age⁸. Another study conducted by the National Institute on Aging (NIA) reported no difference in the Kaplan-Meier estimated survival in DR monkeys (of note, the study was not statistically powered for this outcome), but the average lifespan of DR animals was extremely high: 31.8 years⁹. Whereas the median and 10% survival of rhesus monkeys in captivity is ~26 and ~35 years of age, respectively, about one third of the NIA monkeys on late-onset DR lived over 40 years, and one monkey died at 44 (the equivalent of 135 years for a human)^7,9.

Although the results and interpretations of these trials are still being debated, these findings, in conjunction with emerging new investigations in model organisms and humans, highlight two crucial points: (i) the impact of DR on reducing age-associated ailments appears conserved across species; (ii) the importance of study design, feeding regimen and diet formulation in modulating both lifespan and disease development and progression^7,10. In Figure 2 (and more extensively in Supplementary information 1), we summarize the results of key randomized trials in rodents, monkeys, and humans on the effects of DR in preventing a wide range of diseases, highlighting similarities and differences.

Figure 2 |. Multiple molecular pathways engaged by dietary restriction.

Dietary restriction (DR) results in reduced consumption of most macronutrients, including carbohydrates and specific amino acids, the building blocks of proteins. Reduced levels of glucose and its catabolite dihydroxyacetone phosphate (DHAP) are sensed by AMP-activated protein kinase (AMPK) and mTOR complex 1 (mTORC1), resulting in increased AMPK activity and decreased mTORC1 signaling, mediated through activation of TSC as well as modulation of the Rag-GATOR pathway that controls lysosomal localization of mTORC1. Downstream of mTORC1, ribosomal biogenesis and protein synthesis are downregulated and autophagy is increased. Decreased levels of methionine, branched-chain amino acids (BCAAs), or of protein similarly reduce mTORC1 signaling via the Rag-GATOR pathway. Decreased levels of protein and amino acids are also sensed by the integrated stress response pathway via GCN2, eukaryotic translation initiation factor eIF2α and cAMP-dependent transcription factor ATF4, leading to the induction of the pro-longevity hormone fibroblast growth factor 21 (FGF21). Reduced levels of carbohydrates and calories lead to decreased insulin/insulin-like growth factor 1 (IGF-1) signaling, which leads to decreased activity of the PI3K/mTOR complex 2 (mTORC2)/AKT signaling cascade that normally inhibits forkhead box protein O (FOXO)-dependent gene transcription, as well as decreased mTORC1 activity. Decreased levels of methionine lead to decreased levels of the metabolite S-adenosyl methionine (SAM), altering DNA and histone methylation. Collectively, DR induces repair and recycling pathways, including autophagy, mitophagy, DNA repair, and oxidant defense, and enhances stem cell function. As a result, cell senescence is downregulated and proteostasis is improved. Together these positive effects on cell and tissue function (shown in blue) contribute to extension of lifespan and healthspan. Proteins or protein complexes with kinase activity are depicted in red. SIRT1, sirtuin 1.

DR and intrinsic aging

The reduced prevalence of chronic diseases in animals subjected to DR does not completely explain the increase in maximal lifespan, because DR animals at any time appear physiologically younger and less frail than age-matched controls. Approximately one-third of DR rodents die at a very old age without any gross histological lesion at necropsy¹¹, suggesting that death may be due to intrinsic aging rather than ageing-related pathologies. Similarly, 20% of Ames dwarf mice and 47% of growth hormone receptor knockout (GHR-KO) mice, which are both long-lived and share similar growth hormone signalling deficits, do not develop any obvious lethal pathological lesions (i.e. known cause of death), suggesting that organ integrity can be preserved during aging in mammals¹². For example, normal aging is associated with a progressive decline in left ventricular diastolic function and heart rate variability, and DR counteracts these physiological changes in rodents and humans^13,14. Lower insulin/insulin growth factor-1 (IGF-1) and transforming growth factor beta (TGFβ) signaling^13,15, increase in antioxidant mechanisms^16,17, enhanced mitochondrial function^18,19 and improved proteostasis and autophagy^20–23 may explain some of the beneficial effects of DR.

In mammals ageing is also associated with presbycusis, and DR prevents it, at least in part, by inhibiting apoptosis of the cochlear ganglion cells via sirtuin 3 (SIRT3)-induced activation of isocitrate dehydrogenase 2 (IDH2) and increased mitochondrial NADPH levels²⁴. Sarcopenia is another universal aging feature, and DR delays the decline in skeletal muscle mass and strength, and the incidence of frailty by Fried’s index in both rodents and primates^25,26, likely in part by upregulating proteostatic, lipid synthetic and RNA processing pathways²⁷. In human skeletal muscle, long-term DR upregulates protein quality control (as seen by an increase in the chaperones HSP-70 and Grp78) and the autophagic proteins LC3 and beclin-1, both necessary for the formation of autophagosomes and cargo engulfment²². Moreover, as in rodents, chronic DR induces a PCG-1α-dependent increase of genes involved in mitochondria biogenesis^18,28.

Intense investigations in multiple non-mammalian model organisms are uncovering genetic and epigenetic pathways as well as microbial factors involved in the regulation of ageing and age-associated diseases, including nutrient sensing, genomic and protein homeostasis (for example, AKT/FOXO/mTOR, FGF21/GCN2/ATF4, AMPK and sirtuin pathways) (Fig. 3)²⁹. Remarkably, several of these genes and molecular determinants control mammalian longevity and nutrient-inducible pathophysiological processes, including cellular senescence, intra- and inter-cellular reprogramming, and telomere and stem-cell function. DR has also been shown to promote stem cell self-renewal and physiologic or injury-induced tissue regeneration in multiple organs (e.g. intestine, brain, skeletal muscle and bone marrow), partially through inhibition of mTOR complex 1 (mTORC1) signaling (Box 1)^30,31.

Figure 3 |. Species-specific effects of fasting on ketone bodies production and survival.

Remarkable differences in biological adaptations to fasting exist between mice and humans that should be considered when determining how studies from rodent models can inform human trials. Because of their high-energy metabolism, most strains of mice starve to death after a 48–60 hour fast. In contrast, even lean men and women can undergo a 57–73 days of water-only fasting before death occurs, and some severely obese individuals can fast for more than a year. Similarly, serum ketone levels increase after approximately 4–7 hours of fasting and peak after 24 hours in rodents, whereas in humans ketone bodies usually start to increase after 18–24 hours of fasting and do not peak until 2 weeks.

Box 1: Regulation of mTORC1 Activity by Amino Acids and Growth Factors.

mTORC1 is recruited to the lysosomal surface by heterodimers of the Rag family of small GTPases, which interact with the mTORC1 subunit Raptor when RagA/RagB is bound to GTP and RagC/RagD is bound to GDP (see the figure). The nucleotide binding state of the Rags is controlled by amino acids via several different protein complexes with guanine nucleotide exchange factor (GEF) or GTPase activating protein (GAP) activity, each of which is sensitive to different amino acids. These complexes include the Ragulator, a GEF activity for RagA and RagB; GATOR1, a GAP for RagA and RagB, and the FLCN complex, a GAP for Rag C and Rag D. Despite the massive amount of work already done to define the molecular mechanisms by which nutrients and environmental cues act to regulate mTORC1, more continues to be discovered, and key recent discoveries are highlighted below.

The GEF activity of Ragulator is modulated by multiple amino acids via lysosmal v-ATPase and SLC38A9. Specific sensors for the amino acids leucine (SESTRIN1, SESTRIN2, and SESTRIN3) and arginine (CASTOR1 and CASTOR2) have been identified; when levels of the sensed amino acids are low, these proteins inhibit GATOR2 activity to indirectly modulate the GAP activity of GATOR1 towards RagA and RagB. SAMTOR is a recently discovered indirect methionine sensor for mTORC1; SAMTOR regulates GATOR1 activity in response to levels of the methionine metabolite S-adenosylmethionine (SAM)²¹⁶, which is extremely responsive to methionine levels both in cell culture and in vivo²¹⁰. Another recent finding was the discovery that GATOR1 action upon RagA/B allows the recruitment of the FLCN complex to the lysosome^297,298. At the lysosome, FLCN acts to preserve lysosomal levels of leucine and mTORC1 activity by blocking accumulation of PAT1, a lysosomal amino acid transporter²⁹⁹; phosphorylation of FLCN by CDK4 is required for FLCN to depart the lysosome and allow the Rags to recruit mTORC1³⁰⁰. Finally, a new study suggests that while the RagGTPases are critical for the sensing of exogenous amino acids, lysosomal-derived amino acids activate mTORC1 via a RagGTPase-independent mechanism that requires the homotypic fusion and vacuole protein sorting (HOPS) tethering complex³⁰¹. Surprisingly, in the context of lysosomal derived amino acids the Rag-GATOR pathway acts as a negative regulator of mTORC1. A detailed mechanism for the regulation of mTORC1 by lysosomal-derived amino acids remains to be determined.

The Rags also play a role in glucose sensing by mTORC1^302,303. Glucose itself is not sensed by mTORC1; instead, the glycolytic intermediate dihydroxyacetone phosphate (DHAP) is detected via a GATOR2 and GATOR1 dependent mechanism³⁰⁴. The precise molecular sensor of DHAP that modulates GATOR activity and the nucleotide-loading status of the Rags remains unknown.

At the lysosome, mTORC1 is activated by the Rheb-GTPase, which binds to the mTOR protein kinase and allosterically realigns the kinase-site residues, activating its ability to phosphorylates substrates³⁰⁵; disruption of this interaction inhibits mTORC1 signaling⁸³. The nucleotide loading status of Rheb is controlled by the Tuberous Sclerosis Complex (TSC, which comprises TSC1, TSC2 and TBC1D7³⁰⁶), which acts as a GAP for Rheb; the activity of TSC is controlled by many different kinases, which phosphorylate different residues and proteins within the TSC complex⁵⁶. Intriguingly, the RagGTPases help to recruit TSC to lysosomes in response to amino acid or growth factor restriction³⁰⁷. Many details around this process are unknown, but GATOR2 seems to act as a regulator of TSC2 phosphorylation via this process, and Sestrin2 has likewise been implicated in the phosphorylation of TSC2³⁰⁸.

Geroprotective mechanisms of DR

Several metabolic pathways that have a role in the ageing-associated cellular and organismal decline are modulated by DR (Fig. 3).

Downregulation of growth hormone and Insulin/IGF-1 signaling

In the 1980’s, a mutation in age-1, which encodes a subunit of phosphoinositide 3-kinase (PI3K), was found to extend the lifespan of C. elegans³²; soon after, a mutation in daf-2 was identified which more than doubled the lifespan². Daf-2 is a regulatory gene that encodes a mammalian orthologue of the insulin/IGF-1 receptor (IIR)³³ and requires daf-16, which encodes a FOXO transcription factor². Other mutants in the insulin signaling pathway that can extend lifespan were soon identified in Drosophila melanogaster, including mutants in insulin receptor (InR)³⁴ and insulin receptor substrate (IRS)-like signaling protein, chico^35,36.

Overlapping these breakthroughs, the long-lived Ames Dwarf mice were discovered; these animals carry the ‘longevity’ gene, Prop1^df (a pituitary-specific homeodomain transcription factor) and live ~50% longer than their normal siblings³⁷. These mice, as well as the long-lived Snell Dwarf mice (which have a mutation in the anterior pituitary transcriptional factor, Pit1, also known as Pou1f1), are extremely small, and have very low levels of thyroid-stimulating hormone (TSH), prolactin, growth hormone (GH) and IGF-1³⁷. DR potentiates the longevity effect of dwarfism in the Ames mice, while GH treatment eradicated this beneficial effect³⁸. Consistently, it was found that GH deficiency alone markedly delays aging and produces the longest-lived laboratory mouse on record³⁹. As 30% DR in the GH receptor knockout mice (GHR-KO) mice failed to produce any further extension of overall or median longevity⁴⁰, suppression of the somatotropic axis may be one of the key longevity DR mechanisms.

GHR-KO mice have profoundly suppressed circulating levels of IGF-I, higher insulin sensitivity, and markedly lower cancer incidence¹². Humans born with GHR deficiency are also protected from cancer and diabetes, but do not have an increased lifespan⁴¹. Improved insulin sensitivity is a widely conserved response to DR in mammals and has been proposed as a key longevity mechanism of DR⁴², but recent data show that improvements in organismal insulin sensitivity are not required for DR to promote leanness, reduce frailty, and extend lifespan in mice⁴³.

Both long-lived Snell mice and GHR-KO mice show lower AKT activity, decreased glial fibrillary acidic protein (GFAP) phosphorylation and increased chaperone-mediated autophagy than control animals⁴⁴. Deletion of GH-R in macrophages seems to be one of the important actors linking reduced NLRP3 inflammasome-induced inflammation and longevity in GHRKO mice; inhibition of the GHR/IGF1 axis preserves the naive T-cell pool and prevents the age-associated activation of the inflammasome in response to accumulation of cellular damage⁴⁵.

As reviewed elsewhere^1,46, these and other findings support the idea that reduced insulin/IGF-1 signaling is crucial for lifespan extension. Consistently, genetic and pharmacological interventions that reduce insulin/IGF-1 signaling also extend mice lifespan. Initial work with mice heterozygous for the IGF-1 receptor found that these mice lived approximately ~30% longer than wild-type mice⁴⁷, but the control mice in this study were comparatively short-lived; independent replication of this experiment found only a small, female-specific increase in the lifespan of Igf1r^+/− mice⁴⁸. Deletions of the insulin receptor specifically in adipose tissue, Irs1 in the whole body, and Irs2 heterozygosity in the whole body or selectively in the mouse brain likewise extend lifespan^48–52. Loss of pregnancy-associated plasma protein A (PAPP-A), a metalloproteinase for IGF-1 binding proteins, extends lifespan when deleted constitutively or in adult mice^53,54. Finally, late-life inhibition of IGF-1 signaling using an antibody targeted to the IGF-1 receptor was recently shown to improve longevity⁵⁵. Overall, it is likely that reduced insulin/IGF-1 signaling contributes to the beneficial effects of DR, and strategies to reduce signaling through this pathway may have translatable potential to promote healthy aging.

Reduced mTORC1 signaling is a conserved mechanism for lifespan extension

One of the most important DR-induced molecular mechanisms downstream of insulin/IGF-1 signaling is the mTOR serine/threonine protein kinase (Fig. 2). mTOR is the catalytic core of two distinct protein complexes, mTORC1 and mTORC2, each of which are composed of distinct protein subunits that phosphorylate different substrates⁵⁶. Briefly, mTORC1 is responsive to a wide range of environmental stimuli, including the availability of amino acids, glucose, oxygen, cholesterol, and insulin/IGF-1, whereas mTORC2 is primarily an effector of PI3K signaling. mTORC1 is acutely sensitive to the drug rapamycin, whereas mTORC2, owing to its structure^57,58, is inhibited only by high levels of rapamycin over extended periods of time^59–61. mTORC1 integrates numerous environmental signals that indicate when conditions are favorable for the anabolic processes it controls, which include ribosomal biogenesis, protein translation, autophagy, lipogenesis and nucleotide biogenesis. The regulation of mTORC1 has been reviewed in detail⁵⁶ and is briefly outlined in Box 1.

mTORC1 activity is reduced in Ames and Snell Dwarf mice⁶², and several different genetic models of reduced mTORC1 signaling, including mice heterozygous for mTOR and mLST8 (which encodes a subunit of mTORC1), mice expressing a hypomorphic allele of mTOR, and those lacking the mTORC1 substrate S6K1, have extended lifespan^59,63,64. The initial studies on mTOR were conducted in invertebrates (reviewed in⁶⁵), but in 2009 it was found that rapamycin extends lifespan in aged mice⁶⁶. Since then several independent laboratories have confirmed the ability of rapamycin to prolong lifespan in multiple strains of mice, even when treatment is intermittent or conducted for only a short period of time^67,68 (and reviewed in⁶⁹). As summarized in Supplementary information 2, rapamycin and DR have similar positive effects on many age-related pathologies in mice; and in a recent small study, rapamycin even improved diastolic and systolic cardiac function in middle-aged dogs⁷⁰.

There has been significant interest in rapamycin as a potential DR mimetic, but rapamycin has several side effects that have precluded its wide-scale use; these include the dysregulation of blood glucose and lipid homeostasis as well as immunosuppression^59,71–73. These side effects are probably mediated by “off-target” inhibition of mTORC2, which is disrupted by chronic treatment with rapamycin in vivo in mice^59,60. Although whole body or tissue-specific (brain, liver, or adipose tissue) deficiency of mTORC2 signaling impairs metabolic health and reduces lifespan in wild-type and long-lived mice^43,74–76, mTORC2 activity has been positively associated with longevity in flies and in some (but not all) studies using worms^77–80. Two drugs that extend mice lifespan, acarbose and 17-α estradiol, boost hepatic mTORC2 signaling⁸¹, and mTORC2 activity is elevated in long-lived Snell dwarf mice and GHR-KO mice^62,81. There has been significant interest in identifying rapamycin dosing regimens or drugs with greater specificity for mTORC1 that could deliver the beneficial DR-mimetic effects of rapamycin while minimizing negative side effects^{67,72,73,82,83}.

Given the similarity of the beneficial effects of DR and mTORC1 inhibition, and the clear link between DR and reduced mTORC1 activity, it is widely accepted that mTORC1 has a role in the response to DR. In yeast, deletion of TOR1 is epistatic with DR; that is, DR does not further extend the lifespan of yeast lacking TOR1⁸⁴. However, genetic analysis of the interaction between DR and TOR in worms and flies has not clearly demonstrated an epistatic relationship. In C. elegans, several studies have linked various DR regimens to TOR signaling, but also identified TOR-independent longevity mechanisms activated by DR^85–88, whereas in flies, rapamycin extends lifespan at every level of calorie intake⁸⁹. Extensive mammalian metabolomic and transcriptomic studies in blood, liver, and white adipose tissue suggest that rapamycin and DR have distinct, largely non-overlapping effects^90–94. Additional research is warranted to fully evaluate the impact of different forms of DR on mTORC1 signaling.

Activation of GCN2 and reduced protein synthesis

GCN2 is another evolutionarily conserved serine/threonine kinase that functions as an amino-acid sensing metabolic switch to control various nutritionally responsive mechanisms, including immune system homeostasis⁹⁵ and tumor cell growth⁹⁶, and to coordinate integrated stress responses and the inflammasome⁹⁷. When GCN2 is activated through binding to uncharged tRNAs or by ribosome stalling^98–100, the protein kinase phosphorylates the α-subunit of eukaryotic initiation factor 2 (eIF2)¹⁰¹, blocking the translation of most mRNAs¹⁰¹, while stimulating the translation of a select set of proteins, including activating transcription factor 4 (ATF4)^102,103 (Fig. 3). ATF4 is a transcription factor that upregulates genes necessary to adapt to amino acid or protein restriction (a dietary regimen distinct from DR that involves the specific reduction of protein intake while overall calories are not restricted; see below), including the energy balance hormone fibroblast growth factor 21 (FGF21)¹⁰⁴.

Gcn2-knockout mice (Gcn2 is also known as Eif2ak4) that are subjected to protein restriction have a delayed metabolic response to protein restriction, including a two week delay in the induction of FGF21 and FGF21 mediated metabolic phenotypes (see below)¹⁰⁵. While the response to acute protein restriction may be dependent on activation of FGF21 through the GCN2-ATF4 axis, chronic protein restriction works upstream of GCN2 to directly activate ATF4, thereby stimulating FGF21 through alternate pathways¹⁰⁵. Although activation of GCN2 is not sufficient to directly inhibit mTORC1, it is required for mTORC1 inhibition upon leucine and arginine deprivation¹⁰⁶. Interestingly, this mTORC1 inhibition seems to occur, at least initially, independently of ATF4 activation, as it is not required for early repression of mTORC1 during leucine deprivation; however, phosphorylation of eif2α by GCN2 is essential, though not sufficient alone, to inhibit mTORC1¹⁰⁶.

While GCN2 has not been heavily investigated as a mediator of DR in mammals, GCN2 in C. elegans is required for both DR and TOR inhibition to extend lifespan, linking these two key amino acid sensing and longevity-regulating pathways¹⁰⁷. Understanding how GCN2 regulates the response to DR in mammals will clearly be an important area for future research.

Multiple effects of FGF21 signalling

Liver-derived FGF21 is implicated in many key metabolic pathways that are altered under nutritional stress¹⁰⁸. FGF21 is a potent regulator of the effects of protein as well as specific amino acids restriction on metabolism, notably by increasing insulin sensitivity and energy expenditure¹⁰⁹. FGF21 stimulates insulin-independent glucose uptake in the cells of both mice and humans^109,110, and facilitates adaptive behavioral changes in feeding through signaling to the brain¹¹¹.

The importance of FGF21 in the response to PR has been seen across multiple rodent models as well as in humans. In male Sprague Dawley rats and C57BL/6J mice, chronic PR induces an increase in both hepatic expression and circulating levels of FGF21^109,112. In humans, 4–6 weeks of protein restriction is sufficient to increase circulating FGF21^109,112. Changes in hepatic Fgf21 expression can be induced within 24 hours by switching rats to a low protein diet, followed by a 10-fold increase in circulating FGF21 after only 4 days¹⁰⁹. In mice lacking Fgf21, protein restriction is unable to provoke shifts in food intake, energy expenditure and weight gain^109,111.

As FGF21 is also powerfully induced by fasting, there has been significant interest in understanding if FGF21 mediates the effects of DR. Unfortunately, a straightforward assessment of how FGF21 levels are regulated by DR is confounded by time of feeding effects¹¹³. However, treatment of mice with recombinant FGF21 or transgenic expression of FGF21 promotes glucose tolerance and insulin sensitivity, and FGF21 overexpressing mice are resistant to diet-induced obesity¹¹⁰. Moreover, FGF21 transgenic overexpression extends mouse lifespan without reducing calorie intake or modulating mTORC1 signaling¹¹⁴. While future research will be needed to further clarify the role of FGF21 in the response to protein restriction and DR, FGF21 analogues are being pursed as a possible therapy for age-related diseases, including diabetes¹¹⁵, non-alcoholic fatty liver disease¹¹⁶, and Alzheimer’s disease¹¹⁷.

Activation of sirtuins, conserved regulators of lifespan

The lifespan extending effects of sirtuins, a family of nicotinamide adenine dinucleotide (NAD⁺)–dependent deacetylases, was originally discovered when a mutation in the silencing regulating gene Sir4 was found to extend the replicative lifespan of yeast. It was later demonstrated that this effect on yeast lifespan required silent information regulator 2 (Sir2), the homologue of mammalian SIRT1, and that overexpression of Sir2 increased yeast replicative lifespan¹¹⁸. Importantly, Sir2 and several yeast Sir2 homologues are required for DR to extend yeast lifespan^119,120. Sir2 homologues also regulate the lifespan of worms and flies and play a part in their response to DR^121,122, although these effects have not been observed consistently in different studies¹²³.

There are seven mammalian sirtuin family members (SIRT1–7), with different subcellular localizations. SIRT1 (with highest sequence homology to yeast Sir2), SIRT6 and SIRT7 are nuclear; SIRT3, SIRT4 and SIRT5 are mitochondrial; and SIRT2 is largely cytoplasmic, but shuttles in and out of the nucleus^124,125. While all of the mammalian sirtuins are (NAD⁺)–dependent deacetylases, other enzymatic activities have been reported for some of these enzymes, including ADP-ribosylation (SIRT 4 and 6)^126,127 and demalonylation and desuccinylation (SIRT5)^128,129. SIRT1 and SIRT2 have also been reported to act as lysine decrotonylases, and SIRT1–4 can remove lipoic acid from lysine¹³⁰.

Sirtuins are linked to diet and metabolism via their need for NAD⁺. The ability of DR to extend the lifespan of yeast is dependent not only on Sir2, but also on the NAD⁺ synthesis pathway enzyme Npt1¹¹⁹. Pnc1, a key enzyme in the NAD⁺ salvage pathway, is induced by DR in yeast via a TOR-mediated pathway and is required for DR-mediated lifespan extension in both yeast and worms^131–133. Adipose-specific overexpression of NAMPT, which is the rate limiting enzyme in a major NAD⁺ synthesis pathway, was recently shown to elevate NAD⁺ levels in multiple tissues, improve multiple measures of metabolic health, cognition, and physical performance, and extend the lifespan of female mice¹³⁴. Nutritional supplementation with NAD⁺ or NAD⁺ precursors is being actively investigated as a way to promote healthy aging and intervene in diseases by activating sirtuins (reviewed in¹³⁵).

A number of studies have found that DR induces SIRT1 expression in multiple tissues in rats and in human cells¹³⁶, and humans on DR have increased levels of SIRT1 in skeletal muscle¹³⁷. SIRT3 and SIRT5 are also induced by DR in mice^138,139. Conversely, overnutrition by high fat diet feeding lowers SIRT1 expression in mouse adipose tissue¹⁴⁰, and SIRT1 levels are lower in the adipose tissue of obese humans¹⁴¹, as well in Alzheimer’s disease¹⁴².

The role of mammalian sirtuins has been primarily investigated in mice in which sirtuins are deleted or overexpressed, either in specific tissues or universally. Whole body deletion of Sirt1 blocks the ability of DR to extend lifespan, although these results are complicated by the short lifespan of Sirt1^−/− mice¹⁴³. Mice genetically modified to overexpress SIRT1 are lean, metabolically active with reduced circulating cholesterol, insulin and improved glucose tolerance¹⁴⁴. While whole body overexpression of SIRT1 does not increase murine lifespan¹⁴⁵, hypothalamic-specific overexpression caused a significant extension of lifespan in both males and females¹⁴⁶. SIRT2 is a tumor suppressor¹⁴⁷, and its overexpression rescues the lifespan of progeroid mice lacking BubR1¹⁴⁸. Deletion of Sirt3 shortens lifespan¹⁴⁹, while increased Sirt3 levels improve the regenerative capacity of hematopoietic stem cells¹⁵⁰. Deletion of SIRT6 and SIRT7 increases frailty and shortens lifespan^151,152, while healthspan and lifespan is increased in transgenic male mice overexpressing SIRT6 (MOSES)^153,154; overexpression of SIRT7 extends the lifespan of a mouse model of Hutchinson-Gilford Progeria Syndrome¹⁵⁵.

The molecular mechanisms by which sirtuins regulate metabolism have been actively investigated. SIRT1 deacetylates and activates the transcriptional coactivator PGC1α by promoting its nuclear localization^156,157. SIRT1 in the liver up-regulates the transcription of the mTORC2 component Rictor, enhancing glucose homeostasis, while liver-specific deletion of Sirt1 increases oxidative stress¹⁵⁸. SIRT1 overexpression suppresses senescence, whereas its inhibition accelerates it in human endothelial cells^159,160. Sirt3 is essential for the DR-mediated reduction in oxidative damage, via enhancement of the mitochondrial glutathione antioxidant defense system, not only in the cochlear cell, but also in the neocortex and liver²⁴. This may be due to an inability of SIRT3-KO mice to lower their acetyl CoA levels in response to DR¹⁶¹. Interestingly, a polymorphism in the SIRT3 gene has been associated with male centenarians in a European population¹⁶². SIRT6 promotes DNA stability and suppresses senescence by enhancing DNA double-strand break repair, perhaps via activation of PARP^127,163,164. SIRT6-dependent repression of LINE1 elements, which induce inflammation and DNA damage, may be another important mechanism¹⁶⁵.

Taken together, the data collected thus far suggests that the sirtuin family of enzymes likely plays an important role in the response to DR. Research is continuing to define the targets and enzymatic functions of the sirtuins, and the role of each sirtuin in regulating metabolism, healthspan, and longevity.

Oxidative Stress and AMPK signalling

The oxidative stress theory of aging postulates that the accumulation of oxidative damage shortens lifespan. As reviewed elsewhere¹⁶⁶, several rodent studies have shown that aged DR animals and long-lived mouse mutants have reduced markers of oxidative damage. Similar reduced oxidation has been found in humans subjected to DR¹⁷. In support of this theory, mice overexpressing human catalase, which protects from oxidative stress by converting hydrogen peroxide to water and oxygen, localized to the mitochondria are long-lived¹⁶⁷.

This theory is not supported by experiments manipulating many other oxidative stress response genes. The lifespan of Sod1^−/− mice lacking CuZn superoxide dismutase is reduced^168,169, but mice one copy of Sod2 deleted, leading to increased DNA damage, or lacking Sod3, which have increased lipid peroxidation and higher sensitivity to oxidative stress, have a normal lifespan^169,170. Deletion of Gpx1 (encoding Glutathione peroxidase 1) or loss of MsrA (encoding Methionine sulfoxide reductase A) also result in increased lipid or protein peroxidation and increased susceptibility to oxidative stress without affecting lifespan^171,172. Studies crossing these mice to induce deletions of multiple oxidative stress genes have confirmed that increased oxidative stress is generally not highly deleterious to lifespan^169,171,173.

New surprising findings in C. elegans suggest that DR may increase reactive oxygen species and induce an oxidative stress resistance response that is essential for its pro-longevity effect^16,174. Intriguingly, this effect, named mitochondrial hormesis or mitohormesis, is dependent on AMPK (AMP-dependent kinase), a sensor of nutrient status and mitochondrial stress that regulates many cellular processes, including mTORC1 via phosphorylation of Raptor and TSC2^175,176. While the mitohormesis hypothesis has not been formally tested in mice, chemical inhibition of glycolysis by D-glucosamine extends the lifespan of C. elegans and C57BL/6 mice, possibly via increasing reactive oxygen species¹⁷⁷. Moreover, supplementation with antioxidants block the beneficial effects of exercise in humans^178,179, and might even increase cancer risk¹⁸⁰. In summary, the data suggests that the benefits of DR are probably not mediated by reduced oxidative stress.

Effects of specific dietary manipulations

Until recently, reduced intake of calories, rather than of specific nutrients, was considered key for the life-extension effects of DR. It is now clear that the old adage ‘a calorie is just a calorie’ is incorrect, as new data support a model whereby diet composition as well as timing of food intake have crucial roles in regulating key aging pathways^181,182.

Protein Restriction

Early studies in rodents on protein restriction had mixed results, probably owing to differences in dietary protein quality and the degree of restriction¹⁸². More recently, interest in PR has been rekindled by the finding that, in flies, total protein restriction or specific essential amino acid restrictions can extend lifespan independently of calorie intake^183–185. Studies in rodents have confirmed that protein restriction independent of caloric intake promotes longevity^186,187, and as described in Supplementary information 3, PR can impact a range of ageing-related conditions in both rodents and humans. These observations, and work on fruit flies done primarily by the Partridge laboratory, has resulted in a strengthening of the theory that it is the reduction in protein, and not calories, that drives the lifespan extension of DR¹⁸⁴. However, the amount of protein restriction during DR studies is smaller than during carefully controlled protein restriction studies, and recent analyses suggest that these interventions act through independent mechanisms¹⁸⁸. Indeed in recent studies on rodents short-term protein restriction did not replicate the physiological and metabolic effects of DR^189,190, and long-term protein restriction only extends the lifespan of male mice, while DR extends the lifespan of both sexes^43,187.

Estimated daily protein intake for adults in many Western Societies is approximately 90–100g (of which ~70–85% is animal protein), roughly twice the amount recommended by the US Institute of Medicine¹⁹¹. Accumulating data indicate that excessive protein intake may cause insulin resistance and type 2 diabetes and induce other long-term negative health consequences by overstimulating the the AKT−mTOR pathway and inhibiting FGF21 signaling¹⁹². In a weight loss trial, consuming a relatively high protein diet (1.3 g/kg/day) completely prevented the improved insulin sensitivity observed in women consuming a normal protein diet (0.8 g/kg/day) who lost the same amount of body weight¹⁹³. High protein intake, therefore, counteracts the beneficial effects of weight loss on insulin resistance, and potentiates the pro-aging and pro-cancer effects induced by compensatory hyperinsulinemia in the face of significant reductions in abdominal and liver adiposity.

In contrast, long-term adherence to physiologically adequate protein diets (0.8–1 g/kg per day) has been found to be beneficial in retrospective and prospective studies¹⁹², and a recent randomized clinical trial found that short-term PR significantly reduced fat mass and improved blood glucose levels in middle-aged overweight and mildly obese males without caloric restriction¹¹². The benefits of lower protein consumption may start in early life: newborns fed a lower protein-content formula (similar to that found in human milk) had a reduced risk of childhood obesity than infants consuming an isocaloric high protein-content formula¹⁹⁴. Although the effect of protein restriction on human lifespan is unknown, the longest living population in the world, the Okinawans, have traditionally eaten a diet containing 9% calories from protein¹⁹⁵.

As dietary amino acids are reduced by protein restriction, the activity of mTORC1, which is responsive to amino acid levels, is also reduced^186,196. As discussed above, protein restriction increases FGF21 levels in rodents and humans via activation of GCN2 and ATF4^{105,109,111,112,197}. GCN2 is essential for the acute response to protein restriction. Gcn2 knockout mice have increased FGF21 levels after only 2 weeks on protein restriction¹⁰⁵, indicating that GCN2 is only required for the initial induction of FGF21.

Restriction of specific amino acids: methionine

It has been proposed that a selective reduction of specific, essential amino acids might be sufficient to extend healthspan and lifespan, independent of total protein and calorie intake. Methionine, threonine, tryptophan and branched chain amino acids (BCAAs) have all been identified as potential candidates.

In 1993, Orentreich and colleagues observed that lifelong ~80% methionine restriction (MR) resulted in a 30% increase in lifespan in male rats¹⁹⁸; these findings were subsequently confirmed in mice¹⁹⁹. As illustrated in Supplementary information 4, methionine reduction has strong beneficial effects on the metabolic health of rodents, and reduced consumption of methionine potentially contributes to the health of humans consuming vegan diets. Methionine plays a unique role in translation, as methionine is specified by the AUG start codon and is thus required for translation initiation of most proteins. Methionine restriction thus has a dramatic effect on protein translation, strongly downregulating protein synthesis²⁰⁰. In addition to these effects, which may be beneficial for healthy aging, methionine also has important and unique metabolic roles and effects that have been implicated in the benefits of methionine restriction.

Many of the metabolic adaptations to methionine reduction have been attributed to the actions of the hormone FGF21, levels of which are upregulated by methionine restriction or depletion in both young and aged mice^201–203. As with protein restriction, a reduced dietary intake of methionine increases energy expenditure by promoting a FGF21-dependent browning of white adipose tissue²⁰⁴. Browning refers to a phenotypic switch where white adipose tissue becomes enriched in mitochondria and upregulates expression of uncoupling protein 1 (UCP1). This promotes thermogenesis and energy expenditure, resulting in reduced adiposity. This effect is independent of GCN2, and induced by the activation of the eIF2a kinase PERK²⁰⁵ as a consequence of methionine restriction-induced oxidative stress resulting from a depletion of glutathione. Dietary supplementation with cysteine blocks the effect of methionine restriction on FGF21 levels, adiposity, and energy expenditure^205,206.

Alterations in dietary methionine levels strikingly and rapidly lead to changes in the levels of its metabolites — the universal methyl donor SAM and cysteine, a key precursor of antioxidant glutathione and the gaseous messenger hydrogen sulfide (H₂S)²⁰⁷. SAM is crucial for histone and DNA methylation; and methionine restriction causes substantial reductions of SAM levels and alterations to both DNA and histone methylation^208–210. Changes in the levels of metabolites such as SAM and S-adenosylhomocysteine (SAH) are thought to drive the protection against hepatic DNA hypomethylation with age in adult mice²⁰⁸, and might account for the stronger effects on metabolic health seen with methionine restriction rather than leucine restriction on hepatic lipogenic gene expression and circulating FGF21²¹¹. H₂S, a powerful vasodilator, is endogenously produced via the trans-sulfuration pathway, and may help protect from multiple age-related diseases^207,212. Production of H₂S is required for DR to extend C.elegans lifespan²¹³. Finally, methionine restriction activates AMPK in mice²¹⁴, which is required for the lifespan extension induced by increased synthesis of the SAM in yeast²¹⁵.

SAM levels are also sensed by the mTORC1 protein kinase via SAMTOR, and a reduction in SAM leads to decreased mTORC1 signaling²¹⁶. However, even complete methionine depletion does not significantly alter hepatic mTORC1 activity, and mice lacking liver TSC1, which have constitutively active hepatic mTORC1, respond normally to methionine-deprived diet²⁰³. Thus, at least in the liver, reduced mTORC1 signaling does not mediate many of the metabolic effects of methionine restriction. Further research will be required to fully define the molecular mechanisms underlying the geroprotective effects of methionine restriction.

Restriction of BCAAs

As summarized in Supplementary information 4, increased blood levels of BCAAs (leucine, isoleucine, and valine) are associated with obesity and diabetes in humans. In mice, dietary BCAA restriction recapitulates many of the beneficial effects of protein restriction, including reduced adiposity, improved glucose tolerance, and elevated energy expenditure, but not increased FGF21^112,217. These effects can be largely recapitulated by restricting isoleucine, and the beneficial metabolic effects of protein restriction are dependent on isoleucine restriction²¹⁸. Conversely, high BCAA diets are associated with enlarged fat mass, increased body weight, hyperphagia, and insulin resistance^217,219,220. BCAAs are potent mTORC1 agonists, and dietary and circulating BCAAs levels are correlated with mTORC1 activation in mouse liver¹⁸⁶. Reintroduction of BCAAs to a protein restricted diet blocks the effects on mTORC1 signaling²²¹, highlighting the critical role of dietary BCAAs in mTORC1 activity. Although the Ames dwarf mouse has low levels of circulating BCAAs²²², the effects of BCAAs on the longevity and healthspan in this long-lived genetic model are only now being explored.

A low BCAA diet was recently shown to increase survival in Drosophila²²³, and a 67% restriction of dietary BCAAs increases the survival of two different progeroid mouse models¹⁸⁷. Similarly, initiation of a BCAA-restricted diet very early in life (3 weeks of age) not only improved metabolic heath, but in wild-type male mice reduced frailty and extended lifespan by over 30%¹⁸⁷. This lifespan extension was associated with a male-specific reduction in mTORC1 activity. However, beginning a 50% or 80% BCAAs restricted diet at 12 weeks of age, or a 67% BCAA restricted diet at 16 months of age, improved body composition and glucose homeostasis, but did not increase mouse lifespan^187,219. Finally, further supporting a negative effect of dietary BCAAs on mammalian healthspan and lifespan, dietary supplementation with additional BCAAs results in a hyperphagia-induced reduction in mouse lifespan²¹⁹.

Tryptophan or Threonine Restriction

Studies in the 1970’s and 1980’s reported that tryptophan restriction increases overall and maximum lifespan of mice and rats, and delays aging-associated pathologies including cancers^224–226. However, it was only recently found that tryptophan restriction may be an evolutionarily conserved geroprotective intervention as administration of ibuprofen, an inhibitor of tryptophan uptake, extends the lifespan of yeast, worms and flies²²⁷. Tryptophan is catabolized via the kynurenine pathway, and its metabolites include NAD⁺, which is a key regulator of metabolism and an essential co-factor for the activity of the sirtuin family of enzymes²²⁸.

While the effects of tryptophan restriction on aging in mammals has not been the direct subject of investigation again until recently, one recent study found that the serum level of tryptophan was associated with onset of diabetes²²⁹. There is also a negative correlation between tryptophan levels and cognitive function in humans²³⁰. Finally, restriction of dietary tryptophan induces expression of FGF21 and recapitulates metabolic effects of a protein restriction diet in mice²³¹. Much work remains to be done to examine if tryptophan restriction can promote healthy aging in mice or in humans.

A recent study surveyed the nine dietary essential amino acids to try and identify the amino acids responsible for the metabolic effects of a protein restricted diet, as modeled by a diet in which 5% of the calories were derived from casein. Intriguingly, in this dietary setting, restriction of threonine mimicked the effects of protein restriction, increasing energy expenditure and increasing insulin sensitivity²³¹. Restriction of threonine induces the hormone FGF21, and the effects of threonine restriction were shown to be FGF21-dependent²³¹. Levels of an enzyme in the threonine catabolic pathway, GCAT (glycine-C-acetyltransferase), decline with age in C. elegans and mice, and downregulation of this gene with RNAi extends the healthspan of C. elegans²³². Overall, while this work is suggestive that threonine restriction may promote health and longevity, this remains to be formally tested in mice and in humans.

Impact of meal frequency and timing

Unlike most humans practicing DR who typically cut calories at every single meal, hungry DR rodents devour their once-a-day restricted food allotment in 1 to 4 hours (depending on the mouse strain) followed by a daily prolonged period of fasting^233,234. Recently, and as discussed below, interventions such as meal feeding that involve the imposition of a daily fasting period have been shown to have metabolic benefits and extend the lifespan of mice^235–237. We recently utilized a series of dietary regimens to dissect the contributions of reduced calorie intake and an imposed fasting period to the metabolic, molecular, and geroprotective effects of DR²³⁸. We found that imposed daily fasting is required to observe DR-induced changes in insulin sensitivity and fuel selection, as well as for the geroprotective effects of DR on frailty, cognition, and lifespan²³⁸. Moreover, we determined that a prolonged daily fast without restricting calories is sufficient to recapitulate the metabolic phenotypes and transcriptional signature of a DR²³⁸. Thus, fasting between meals is a critical component of DR in rodents.

These observations coupled with the beneficial effect on murine healthspan and lifespan of alternate-day fasting, even independently of weight loss, has opened a new exciting field of translational research that is gaining momentum. Intermittent fasting in rodent models usually refers to a 24-hr complete fast every other day, while in humans it refers to a variety of regimes such as complete fasting or severe DR (e.g. 500–600 calories per day) on alternate days or 2 non-consecutive days per week (5:2 diet). Another form of intermittent fasting is time-restricted feeding, which involves consuming all daily food in a 4- to 12-h time window and fasting for the remainder of the day. Prolonged, periodic fasting, lasting for two to 7 consecutive days, and repeated cyclically, is an extreme form of DR that might have benefits for specific clinical indications.

An often overlooked problem in translating rodent findings to humans is that humans can fast for much, much longer than mice (Fig. 4). A 24-hr fast-feed cycle in mice most likely equates to recurrent ~5 days fast-feed cycles in humans. Another consideration is that laboratory rodents during the feast days consume a nutritionally balanced chow; in contrast, most people practicing different forms of intermittent fasting, prolonged, periodic fasting or time-restricted feeding eat the unhealthy Western-like obesogenic food that has been shown to cause negative consequences on metabolic and gut microbiome health even during DR, and potentiate vitamin and mineral deficiencies^{193,239–241}. Finally, unlike laboratory rodents many obese adults take a range of medications (e.g., antidiabetic and antihypertensive agents) that could have serious negative and potentially fatal consequences when coupled with fasting, including hypotension and severe hypoglycemia²⁴².

Impact of intermittent fasting on lifespan and healthspan

A 24-hour fast every other day or twice per week extends lifespan in mice and rats. The ability of IF to extend lifespan, as well as the magnitude of life extension, depends on background strain and age of initiation; intermittent fasting started at 10 months of age caused a 15% reduction in mean and maximal lifespan of A/J mice²⁴³. Intermittent fasting in rodents also reduces the incidence of a wide range of chronic diseases, including stroke, cardiomyopathy, hypertension, diabetes, and several neurodegenerative diseases due to its stimulatory effect on synaptic plasticity²⁴⁴. Cancer protection, however, is not universal, and several studies (especially in laboratory rats)^245,246 report a cancer promoting effect of intermittent fasting that may be mediated at least in part by TGF-β1²⁴⁷.

Several common pathways are induced by chronic DR and intermittent fasting. These include activation of multiple transcriptional factors induced by metabolic/hormonal modifications that lead to reduced oxidative stress and inflammation, and enhanced autophagy, mitophagy, and tissue repair capacity^248,249. In mice, a 24-hr fast increases intestinal stem cell function through induction of fatty acid oxidation; molecular deletion of carnitine acyltransferase I (the rate-limiting enzyme in fatty acid oxidation) reduces the numbers and function of gut stem cells²⁵⁰. Alterations of gut microbiota composition seem to play a role in mediating some of the effects of intermittent fasting on energy expenditure by selectively upregulating monocarboxylate transporter 1 and UCP1 expression in brown adipose tissue in rodents²⁵¹, but not in humans²⁵².

A peculiar characteristic of intermittent fasting that does not apply to chronic DR (when the restricted allotment of food is equally distributed during the day) is the metabolic switch with a transient elevation of plasma non-esterified fatty acids and ketone bodies that occur during fasting. β-OH butyrate binds to two G protein-coupled receptors, GPR109a and GPR41,^253,254 and by acting as an endogenous histone deacetylase inhibitor causes a wide range of modifications of gene expression and downstream signaling pathways that protects against oxidative stress²⁵⁵. However, mice and humans regulate ketones differently in response to acute fasting; in adult C57BL/6J mice, plasma ketones begin to increase after just 4–7 hours of fasting, and peak at about 24 hours²⁵⁶. In contrast, in adult humans the production of ketone bodies after an overnight fast is negligible; plasma β-OH-B levels start to creep up after 18–24 hours, and progressively increase (more rapidly in women than in men), peaking at 4–7 mM after 2 weeks fasting^257,258. Intermittent fasting in rodents exerts neuroprotective effects by increasing Brain-derived neurotrophic factor (BDNF) concentrations²⁵⁹. However, several clinical trials of intermittent fasting have reported a significant reduction of circulating BDNF levels^260,261, confirming that the metabolic adaptations to alternate day fasting in mammals with a very high-metabolic rate such as young rodents cannot be compared with those of middle-aged human adults.

While there have been no studies using intermittent fasting in non-human primates, several short-term randomized clinical trials (summarized in Supplementary information 5) have shown potentially favorable effects of fasting in humans. However, not all studies have demonstrated beneficial effects, and unlike in rodents where some of the metabolic adaptations of intermittent fasting are independent of food intake and weight loss, well-conducted human studies suggest that energy restriction is required to improve health²⁶². Thus, the translatable lessons that can be learned from intermittent fasting studies in rodents may be limited.

Prolonged, periodic fasting

In contrast to intermittent fasting, which consists of frequent short periods without or with limited amounts of food, prolonged, periodic fasting lasts more than 24 hours and is repeated once or twice a month. Because water-only fasting for more than 48–60 hours is deadly in mice, an alternative is to feed them reduced portions of a diet low in protein and carbohydrate, and high in unsaturated fat, providing between 10 and 50% of normal ad libitum intake. 16-mo old C57BL/6 mice fed a fasting-mimicking diet for four days twice per month had reduced weight and visceral fat accumulation, while preserving lean and bone mass. This fasting-mimicking diet regimen reduced cancer burden, rejuvenated the immune system, improved motor and memory performance, and increased median, though not maximum lifespan by 11%²⁶³.

Weekly 3-days of fasting-mimicking diet suppresses autoimmunity, and induces oligodendrocyte precursor cell regeneration and axonal remyelination in a murine experimental autoimmune encephalomyelitis model²⁶⁴. This suggests that a fasting-mimicking diet might work by activating stem cell-based or other regenerative processes through transient inhibition of the AKT, PKA and mTOR pathways²⁶⁵. Similarly, weekly 4-days of fasting-mimicking diet can cure both type 2 and type 1 diabetes, restoring pancreatic insulin production in a streptozotocin mouse model of type 1 diabetes by causing a stepwise expression of SOX17 and PDX1, followed by NGN3-driven generation of insulin-producing β cells²⁶⁶. Whether or not a fasting-mimicking diet can regenerate β-cells in patients with type 1 diabetes can be easily tested in a clinical study, because unlike multiple sclerosis, this autoimmune disease is not relapsing-remitting, and C-peptide is an excellent marker of therapeutic efficacy.

Preclinical data have shown that both prolonged fasting and fasting-mimicking diets can induce a differential stress resistance response in tumor-bearing mice that has the potential to maximize chemotherapy toxicity to cancer cells while protecting normal cells^267,268. Increased resistance to stress during fasting is associated with improved activity of chemotherapeutic agents by reducing circulating IGF-1, insulin and leptin and by inhibiting AKT–mTOR signalling via upregulation of EGR1 and PTEN²⁶⁹. By inhibiting the stress-responsive enzyme heme oxygenase-1 (HO-1), fasting increases T cell-dependent targeted killing of cancer cells in murine models of breast and melanoma cancer²⁷⁰.

In a recent randomized clinical trial of women with HER2-negative stage II/III breast cancer, 4 days of fasting-mimicking diet a week was able to reinforce the radiological and pathological tumor response to neoadjuvant chemotherapy and reduced DNA damage to T-cells²⁷¹. In this trial (in contrast to previous results reported in a non-randomized parallel study²⁶³), a fasting-mimicking diet significantly reduced circulating insulin levels but did not change serum IGF-1 or IGFBP-3 concentrations, while increasing inflammation as assessed by C-reactive protein²⁷¹. Very small preliminary clinical studies suggest that fasting before and during chemotherapy may reduce adverse events of chemotherapy and improve quality of life^272,273.

Time restricted feeding

Preclinical and epidemiological data show that consuming food out of sync with the day/night cycle and/or over an extended time frame impairs metabolic health²⁷⁴. In mice, time restricted feeding (8-hour daily access to food) is thought to regenerate disrupted circadian clock rhythms, and protect against obesity, fatty liver disease, insulin resistance, hyperinsulinemia and inflammation independently of caloric and fat intake²³⁵. These positive metabolic effects are attributed to a fine tuning of circadian clock genes in response to metabolic modulation of mTOR, CREB and AMPK pathway activity^235,236, even when weekend ad libitum “cheat days” are permitted. Disrupting circadian oscillations by genetic manipulations of clock genes that encode circadian rhythmicity (e.g. Clock, Per1, and Cry genes) induces obesity, alters glucose metabolism and reduces lifespan in mouse models^275–277, and irregular feeding uncouples peripheral clocks from the central pacemaker inducing insulin resistance and glucose intolerance²⁷⁸.

Data on early time-restricted feeding (from 6am to 2pm) in non-human primates are limited to a study of geriatric vervet monkeys showing a significant increase in HDL-cholesterol and reverse cholesterol efflux but not change in adiposity²⁷⁵. This is consistent with findings from a randomized clinical trial involving women affected by polycystic ovary syndrome showing that early meal timing (980 kcal breakfast, 640 kcal lunch, and 190 kcal dinner) is associated with more weight loss, higher insulin sensitivity, lower circulating testosterone, and increased ovulation rate than controls eating isocaloric diets with a later meal pattern (190 kcal breakfast, 640 kcal lunch, and 980 kcal dinner)²⁷⁹. A five-week, randomized, crossover, isocaloric and eucaloric controlled feeding trial in males with prediabetes found that independently of weight loss early time-restricted feeding (6-hr feeding period, with dinner before 3 pm) did not improve glucose, IL-6 and C-reactive protein levels, but ameliorated insulin sensitivity, β cell responsiveness and blood pressure; however, early- time-restricted feeding markedly increased plasma triglycerides and total cholesterol²⁸⁰. A number of small short-term (2–4 months) pilot studies in overweight individuals suggest a beneficial effect of time-restricted feeding (~8–12 hours per day) on body weight and composition²⁸¹, but the large TREAT randomized clinical trial challenges these findings. It shows that overweight or obese men and women randomized to time-restricted feeding diets(access to food for 8 hours per day) did not experience weight loss or any cardiometabolic or glucose homeostatic improvements over 3 months²⁸². While the underlying molecular changes that may dictate the beneficial effect of time-restricted feeding in humans is largely unknown, 30 consecutive days of dawn till dusk fasting (~14 hour fast) in healthy humans showed an anti-cancer and anti-diabetes serum proteomic signature²⁸³. Long-term randomized clinical trials are warranted to establish the efficacy of time-restricted feeding to improve metabolic health in primary and secondary prevention of other cardiometabolic conditions.

Conclusions

Many questions about the effects of DR and its clinical translatability remain unanswered. Research over the next decade will focus on understanding how precise dietary components – amino acids, specific sugars, fats and microbial metabolites – regulate health and longevity, and how these components interact with a DR diet, exercise and cognitive training and other lifestyle factors in a mechanistic way. In addition, the continued development of tools for conditional gene inactivation and mutation will permit a greater range of epistasis experiments to identify pathways required for DR to extend healthspan and lifespan. While we believe that promoting healthy eating and habits is the most cost-effective way to prevent multiple chronic diseases and promote human and environmental health²⁸⁴, identification of these pathways may also help to further the development of geroprotective agents that might potentiate the effects of healthy lifestyles.

Finally, a crucial point to consider is that humans are genetically heterogeneous, and experiments examining the interaction of diet and strain in mice have demonstrated that genetic and epigenetic background determines the response to dietary interventions, including DR. Our hope is that in the near future, these findings will be clinically translated using a personalized food-as-medicine approach, to identify how each person can best improve their health and potentially extend their lifespan by optimizing what, when and how much they eat.

Table 2 |.

Impact of dietary restriction on common ageing-associated pathologies and physiological decline on different mammalian species.

Organism	Human	Non-human primates	Rodents
Effects on body weight and metabolism
Body weight	↓	↓	↓
Fat mass	↓	↓	↓
Insulin sensitivity	↑	↑	↑
Inflammation	↓	↓	↓
Effects on cardiovascular disease
Atherosclerosis	↓	?	↓ a
Diastolic dysfunction	↓	?	↓
Cardiomyopathy	?	↓	↓
Blood pressure	↓	↓	↓
Cholesterol	↓	↓	↓
Heart rate variability	↑	?	↑
Effects on neurodegenerative disease
Memory	?	?	↑
Cognitive function	?	?	↑ c
Neuronal survival	?	↑	↑
Amyloid deposition	?	? b	↓ c
Effects on longevity and other chronic diseases
Lifespan	?	↑	↑
Obesity	↓	↓	↓
Type 2 diabetes	↓	↓	↓
Fatty liver disease	↓	?	↓
Kidney disease	?	?	↓
Cancer incidence	↓	↓	↓
Autoimmune disease	?	?	↓
Sarcopenia and frailty	?	↓	↓

Footnote: Up arrows: increase or improvement; down arrows: decrease or decline; question marks: unknowns.

^ain apolipoprotein E deficient mice.

^bAmyloid peptides were decreased in DR squirrel monkeys but were unchanged in DR rhesus macaques.

^cIn Alzheimer’s Disease mouse model. Relevant references can be found in Supplementary information 1.