Two Conserved Histone Demethylases Regulate Mitochondrial Stress-Induced Longevity

Summary

Across eukaryotic species, mild mitochondrial stress can have beneficial effects on the lifespan of organisms. Mitochondrial dysfunction activates an unfolded protein response (UPR^mt), a stress signaling mechanism designed to ensure mitochondrial homeostasis. Perturbation of mitochondria during larval development in C. elegans not only delays aging but also maintains UPR^mt signaling, suggesting an epigenetic mechanism that modulates both longevity and mitochondrial proteostasis throughout life. We identify the conserved histone lysine demethylases jmjd-1.2/PHF8 and jmjd-3.1/JMJD3 as positive regulators of lifespan in response to mitochondrial dysfunction across species. Reduction-of-function of the demethylases potently suppresses longevity and UPR^mt induction while gain-of-function is sufficient to extend lifespan in an UPR^mt-dependent manner. A systems genetics approach in the BXD mouse reference population further indicates conserved roles of the mammalian orthologs in longevity and UPR^mt signaling. These findings illustrate an evolutionary conserved epigenetic mechanism that determines the rate of aging downstream of mitochondrial perturbations.

Introduction

Aging is a deleterious and complex process characterized by the progressive decline of cellular and organismal homeostasis amid constantly increasing levels of entropy (Kirkwood, 2005) and represents the primary risk factor in major human pathologies including cancer, diabetes, cardiovascular disorders, and neurodegenerative diseases (Lopez-Otin et al., 2013). Aging, however, is not only a response to experiences incurred toward the end of an organism's life: it is shaped and determined by experiences that have accumulated from the earliest stages of development, and even during the generations that came before it. The organism's perpetuation of historic cues – how its patterns of gene expression change and become cemented in response to stresses that may have occurred long in the past or in parental populations – have proven to significantly affect long-term health and longevity, the mechanistic details of which are only beginning to emerge. It is thus that we now know, for example, that the transgenerational inheritance of chromatin marks can increase lifespan for multiple generations in C. elegans (Greer et al., 2010; Han and Brunet, 2012; Lopez-Otin et al., 2013; Rando and Chang, 2012) underscoring the importance of understanding not only the genetic but also the epigenetic contributions to the complex and disordered aging process.

One of the most dramatic examples in which early events have effects on longevity is found in the nematode C. elegans, in which mitochondrial stress during development can cause nearly a doubling of the animal's lifespan (Dillin et al., 2002a). The timing and degree of mitochondrial dysfunction is highly selective: it must occur during a specific L3/L4 larval transition in order to cause lifespan extension, a time during which a heavy amount of germline-specific mitochondrial biogenesis also occurs (Rea et al., 2007; Tsang and Lemire, 2002). In contrast, mitochondrial dysfunction that is too severe or which is implemented too early or late can have a negative effect on lifespan. In many cases, titrating a level of dysfunction is absolutely required in order to observe an extension of lifespan (Rea et al., 2007). Longevity caused by mitochondrial dysfunction also often fails to generate universal health benefits, as organisms may live longer but exhibit developmental delay and a drastic reduction in reproductive fitness (Dillin et al., 2002a; Lee et al., 2003). These effects are surprisingly conserved: in yeast, flies, and mice, mitochondrial dysfunction can delay the aging process (Copeland et al., 2009; Dell'agnello et al., 2007; Dillin et al., 2002a; Feng et al., 2001; Kirchman et al., 1999; Lee et al., 2003; Liu et al., 2005), but, when occurring later in life, has deleterious effects and is associated with age-onset neurodegenerative diseases, directly contributing to their pathologies (Schon and Przedborski, 2011).

Within mitochondria, intricate surveillance systems monitor the quality of existing and newly synthesized mitochondrial proteins to ensure mitochondrial homeostasis (Baker et al., 2011). The mitochondrial unfolded protein response (UPR^mt) consists of a signaling cascade that results in upregulation of nuclear-encoded genes to alleviate the stress sensed in mitochondria. Perception of misfolding in mitochondria requires the protease ClpP, which generates a mitochondrial derived signal to activate downstream genes (Haynes et al., 2007). ClpP triggers the activation of the ubiquitin-like protein UBL-5, which acts as a coactivator of the transcription factor DVE-1. UBL-5 and DVE-1 respond to mitochondrial perturbation to increase expression of mitochondrial chaperones, including hsp-6 and hsp-60 (Benedetti et al., 2006). In parallel, the bZIP transcription factor ATFS-1 coordinates a wide cellular response to mitochondrial stress (Nargund et al., 2012, 2015).

Previously, several genes characterized for their role in the regulation of the UPR^mt were identified as specific requirements for the long lifespan of animals with reduced mitochondrial function (Durieux et al., 2011; Houtkooper et al., 2013). The requirements for these genes in electron transport chain (ETC)-mediated longevity suggested that the function of the UPR^mt might have a beneficial effect on the organism and was required to maintain the longer lifespans in the mitochondrial mutants. In this model, mitochondrial dysfunction early in development was capable of imposing a mild, hormetic stress that could remodel gene expression patterns, perpetuating a beneficial effect long into adulthood (Durieux et al., 2011). In keeping with this hypothesis, induction of the UPR^mt during early larval stages persists long into adulthood, suggesting that the animal may possess an epigenetic mechanism to maintain the activation of stress responses and ensure increased resistance to future mitochondrial insults (Durieux et al., 2011). In yeast, mitochondrial stress generated by reactive oxygen species (ROS) causes an epigenetic remodeling that extends lifespan (Schroeder et al., 2013), and forced expression of UPR^mt genes in Drosophila is sufficient to preserve mitochondrial function (Owusu-Ansah et al., 2013).

In this study we identify two conserved histone lysine demethylases as regulators of the ETC longevity pathway. Using RNAi-based screens in C. elegans, we isolated the conserved histone demethylases jmjd-1.2 and jmjd-3.1 as potent suppressors of longevity in response to mitochondrial perturbations. We demonstrate that both jmjd-1.2 and jmjd-3.1 are necessary and sufficient for activation of the UPR^mt in C. elegans. Moreover, our experiments identify jmjd-1.2 and jmjd-3.1 as positive regulators of a longevity response that genetically requires UPR^mt signaling. Using transcriptome analysis, we demonstrate that jmjd-1.2 and jmjd-3.1 coordinate the transcriptional response to mitochondrial stress. Furthermore, using a systems genetics approach, we find that the mammalian orthologs exhibit positive genetic correlations with UPR^mt core components in the BXD mouse genetic reference population (Andreux et al., 2012; Wu et al., 2014). Together, these data reveal a conserved epigenetic mechanism that determines longevity and stress signaling in response to mitochondrial dysfunction.

Results

Mitochondrial ETC-Mediated Longevity Requires the Histone Lysine Demethylases jmjd-1.2 and jmjd-3.1

We performed an RNAi-based screen to identify genes specifically required for the ETC-mediated longevity in C. elegans. Through these analyses, we identified a putative histone lysine demethylase of the JumonjiC (JmjC)-domain-containing protein family, jmjd-1.2 (F29B9.2), (Kooistra and Helin, 2012), as a potent suppressor of the ETC longevity pathway (Figure 1A and Table S1). We hypothesized that other JmjC domain containing demethylases may have similar effects on the regulation of lifespan. From the JmjC orthologs found in C. elegans, we identified a second histone lysine demethylase, jmjd-3.1 (F18E9.5), that was also an effective suppressor of ETC-mediated longevity (Figure 1B and Table S1). In their roles as demethylases, jmjd-1.2 and jmjd-3.1 have both distinct and overlapping specificity for histone modifications, most notably against H3K27me2/3 (Agger et al., 2007; Kleine-Kohlbrecher et al., 2010). We found that both jmjd-1.2 and jmjd-3.1 were required for the extended lifespan caused by point mutation in the Rieske iron-sulfur protein (complex III), isp-1(qm150) (Feng et al., 2001) (Figures 1C, D). Downregulation of the closely related H3K27me2/3 demethylases jmjd-3.2, jmjd-3.3 and utx-1, however, did not significantly affect lifespan extension by cco-1 RNAi (Figure S1A), highlighting a specific role of jmjd-1.2 and jmjd-3.1 for ETC-mediated longevity.

Figure 1. Lifespan Extension by Mitochondrial ETC Perturbation Requires the Histone Demethylases jmjd-1.2 and jmjd-3.1.

(A) and (B) Knockdown of jmjd-1.2 (A) or jmjd-3.1 (B) suppresses cco-1-mediated lifespan extension.

(C) jmjd-1.2 is partially required for longevity of isp-1(qm150) mutant animals.

(D) jmjd-3.1 is required for longevity of isp-1(qm150) mutant animals.

(E) and (F) Dietary restriction-mediated longevity of eat-2(ad1116) animals is not affected by jmjd-1.2 (E) or jmjd-3.1 (F) knockdown.

(G) Longevity of daf-2(e1370) mutant animals is not affected by jmjd-1.2 knockdown.

(H) Longevity of daf-2(e1370) mutant animals partially depends on jmjd-3.1. See also Figure S1 and Table S1 for lifespan statistics.

Next, we assessed whether jmjd-1.2 and jmjd-3.1 were specific to lifespan extension in response to mitochondrial perturbation. The mitochondrial ETC longevity pathway acts in parallel to several other lifespan extending paradigms, including the insulin/IGF longevity response (Kenyon et al., 1993), dietary restriction (DR) (Lakowski and Hekimi, 1998), and germline-mediated longevity (Hsin and Kenyon, 1999). Neither jmjd-1.2 nor jmjd-3.1 downregulation altered longevity of dietary restricted eat-2(ad1116) animals (Figures 1E, F). We further found that jmjd-1.2 had no effect on the lifespan of animals in which stress responses of the ER or cytosol were constitutively activated (Baird et al., 2014; Taylor and Dillin, 2013) (Figures S1B-D) while transcription factors FoxO/daf-16 and FoxA/pha-4 successfully suppressed lifespan in our analyses (Figures 1E-H, Table S1). In contrast, while jmjd-1.2 RNAi did not affect the extended longevity seen in daf-2(e1370) mutant animals or germline deficient glp-1(e2141) animals (Figures 1G, S1B), we observed that the long lifespan daf-2(e1370) mutant worms depended on jmjd-3.1, albeit partially (Figure 1H). Recently jmjd-3.1 was also reported to be required for the lifespan extension of germline-deficient glp-1 animals (Labbadia and Morimoto, 2015). Collectively, these data indicate that while jmjd-3.1 is involved in several lifespan extension paradigms, both histone demethylases are required for ETC-mediated longevity.

Overlapping Temporal Requirements of JmjC Demethylase Activity and ETC-Mediated Longevity

One of the most remarkable features of ETC-mediated longevity in C. elegans lies in its precise temporal requirements: Reduction of ETC function during a narrow window of development is sufficient to extend lifespan throughout life (Dillin et al., 2002a; Rea et al., 2007). In contrast, loss of ETC function in adulthood is sufficient to reduce ATP production but can no longer affect lifespan. We reasoned that the activity of histone demethylases might be required during development and therefore performed RNAi shifting experiments in which we reduced ETC activity during development and restored them during adulthood by shifting young adults to Dicer (dcr-1) RNAi to inhibit the RNAi machinery (Dillin et al., 2002a, 2002b). We found that reduction of jmjd-1.2 during larval development was sufficient to block cco-1 mediated longevity (Figure S1E). jmjd-1.2 RNAi during adulthood, in contrast, reduced cco-1 lifespan partially (Figure S1F). jmjd-3.1 was partially required for cco-1 mediated longevity during larval development but had only a minor effect during adulthood (Figures S1G, H). These data suggest that both jmjd-1.2 and jmjd-3.1 are required for an initial response to ETC deficiency during development that is maintained into adulthood.

JMJD-1.2 and JMJD-3.1 Regulate the Mitochondrial Unfolded Protein Response (UPR^mt)

Previously, the UPR^mt was found to be required for the extended longevity of mitochondrial mutants (Durieux et al., 2011). We therefore asked whether jmjd-1.2 and jmjd-3.1 had any effect on the induction of the UPR^mt in C. elegans. In these analyses, cco-1 RNAi was used as a potent inducer of both hsp-6 and hsp-60 endogenous transcripts, robustly turning on the transcriptional hsp-6p∷gfp reporter (Yoneda et al., 2004). As reported, cco-1 mediated UPR^mt activation required the ClpP protease clpp-1 and the homeodomain transcription factor dve-1 (Haynes et al., 2007) (Figures 2A and S2B). Similarly, RNAi of either jmjd-1.2 or jmjd-3.1 suppressed the hsp-6p∷gfp reporter activation elicited by cco-1 (Durieux et al., 2011) (Figures 2A and S2A), spg-7 (Yoneda et al., 2004) (Figures 2B and S2C) or mrps-5 (Houtkooper et al., 2013) knockdowns (Figure S2D).

Figure 2. jmjd-1.2 and jmjd-3.1 are Necessary and Specific for Induction of the UPR^mt.

(A) Fluorescent micrographs of hsp-6p∷gfp UPR^mt reporter animals treated with the indicated RNAi at day 1 of adulthood. Knockdown of jmjd-1.2 suppresses cco-1-mediated UPR^mt induction in hsp-6p∷gfp reporter animals.

(B) Knockdown of jmjd-3 histone demethylase family members suppresses spg-7-mediated UPR^mt induction.

(C) and (D) Fluorescent micrographs of hsp-4p∷gfp UPR^er reporter animals. Induction of the UPR^er response in hsp-4p∷gfp UPR^er reporter animals by tunicamycin treatment is not affected by neither jmjd-1.2 (C) or jmjd-3.1 (D) RNAi.

(E) and (F) Fluorescent micrographs of hsp-16.2p∷gfp reporter animals. Induction of the heat shock response in hsp-16.2p∷gfp reporter animals occurs independently of jmjd-1.2 (E) or jmjd-3.1 (F) RNAi.

(G) and (H) Fluorescent micrographs of sod-3∷gfp reporter animals treated with the indicated RNAi. daf-16 RNAi was used as a positive control. See also Figures S2 and S3.

We tested additional JmjC-containing proteins for their effects on UPR^mt following mitochondrial stress (Figure S2A). Interestingly, RNAi against the H3K27 demethylase family members jmjd-3.2, jmjd-3.3 and utx-1 also substantially decreased UPR^mt activation, while other JmjC-domain containing demethylases had no significant effect on UPR^mt activation (Figures 2B, S2A). As an alternative approach, we used the H3K27 demethylase inhibitor GSK-J4 (Heinemann et al., 2014) to examine potential activity towards UPR^mt induction. GSK-J4 treatment of hsp-6p∷gfp reporter animals grown on mrps-5 RNAi suppressed UPR^mt activation (Figure S2E) indicating that H3K27 demethylase activity was required for the perpetuation of mitochondrial stress signaling. As H3K27 methylation can be established by the Polycomb repressive complex 2 (PRC2) in C. elegans (Bender et al., 2004), we asked whether reduction of polycomb components is sufficient to trigger the UPR^mt. However, knockdown of individual PRC2 components mes-2, mes-3 and mes-6 in C. elegans did not activate UPR^mt signaling (Figure S2F).

We next examined a potential contribution of jmjd-1.2 and jmjd-3.1 on additional cellular stress responses including the unfolded protein response of the endoplasmic reticulum (UPR^er) (Walter and Ron, 2011) and the cytosolic heat shock response (HSR) (Morimoto, 2011). RNAi against jmjd-1.2 did not affect induction of UPR^er target genes, including the ER chaperone BiP (hsp-4), in response to UPR^er stimulation with tunicamycin (Figures 2C, S3A). In contrast, jmjd-3.1 RNAi revealed minor but statistically significant effects on several UPR^er target genes, which became evident only upon monitoring gene expression independent of the hsp-4p∷gfp reporter (Figure 2D, S3B), consistent with recent findings (Labbadia and Morimoto, 2015). Moreover, RNAi against jmjd-1.2 or jmjd-3.1 did not affect the heat shock-induced expression of the HSR chaperone hsp-16.2 (Figures 2E, F; S3C, D). Finally, jmjd-1.2 and jmjd-3.1 had no effect on the expression of the antioxidant enzyme sod-3, which is induced by reduced insulin/IGF-1 signaling (Figures 2G, H). Collectively, these data indicate that jmjd-1.2 specifically modulates the UPR^mt, while jmjd-3.1 appears to be moderately involved in the transcriptional response to ER stress.

Overexpression of jmjd-1.2 or jmjd-3.1 is Sufficient for Lifespan Extension and UPR^mt Induction

The results of the previous experiments indicated that both jmjd-1.2 and jmjd-3.1 were required for ETC-mediated longevity and UPR^mt induction. We next tested their sufficiency to extend lifespan and increase mitochondrial stress signaling. We generated transgenic strains expressing jmjd-1.2 under control of the ubiquitous sur-5 promoter and monitored lifespan. For control, we generated an enzymatically inactive version of jmjd-1.2 that harbors a point mutation in the catalytic histidine of the JmjC domain (jmjd-1.2^H508A). jmjd-1.2 overexpression significantly increased longevity when compared to wild type animals (Figure 3A), while the catalytically-inactive jmjd-1.2^H508A had no effect (Figure S4A). We also generated strains overexpressing jmjd-3.1 under the control of its endogenous promoter (jmjd-3.1p∷jmjd-3.1) and found these animals to be long-lived (Figure 3B), in agreement with previous reports (Labbadia and Morimoto, 2015).

Figure 3. jmjd-1.2 and jmjd-3.1 Overexpression is Sufficient for Lifespan Extension and UPR^mt Induction.

(A) and (B) Overexpression of jmjd-1.2 and jmjd-3.1 extends C. elegans lifespan. Lifespan analysis of two independent transgenic lines of sur-5p∷jmjd-1.2 (A) or jmjd-3.1p∷jmjd-3.1 (B) expressing animals compared to WT (N2) animals.

(C) and (D) Fluorescent micrographs of hsp-6p∷gfp UPR^mt reporter animals expressing either sur-5p∷jmjd-1.2 (C) or jmjd-3.1p∷jmjd-3.1 (D) transgenes in two independent lines analyzed at day 1 of adulthood.

(E) and (F) Transcript levels in two independent lines of sur-5p∷jmjd-1.2 (E) or jmjd-3.1p∷jmjd-3.1 (F) expressing animals at day 1 of adulthood were measured by qRT-PCR. Results are shown relative to transcript levels in WT (N2) animals, with error bars indicating mean ± SEM. *** denotes p < 0.001, **p < 0.01, *p < 0.05.

(G) Neuron-specific overexpression of jmjd-1.2 is sufficient to extend lifespan. Lifespan analysis of two independent lines expressing either pan-neuronal (rgef-1p∷jmjd-1.2) or intestinal (gly-19p∷jmjd-1.2) transgenes compared to WT (N2) animals.

(H) Transcript levels in animals expressing either pan-neuronal (rgef-1p∷jmjd-1.2) or intestinal (gly-19p∷jmjd-1.2) transgenes at day 1 of adulthood were measured by qRT-PCR. Results are shown relative to transcript levels in WT (N2) animals, with error bars indicating mean ± SEM. ** denotes p < 0.01, *p < 0.05, n.s. = p > 0.05. See also Figure S4 and Table S1 for lifespan statistics.

Intriguingly, overexpression of either jmjd-1.2 or jmjd-3.1 was sufficient to induce the UPR^mt (Figures 3C-F). To identify the tissues in which expression was required, we generated transgenic strains expressing jmjd-1.2 in either neurons or the intestine. Neuronal overexpression of jmjd-1.2 was sufficient for lifespan extension and UPR^mt activation, while expression in the intestine did not (Figures 3G, H, S4B). Similarly, overexpression of the catalytically inactive jmjd-1.2^H508A variant failed to activate the UPR^mt (Figure S4B). Moreover, jmjd-1.2 overexpression had no effect on mRNAs levels of UPR^er and HSR target genes indicating that jmjd-1.2 selectively modulates the UPR^mt (Figures S4C, D). jmjd-3.1 overexpression, however, lead to a significant increase in expression of some UPR^er target genes, but not of the HSR (Figures S4E, F).

The UPR^mt Is a Genetic Requirement for JmjC demethylase-mediated Longevity

The ubiquitin-like protein UBL-5 is required for increased longevity of the ETC mutant isp-1(qm150) (Durieux et al., 2011) (Figure S5A). Therefore, we tested whether the long lifespan of jmjd-1.2 and jmjd-3.1 overexpressing animals depended upon ubl-5. Intriguingly, ubl-5 RNAi fully suppressed increased longevity of both jmjd-1.2 and jmjd-3.1, as well as neuron-specific jmjd-1.2 overexpressing animals (Figures 4A-C). Notably, ubl-5 was not required for increased longevity of animals ubiquitously expressing heat shock factor-1 (HSF-1) (Baird et al., 2014) (Figure S5B) and had no effect on wild type lifespan (Durieux et al., 2011). These data strongly suggest that the UPR^mt is an essential and specific requirement in H3K27 demethylase gain-of-function models.

Figure 4. The UPR^mt is a Genetic Requirement for the Pro-Longevity Response upon jmjd-1.2 and jmjd-3.1 Overexpression.

(A), (B) and (C) ubl-5 RNAi suppresses lifespan extension upon jmjd-1.2, jmjd-3.1 or neuronal jmjd-1.2 overexpression. Lifespan analysis of two independent lines of sur-5p∷jmjd-1.2 (A), jmjd-3.1p∷jmjd-3.1 (B), neuronal rgef-1p∷jmjd-1.2 (C) and WT (N2) animals grown on empty vector control or ubl-5 RNAi.

(D), (E) and (F) Fluorescent micrographs of hsp-6p∷gfp UPR^mt reporter animals expressing the indicated transgenes treated with the indicated RNAi, at day 1 of adulthood.

(G), (H) and (I) Transcript levels of canonical UPR^mt targets assessed by qRT-PCR in sur-5p∷jmjd-1.2 (G), jmjd-3.1p∷jmjd-3.1 (H) or neuronal rgef-1p∷jmjd-1.2 (I) transgenic animals at day 1 of adulthood treated with the indicated RNAi. Results are shown relative to transcript levels in WT (N2) animals grown on the indicated RNAi, with error bars indicating mean ± SEM. *** denotes p < 0.001, **p < 0.01, *p < 0.05, n.s. = p > 0.05. See also Figure S5 and Table S1 for lifespan statistics.

Since both jmjd-1.2 and jmjd-3.1 overexpression induced the UPR^mt, we also tested the requirement of the UPR^mt core components clpp-1 and atfs-1 for UPR^mt activation in response to demethylase overexpression. RNAi against atfs-1 and clpp-1 in strains overexpressing jmjd-1.2 or jmjd-3.1 efficiently abrogated UPR^mt signaling, suggesting that jmjd-1.2 and jmjd-3.1 function upstream of the core transcriptional UPR^mt machinery in C. elegans (Figures 4D-I).

JMJD-1.2 and JMJD-3.1 Overexpression Recapitulates the Transcriptional Response to Mitochondrial Stress

To understand the molecular mechanisms that underlie the extended longevity in response to mitochondrial ETC perturbation and histone demethylase overexpression on a whole genome level, we performed RNA-seq analysis of strains ubiquitously overexpressing jmjd-3.1 or jmjd-1.2, or neuronally-expressed jmjd-1.2, and compared them to cco-1 RNAi treated animals. Our analysis identified ∼3,000-7,500 differentially expressed genes (DEGs) in each condition (adjusted p-value < 0.05), with the greatest number of changes observed in the jmjd-3.1p∷jmjd-3.1 strain (Figure 5A). Strikingly, almost half the genes differentially expressed upon cco-1 RNAi treatment (1405/2979) were also significantly differentially expressed in all three demethylase overexpression strains, and 84% (2505/2979) were in common with at least one of the examined overexpression strains (Figure 5A). Among the overlapping DEGs, 99% (1385/1405) of gene expression changes go in the same direction in all 4 conditions (Figure 5E). These data indicate that an overwhelming majority of the gene expression changes induced by cco-1 RNAi are recapitulated by overexpression of jmjd-1.2 or jmjd-3.1. Furthermore, an examination of all nine JmjC-domain encoding genes revealed that jmjd-1.2 and jmjd-3.1 were specifically upregulated in response to cco-1 RNAi treatment (Figure 5B). These data place jmjd-1.2 and jmjd-3.1 downstream of mitochondrial stress and support a model in which a majority of the transcriptional response to mitochondrial stress is mediated by jmjd-1.2 and jmjd-3.1.

Figure 5. ETC Perturbation and JMJD Overexpression Share Common Lifespan Extension Mechanisms.

(A) Venn diagram of differentially expressed genes (DEGs) in cco-1 RNAi treated worms and transgenic overexpression lines jmjd-3.1p∷jmjd-3.1, sur-5p∷jmjd-1.2, rgef-1p∷jmjd-1.2, compared to wild type N2 worms on empty vector control, as measured by RNA-seq (Benjamini-Hochberg adjusted p value < 0.05). See Table S2 for complete list of DEGs.

(B) Transcriptional upregulation of jmjd-1.2 and jmjd-3.1 upon cco-1 RNAi. Gene expression analysis of all nine JmjC domain encoding genes in RNA-seq samples, expressed as fold change relative to wild type N2 on empty vector control. Results are expressed as mean ±SEM of normalized count values (n=3, Benjamini-Hochberg adjusted p values (padj) calculated by DESeq2, jmjd-1.2 padj=6.93E-09, jmjd-3.1 padj=0.002, jmjd-3.3 padj=0.452, jmjd-4 padj=0.085. n.d.= not detected, ** denotes p < 0.01, ***p < 0.001).

(C) UPR^mt gene expression in RNA-seq samples, expressed as fold change relative to wild type N2 on empty vector control. Results are expressed as mean ±SEM of normalized count values. * denotes p < 0.05, **p < 0.01, ***p < 0.001).

(D) Representative top GO terms of upregulated and downregulated genes in the 1405 overlapping DEGs (Bonferroni adjusted p value<0.05). See also Table S2.

(E) Gene expression heatmap of 1405 overlapping DEGs in all 4 conditions described in(A). DESeq2-normalized count values were used for calculations. The indicated Row Z-scores reflect the number of standard deviations each replicate is apart from the mean gene expression value over all conditions. See also Table S2.

(F) Heatmap of 470 mitochondrial genes from GO cellular component category mitochondrion (GO:0005739). Fold change (FC) was calculated by comparing normalized count values of each condition to wild type N2 empty vector control, then transformed to log2 scale. See also Table S2.

(G) Heatmap of 111 OXPHOS genes from GO biological process category oxidative phosphorylation (GO:0006119) and manual annotation. See also Table S2.

The 1405 overlapping genes were subjected to Gene Ontology (GO) analysis using the DAVID database (Huang et al., 2009). Among the upregulated genes, signaling and cell communication related terms were enriched (Figure 5D). Importantly, larval development and growth, as well as multiple protein processing pathways, such as translation and proteasome were extensively downregulated, consistent with global remodeling of the proteome in response to stress (Figure 5D). Whereas the majority of mitochondrial genes were significantly reduced in all examined conditions (Figure 5F, G), many of the UPR^mt components, such as the chaperone hsp-6, the proteases ymel-1 and spg-7 as well as the transcriptional regulator dve-1 were induced (Figure 5C). Remarkably, OXPHOS genes in jmjd-1.2 and jmjd-3.1 overexpression strains were downregulated to a similar extent as in cco-1 RNAi treated animals (Figure 5G). Surprisingly, there were only a few OXPHOS components that were significantly upregulated in all conditions, specifically cyc-1 from complex III and mtDNA encoded nduo-2, nduo-4 and nduo-5. Intriguingly, this finding is reminiscent of mito-nuclear protein imbalance, a phenomenon of contrasting expression between nuclear- and mtDNA encoded OXPHOS components which tightly couples UPR^mt activation with longevity across organisms (Houtkooper et al., 2013).

Mammalian PHF8 and JMJD3 Correlate with Lifespan and UPR^mt Activation

To assess the potential roles of jmjd-1.2 and jmjd-3.1 in regulating UPR^mt and longevity in mammals, we investigated their murine homologs Phf8 and Jmjd3, respectively, in the GeneNetwork database (www.genenetwork.org), which contains a vast collection of clinical and molecular (transcript and protein expression) phenotypes from the BXD mouse genetic reference panel (GRP) (Andreux et al., 2012; Wu et al., 2014). Importantly, variations within these datasets reflect mild, natural variations in gene expression patterns found in isogenic populations and are not reliant upon more deleterious genetic manipulations of mitochondrial function.

Both the transcripts for Phf8 and Jmjd3 showed high levels of variability in expression across the tissues and strains examined (Figure 6A). In tissue-specific datasets, natural variations in Phf8 expression in the hypothalamus, spleen, and amygdala positively correlated with Jmjd3 expression (Figure 6B), suggesting a correlative genetic interaction between the two enzymes. Importantly, expression levels of both histone demethylases were also positively correlated with UPR^mt related genes, including Hspd1 (encoding the mitochondrial chaperone HSP60), Satb1, Satb2 (orthologs of the C. elegans UPR^mt regulator dve-1), Abcb10 (ortholog of the mitochondrial peptide exporter haf-1) (Haynes et al., 2010) and Ubl5 (ortholog of ubl-5) across an array of tissues (Figures 6C, D). The strongest associations were observed in the hypothalamus, in which UPR^mt related genes in addition to Phf8 and Jmjd3 formed a connected correlation network (Figure 6E). In accordance with the impact of jmjd-1.2 and jmjd-3.1 on lifespan regulation in C. elegans, we observed correlations between Phf8 and Jmjd3 expression and lifespan in the spleen and hypothalamus or pituitary and adrenal glands of mice, respectively (Figures 6F, G).

Figure 6. Positive Correlations between Phf8, Jmjd3, Lifespan and UPR^mt transcripts in the BXD Mouse Genetic Reference Population.

(A) Variation of Phf8 and Jmjd3 mRNA levels in hypothalamus (n=44) and amygdala(n=56) across BXD mouse strains. Each bar represents mRNA levels from a pool of approximately five animals per strain.

(B) Positive correlations between Phf8 (y-axis) and Jmjd3 (x-axis) expression in hypothalamus (n=44), amygdala (n=56) and spleen (n=67).

(C) Positive correlations between Phf8 (y-axis) and selected UPR^mt genes (x-axis) transcripts in various tissues (n=46 for adrenal gland, n=49 for pituitary gland).

(D) Positive correlations between Jmjd3 (y-axis) and selected UPR^mt genes (x-axis) transcripts in various tissues (n=46 for liver).

(E) Spearman's correlation co-expression network for Phf8, Jmjd3 and UPR^mt genes in hypothalamus. Blue correlations are positive, red correlations are negative – intensity of the colors corresponds to correlation coefficients.

(F) Pearson correlations of Lifespan versus Phf8 transcript levels in either spleen (left) or hypothalamus (right) of BXD mice.

(G) Pearson correlations of Lifespan versus Jmjd3 transcript levels in either pituitarygl and (left) or adrenal gland (right) of BXD mice.

Across strains expressing variable levels of Phf8 mRNA, using immunoblotting, we found a correlative change of PHF8 protein, which was paralleled by a reduction of global H3K27me2 levels (Figure 7A). Increased PHF8 protein levels were also associated with higher abundance of UPR^mt marker proteins, such as the chaperones mtHSP70 and HSP60 (Figure 7B), reminiscent of the transcript correlations (Figure 6C). Similarly, higher JMJD3 levels were associated with an increase of HSP60 protein (Figure 7C).

Figure 7. Conserved UPR^mt Gene Regulation Mechanisms Through H3K27 Demethylases.

(A) and (B) Immunoblot analysis of tissue protein lysates using the indicated antibodies. Increased Phf8 expression correlates with reduced H3K27me2 (A) and higher levels of mitochondrial chaperones (B) in the hypothalamus of the indicated BXD mouse strains. β-actin and Histone H3 were used as loading controls (upper panels). Densitometric quantifications of immunoblot signals normalized to either β-actin or Histone H3 (lower panels). Data represent the mean+ SEM. *** denotes p < 0.001, **p < 0.01.

(C) Immunoblot analysis of BXD liver tissue protein lysates using the indicated antibodies. HSP90 was used as a loading control.

(D) Heatmap of selected UPR^mt transcripts in WT and Jmjd3 KO mice embryos at E9.5 (GSE40332). Low expression is shown in blue, while high expression is in red.

(E) Heatmap of fold change in UPR^mt transcripts upon shJMJD3 treatment relative to shRenilla in human T-cell lymphoblastic leukemia CEM and CUTLL1 cell lines (GSE56696).

(F) Heatmap of selected UPR^mt transcripts in the human T-cell lymphoblastic leukemia CUTLL1 cell line upon treatment with the H3K27 demethylase inhibitor GSK-J4 (GSE56696).

(G) ChIP-Seq profiles of H3K27me3 enrichment at selected UPR^mt genes in the CUTLL1 cell line upon shJMJD3 and shRenilla treatments (GSE56696).

(H) ChIP-Seq profiles of PHF8 binding at selected UPR^mt gene promoters in HeLa and 293T cells (GSE20725).

(I) ChIP-qRT-PCR analysis of H3K27me3 enrichment at UPR^mt genes at L3 stage of cco-1 RNAi treated worms compared to empty vector control. IgG antibody was used as a control. Results are expressed as percent of input, with error bars indicating mean ± SEM (n=7, * denotes p < 0.01).

We hypothesized that a relationship between the expression of markers of UPR^mt, Phf8, and Jmjd3 also might be detectable in mammalian Jmjd3 loss-of-function experiments, which have been published previously (GEO datasets GSE40332, GSE56696) (Ntziachristos et al., 2014). Consistent with our prediction, in Jmjd3 knockout mouse embryos (GSE40332), the expression of various UPR^mt related transcripts was decreased (Figure 7D). Likewise, knockdown of JMJD3 by shRNA in two human T-cell lymphoma cell lines (CUTLL1 and CEM) decreased the expression levels of multiple transcripts related to the UPR^mt (Figure 7E). Similarly, treatment of CUTLL1 cell line with the H3K27 demethylase inhibitor GSK-J4 (GSE56696) (Ntziachristos et al., 2014) also decreased the levels of multiple UPR^mt transcripts in a time-dependent fashion (Figure 7F). Overall, these data in both the murine GRP, mice with loss-of-function alleles, or cell lines with modified levels or activity of PHF8 and/or JMJD3 collectively suggest that expression levels of these demethylases positively correlate with UPR^mt expression.

Finally, we examined ChIP-seq analysis of CUTLL1 cells treated with shJMJD3 (GSE56696) for evidence of direct JmjC regulation of chromatin at coding regions of UPR^mt target genes. In these analyses, we found that H3K27me3 enrichment on HSPD1, HSPE1, and SATB1 genes was increased upon JMJD3 knockdown (Figure 7G), indicating that the repressive H3K27me3 mark may be actively removed from the coding regions of these genes by JMJD3, thereby allowing for their expression. In another independent ChIP-seq dataset (GSE20725) (Fortschegger et al., 2010), PHF8 was found to bind to coding regions of UPR^mt genes HSPD1, HSPE1 and UBL5 in both HeLa and 293T cells (Figure 7H).

In addition to these in silico analyses, we examined H3K27me3 occupancy at coding regions of UPR^mt regulators during mitochondrial stress in larval stage C. elegans by ChIP-qRT-PCR. We observed a substantial decrease in H3K27me3 abundance on hsp-6, clpp-1 and atfs-1 genes upon cco-1 RNAi during early larval development (Figure 7I). Taken together, these data suggest that PHF8 and JMJD3 may be conserved positive regulators of UPR^mt and lifespan from C. elegans to mammals, through the modulation of the H3K27 methylation status at coding regions of key UPR^mt genes.

Discussion

The distinct timing requirements of lifespan extension due to mild mitochondrial dysfunction during larval development led to the early proposal of an epigenetic mechanism that determines ETC-mediated longevity of C. elegans (Dillin et al., 2002a; Durieux et al., 2011). Remarkably, the discovery that the UPR^mt is not only a genetic requirement but also shares these overlapping timing requirements with ETC-mediated longevity reinforces the idea that primary mitochondrial perturbations establish an epigenetic memory which sets the rate of aging of the entire organism and protects its mitochondrial proteome from future insults (Durieux et al., 2011). The identification of jmjd-1.2/PHF8 and jmjd-3.1/JMJD3 provides a molecular explanation for these observations and suggests the epigenetic regulation of transcriptional outputs during mitochondrial stress. We find jmjd-1.2 to be not only necessary and specific, but also sufficient for both induction of the UPR^mt and extension of lifespan. jmjd-3.1 is also necessary and sufficient for UPR^mt induction and lifespan extension, but has overlapping roles in ER stress regulation (Labbadia and Morimoto, 2015).

The two histone demethylases differ in their substrate specificity: While jmjd-3.1/JMJD3 is associated with the removal of H3K27me2/3 epigenetic marks (Agger et al., 2007), jmjd-1.2/PHF8 has activity towards a wider range of substrates, including H3K9me1/2, H3K27me2 and H4K20me1 (Feng et al., 2010; Fortschegger et al., 2010; Kleine-Kohlbrecher et al., 2010; Liu et al., 2010; Qi et al., 2010). Both genes are associated with the positive regulation of gene expression and the removal of repressive marks. This raises the intriguing possibility that both enzymes might function in a linear pathway to sequentially demethylate H3K27me3 and thereby activate gene expression in response to mitochondrial perturbations. This hypothesis is supported by the mitochondrial stress-induced reduction of H3K27me3 occupancy at UPR^mt response genes during larval development in worms (Figure 7I). Our results are therefore consistent with the idea that removal of repressive marks might allow access of the core UPR^mt transcriptional machinery to induce mitochondrial stress signaling (Haynes et al., 2010; Nargund et al., 2012).

How is mitochondrial stress sensed by jmjd-1.2/PHF8 and jmjd-3.1/JMJD3? An attractive idea is that both jmjd-1.2 and jmjd-3.1 themselves are targets of the transcriptional response to mitochondrial stress as indicated in our transcriptomics analyses (Figure 5B). Another possibility is that acute adaptations in mitochondrial metabolism may promote activity of JmjC demethylases. Since both JMJD3 and PHF8 belong to the family of 2-oxoglutarate dependent oxygenases, it appears likely that elevated levels of the TCA cycle intermediate alpha-ketoglutarate (α-KG) might contribute to their increased activity (Teperino et al., 2010). Intriguingly, the exogenous supplementation of α-KG has recently been found to extend lifespan of C. elegans (Chin et al., 2014). While RNAi knockdown experiments for both jmjd-1.2 and jmjd-3.1 revealed robust suppression of ETC-mediated longevity and UPR^mt induction, we did not find strong effects on lifespan and hsp-6 transcriptional induction in the analysis of the respective mutant strains suggesting that an acute depletion of the enzymes during larval development rather than chronic deficiency is necessary to unmask their role for these phenotypes (Figure S6).

In this work, we identify neurons as a key tissue to promote longevity and UPR^mt induction upon neuron-specific jmjd-1.2 overexpression. Neuronal jmjd-1.2 is not only sufficient to mediate a robust UPR^mt induction but also extends C. elegans lifespan in an UPR^mt-dependent manner. These findings were corroborated in mice and appear to be conserved in the BXD mouse reference population. Across various neuronal subregions, high levels of PHF8 and JMJD3 correlate with increased expression of mammalian UPR^mt core components (CLPP, ABCB10, SATB1/2, UBL5) and downstream mitochondrial chaperones HSP60 (HSPD1) and mtHSP70 (HSPA9) illustrating that both demethylases might control an integrated transcriptional network promoting mammalian longevity.

We previously demonstrated that mito-nuclear protein imbalance induces a robust UPR^mt and is linked to increased lifespan, both in nematodes and in BXD mice (Houtkooper et al., 2013). Interestingly, mito-nuclear imbalance might also play a role in longevity mediated by overexpression of jmjd-1.2 and jmjd-3.1, as suggested by our RNA-seq analysis (Figure 5G). As further proof of concept that similar stressors could activate the UPR^mt across species, we recently found that expression of Cox5b and Spg7 correlate negatively with the UPR^mt network, indicating that their low abundance likely triggers the UPR^mt in mammals (Wu et al., 2014). These data indicate that the UPR^mt pathway is active in vivo in mammals under physiological, non-stress conditions. Based on the observed positive correlations between PHF8, JMJD3 and UPR^mt in multiple tissues, we suggest that the regulation of UPR^mt by the H3K27 demethylases PHF8 and JMJD3 is also conserved in BXD mice under basal conditions. Of note, this association between histone demethylases and UPR^mt seems in certain tissues linked to mouse lifespan regulation, although further mechanistic work is required to ascertain this relation. Our bioinformatics analysis and literature discussed above suggests that PHF8 and JMJD3 regulate UPR^mt genes by removing repressive H3K27 methylation marks from their coding regions. Identifying the exact mechanism of regulation and whether there is a developmental aspect of the UPR^mt pathway in mammals, as demonstrated in C. elegans (Durieux et al., 2011), remains an important direction for future work.

Collectively, our data corroborate the increasing body of literature in which epigenetic marks that control chromatin states, including histone methylation, represent a hallmark of aging (Greer et al., 2010; Han and Brunet, 2012; Lopez-Otin et al., 2013; Rando and Chang, 2012). Mitochondrial perturbations early in life have long-lasting effects on gene expression, and within this work we provide a mechanistic understanding of how this might be achieved. Our results thus reveal a conserved mode for the regulation of stress response and lifespan dependent on mitochondrial function.

Experimental Procedures

Additional details are provided in the Extended Experimental Procedures.

Lifespan Analysis

Lifespan experiments were conducted at 20°C as previously described (Durieux et al., 2011). Lifespans were performed on nematodes fed HT115 bacteria expressing the indicated RNAi, using the pre-fertile period of adulthood as day 0. Animals were transferred to fresh plates every other day until day 12. Prism 5 software was used for statistical analysis to determine significance calculated using the log-rank (Mantel-Cox) method. Lifespan experiments involving RNAi shifting to Dicer (dcr-1) RNAi were performed as described (Dillin et al., 2002a, 2002b). Briefly, for lifespans with selective RNAi only during development, animals were grown on cco-1+jmjd-1.2 or cco-1+jmjd-3.1 RNAi bacteria and transferred to dcr-1 dsRNA expressing bacteria at L4 stage. For lifespans with RNAi during adulthood, animals were grown on empty vector RNAi control bacteria and transferred to cco-1+jmjd-1.2 or cco-1+jmjd-3.1 RNAi bacteria at L4 stage.

Gene Expression Analysis

C. elegans were age synchronized by egg bleaching and cultivated on nematode growth (NGM) plates containing HT115 bacteria expressing the indicated RNAi constructs at 20°C and harvested at day 1 of adulthood. Animals were collected in M9 buffer, centrifuged at 1000 × g for 30 sec, resuspended in Trizol (Life Technologies) and snap frozen in liquid nitrogen. After several freeze-thaw cycles, total RNA was isolated using the RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions. 1 μg of total RNA was subjected to cDNA synthesis using the QuantiTect Reverse Transcription Kit (Qiagen). Quantitative Real-Time PCR reactions were performed with the SYBR Select Master Mix (Applied Biosystems) in Optical 384-well MicroAmp plates (Applied Biosystems) using a QuantStudio 6 Flex (Applied Biosystems).

Fluorescence Microscopy

For fluorescence microscopy, animals were blindly chosen under the light microscope (at random) from a population, immobilized with 100 μg/ml levamisole (Sigma) and images were then captured using a Leica M250FA automated fluorescent stereo microscope equipped with a Hamamatsu ORCA-ER camera.

Immunoblot Analysis

Mouse tissue extracts were prepared in modified RIPA buffer using a hand-held homogenizer (UltraTurrax). Crude lysates were centrifuged at 10,000 × g at 4°C for 10 min and total protein amount was determined with the DC Protein Assay (BioRad). Supernatants were supplemented with 4x SDS sample buffer, boiled for 5 min at 95°C and resolved by standard SDS-PAGE. Proteins were transferred to PVDF membranes (Immobilon). Equal loading was assessed with anti-β-actin (Abcam) antibodies.