H3K9me2 protects lifespan against the transgenerational burden of germline transcription in C. elegans

During active transcription, the COMPASS complex methylates histone H3 at lysine 4 (H3K4me). In Caenorhabditis elegans, mutations in COMPASS subunits, including WDR-5, extend lifespan and enable the inheritance of increased lifespan in wild-type descendants. Here we show that the increased lifespan of wdr-5 mutants is itself a transgenerational trait that manifests after eighteen generations and correlates with changes in the heterochromatin factor H3K9me2. Additionally, we find that wdr-5 mutant longevity and its inheritance requires the H3K9me2 methyltransferase MET-2 and can be recapitulated by a mutation in the putative H3K9me2 demethylase JHDM-1. These data suggest that lifespan is constrained by reduced H3K9me2 due to transcription-coupled H3K4me. wdr-5 mutants alleviate this burden, extending lifespan and enabling the inheritance of increased lifespan. Thus, H3K9me2 functions in the epigenetic establishment and inheritance of a complex trait. Based on this model, we propose that lifespan is limited by the germline in part because germline transcription reduces heterochromatin.


INTRODUCTION 14
Lifespan is governed by complex interactions between genetics and the environment. Despite its 15 complexity as a trait, lifespan is limited by the germline in a wide range of metazoans. For example, genetic 16 or physical ablation of the germline extends lifespan in C. elegans (Arantes-Oliveira et al. 2002;Flatt et al. 17 2008;Hsin & Kenyon 2009;Kenyon 2010). These observations led to the disposable soma theory of aging, 18 which posits that resources are shifted to the germline to promote progeny fitness at the expense of 19 maintaining the parental soma (Kenyon 2010;Kirkwood & Holliday 1979). Some have theorized that one 20 cost of a germline is transcription during gametogenesis, but the molecular details of how germline 21 transcription affects lifespan remain elusive (Ghazi, Henis-Korenblit, & Kenyon 2009;Greer et al. 2010). 22 Regions of active transcription are marked by post-translational histone modifications like the 23 methylation of histone 3 at lysine 4 (H3K4me) (Shilatifard 2008). H3K4me is deposited by the 24 MLL/COMPASS complex, which travels with elongating RNA Polymerase II during transcription (Wood et al. 25 2007). In the nematode C. elegans, animals with reductions in COMPASS complex subunits (wdr-5, ash-2, 26 and the methyltransferase set-2) live longer than wild-type individuals (Greer et al. 2010). The lifespan 27 extension of COMPASS mutants requires the presence of an actively proliferating germline, suggesting that 28 COMPASS acts specifically in the germline to limit lifespan (Greer et al. 2010). Subsequent work showed 29 that reducing COMPASS solely in the germline activates a fatty acid desaturation pathway in the intestine 30 that increases lifespan, in part by down-regulating an S6 kinase normally expressed in the germline (Han et 31 al. 2017). Activation of this pathway causes the accumulation of mono-unsaturated fatty acids, and this 32 increase is sufficient to extend lifespan (Han et al. 2017). 33 Greer and colleagues also show that longevity in COMPASS mutants was a heritable trait. The 34 lifespan extension of COMPASS mutants was inherited by their wild-type descendants for four generations, 35 before reverting to wild-type lifespan (Greer et al. 2011). The transgenerational inheritance of longevity 36 suggests that it is an epigenetic trait, but the mechanism of inheritance remains unknown. In particular, it is 37 difficult to reconcile how reduced H3K4me in COMPASS mutants can block COMPASS from restoring a 38 normal chromatin environment in wild-type descendants of COMPASS mutants. Thus, the longevity in 39 COMPASS mutants may be due to another heritable factor. 40 H3K9me2 is classically considered a repressive modification and often associated with 41 heterochromatin. In C. elegans, H3K9me2 and H3K9me3 each have distinct roles in the genome, with 42 H3K9me2, rather than H3K9me3, most closely associated with canonical heterochromatin factors like HP1 43 (Garrigues et al. 2015;Liu et al. 2011;Mcmurchy et al. 2017). Disruption of heterochromatin has been 44 broadly linked to aging, with evidence from budding yeast, C. elegans, Drosophila, mammals, and human 45 diseases (Haithcock et al. 2005;Kaeberlein et al. 1999;Ni et al. 2012;Shumaker et al. 2006;Tsurumi & Li 46 2012;Villeponteau 1997;Wood et al. 2010;Wood & Helfand 2013). Additionally, mutations that increase 47 repressive chromatin extend lifespan (Jin et al. 2011;Kennedy et al. 1995;Larson et al. 2012;Maures et al. 48 2011). We previously found that a number of examined loci have increased levels of H3K9me2 when the 49 COMPASS subunit WDR-5 is mutated (Kerr et al. 2014). Therefore, we investigated the role of repressive 50 H3K9me2 in the longevity of wdr-5 mutants over generational time. 51 52 RESULTS 53 Transgenerational longevity in wdr-5 mutants 54 It has previously been reported that animals mutant for genes encoding components of the 55 COMPASS complex have a lifespan extension of up to 28% (Greer et al. 2010), and this lifespan extension is 56 inherited by wild-type descendants of these mutants before reverting back to wild-type levels in the fifth 57 generation (Greer et al. 2011). To investigate the nature of this inheritance, we first attempted to 58 recapitulate the original observation using fertile, homozygous wdr-5 (ok1417) mutants. Since the 59 transgenerational effects of a wdr-5 mutation lasts for four generations, we crossed wdr-5 mutants to wild 60 type to generate heterozygous wdr-5/+ progeny and maintained populations as wdr-5/+ heterozygotes for 61 five generations. We then used homozygous wdr-5 mutant progeny from F5 wdr-5/+ heterozygotes as our 62 P0 founding population, comparing them to P0 wild-type animals descended from survivors recovered from 63 a thaw. In contrast to prior observations, P0 wdr-5 mutants were never long-lived compared to their wild-64 type counterparts (as observed in eight biological replicates) ( The initial observations of wdr-5 mutant longevity did not report how many generations lacking 66 WDR-5 activity were required to reach the 28% lifespan extension in wdr-5 mutants, so we considered the 67 possibility that lifespan may gradually change over a number of generations in COMPASS mutants. Starting 68 with P0 as described above, we followed a single population of wdr-5 mutants and assessed lifespan every 69 three generations. For the first six generations (which are hereafter referred to as early-gen populations), 70 wdr-5 mutants had slightly shorter lifespans than wild-type populations of the same generation, though the 71 decrease was not statistically significant ( we conclude that wdr-5 mutants had longer lifespans, but only after many generations of lacking  In many taxa, fecundity is inversely correlated with lifespan (Kenyon 2010). To determine whether 82 the appearance of longevity in wdr-5 mutant populations correlates with a decrease in fecundity, we 83 measured progeny number across generational time. It has previously been shown that wdr-5 mutants 84 have decreased broods and increased embryonic lethality (Li & Kelly 2011;Simonet et al. 2007). Consistent 85 with prior observations, early-gen wdr-5 mutants had 27-47% fewer progeny than wild-type populations 86 (supplementary table 2). However, we did not observe a correlation between the onset of longevity and the 87 decrease in progeny number, as long-lived late-gen wdr-5 mutants did not have significantly fewer progeny 88 than early-or mid-gen populations (Figure 1 -figure supplement 2A and B, P = 0.11 and P = 0.13, 89 respectively, unpaired t test). This result suggests that lifespan extension in wdr-5 mutants was not a direct 90 consequence of reduced fecundity, despite the fact that it requires a proliferative germline (Greer et al. 91 2010). 92 The appearance of longevity in wdr-5 mutants could be caused by the acquisition of background 93 mutations in lifespan-determining genes, but several observations make this scenario unlikely. First, we 94 have repeated the transgenerational analysis of lifespan in these populations seven additional times. In 95 each replicate, lifespan increased gradually between early-, mid-, and late-gen populations, with late-gen 96 wdr-5 mutants consistently living longer than late-gen wild-type populations (Figure 1 -figure supplement 97 1). Additionally, when long-lived late-gen wdr-5 mutants were subjected to either starvation or freezing, 98 lifespan reverted back to wild-type levels (Figure 1 -figure supplement 2C, Figure 2). We have also 99 outcrossed long-lived late-gen wdr-5 mutants to wild-type populations, maintained these populations as 100 heterozygotes for five generations, and re-selected new homozygous P0 wdr-5 mutant populations. These 101 P0 wdr-5 mutants reverted back to being short-lived, even as the original late-gen wdr-5 mutant 102 populations remained long-lived (Figure 1 -figure supplement 2D). 103 104 Relative changes in lifespan across generations 105 The determination of longevity in a population is always relative to a control. In our 106 transgenerational lifespan experiments, wdr-5 mutant populations were compared to wild-type 107 populations of the same generation. Therefore, the appearance of longevity could be caused by a relative 108 extension in wdr-5 mutant lifespan, a relative decrease in wild-type lifespan, or both changes. To 109 distinguish between these possibilities, we directly compared the lifespan of early-, mid-, and late-gen wdr-110 5 mutant and wild-type populations in the same lifespan assay ( Figure 2A). Surprisingly, wild-type lifespan 111 decreased over generational time by 12%, primarily between early-and mid-gen populations (Figure 2A'), 112 despite the fact that progeny number did not change in these populations (supplementary table 2). As 113 mentioned previously, we assayed wild type lifespan in descendants of animals recovered from a thaw, 114 which we consider the P0 population. The process of freezing a strain involves an L1 larval diapause 115 induced by starvation (Baugh 2013). The generational change in wild-type lifespan after recovery from a 116 thaw suggested that some aspect of starvation may increase lifespan in descendants, as has been 117 previously shown in populations recovered from L1 diapause or from dauer diapause (Rechavi et al. 2014;118 Webster et al. 2018). To investigate the effect of starvation on wild-type populations, we starved a late-gen 119 population that had already experienced a decrease in lifespan. Wild-type descendants of starved animals 120 lived longer than their non-starved cousins (Figure 1 -figure supplement 2C). 121 In contrast to what occurred in wild-type animals, lifespan in wdr-5 mutants increased over 122 generational time by 20%, primarily between mid-and late-gen populations (Figure 2A''). The decrease in 123 wild-type lifespan was large enough that even early-gen wdr-5 mutants had significantly longer lifespans 124 when compared to those of late-gen wild-type populations (Figure 2A, P < 0.05, log-rank test). However, 125 changes in wild-type lifespan were not the sole driver of longevity in late-gen wdr-5 mutants, since late-gen 126 wdr-5 mutants lived significantly longer than even early-gen wild-type populations (P < 0.001, log-rank 127 test). Overall, concurrent and opposite lifespan changes in both wild type and wdr-5 mutants accounted for 128 the full increase in lifespan in late-gen wdr-5 mutant populations. 129 130 Correlation of longevity with repressive H3K9me2 131 Our previous work demonstrates that levels of H3K9me2 are enriched at certain loci in wdr-5 132 mutants compared to wild type (Kerr et al. 2014). Based on this finding, we asked whether an increase in 133 H3K9me2 could account for the extended lifespan of wdr-5 mutants. To address this possibility, we first 134 compared global levels of H3K9me2 by immunoblot, and chromatin immunoprecipitation followed by next-135 generation sequencing (ChIP-seq). We chose to focus on H3K9me2 because it is more closely associated 136 with canonical heterochromatin factors in C. elegans. By immunoblot of mixed-stage populations, long-137 lived late-gen wdr-5 mutants had a global increase of H3K9me2 compared to late-gen wild-type populations 138 ( Figure 3A). 139 We next examined the genomic enrichment of H3K9me2 using ChIP-seq across generations in wdr-140 5 mutant and wild-type populations. At H3K9me2 peaks, early-gen wdr-5 mutants had slightly less 141 enrichment than their wild-type counterparts ( Figure 3B). In wild type, H3K9me2 enrichment decreased 142 ( Figure 3B') at the same time that lifespan decreased (Figure 2A'), both between early-and mid-gen 143 populations as well as between mid-and late-gen populations. wdr-5 mutants experienced a similar 144 H3K9me2 decrease between early-and mid-gen populations, but H3K9me2 enrichment subsequently 145 increased between mid-and late-gen populations ( Figure 3B (Han et al. 2017). Overall, as lifespan increases across generations in wdr-5 150 mutants compared to wild type (early-to mid-to late-gen) (Figure 1), we observed a corresponding 151 increase in the ratio of mutant to wild-type coverage at each peak ( Figure 3C). 152 Because transgenerational phenotypes must be inherited through the germline, we would expect 153 that H3K9me2 is most affected at germline-expressed genes in wdr-5 mutants. To address this possibility, 154 we examined H3K9me2 at germline-expressed genes (hereafter referred to as germline genes), including 155 those expressed exclusively in the germline and those that are expressed in both the germline and soma (V 156 Reinke et al. 2000). At all genes, H3K9me2 enrichment was low and decreased from early-to late-gen in 157 both wild-type and wdr-5 mutant populations ( Figure 3D). Likewise, at germline genes, wild-type H3K9me2 158 enrichment decreased between early-and late-gen populations ( Figure 3E). In contrast, in wdr-5 mutants, 159 H3K9me2 enrichment at germline genes increased between early-and late-gen populations ( Figure 3E). 160 This increase countered the slight global decrease observed in wdr-5 mutants between early-and late-gen 161 populations ( Figure 3B in wdr-5 mutants, we would expect wdr-5 mutants lacking H3K9me2 to have a short lifespan like met-2 173 mutants. Furthermore, in the continued absence of H3K9me2, they should never be able to acquire 174 longevity, even after many generations without WDR-5. To test these possibilities, we generated wdr-5 175 met-2 double mutants and followed them for more than twenty generations, assessing lifespan 176 periodically. The lifespan of wdr-5 met-2 double mutants consistently resembled that of short-lived met-2 177 mutants, and wdr-5 met-2 mutants never became long-lived ( Figure 4C). 178 179 Increased H3K9me2 and lifespan in jhdm-1 mutants 180 To determine whether higher levels of H3K9me2 can confer longevity directly, rather than as a 181 consequence of changes in H3K4me, we examined mutant animals lacking a predicted H3K9 demethylase, 182 JHDM-1. JHDM-1 has homology to S. pombe Epe1, a putative demethylase that prevents the inheritance of 183 H3K9me2 across cell divisions (Audergon et al. 2015;Ragunathan et al. 2014). Consistent with a role in 184 removing H3K9me, jhdm-1 mutants have higher levels of global H3K9me2 compared to mid-gen wild-type 185 animals by immunoblot ( Figure 4A). To investigate this increase across the genome, we next examined 186 genomic H3K9me2 and compared it to lifespan in jhdm-1 mutant populations across generational time. 187 Early-gen jhdm-1 mutant populations were slightly longer lived than wild type ( Figure 4D), and had similar 188 levels of H3K9me2 enrichment at ChIP-seq peaks ( Figure 4E). By mid generations, jhdm-1 mutants had 189 significantly longer lifespans than their wild-type counterparts, despite having no difference in progeny 190 number ( Figure 4D and supplementary table 2). With lifespans averaging 30% longer than wild type, jhdm-1 191 mutants experienced a more robust longevity effect than wdr-5 mutants (supplementary table 1 and 3). It 192 was notable that mid-gen jhdm-1 populations are significantly long-lived regardless of which generation of 193 wild type was used for comparison ( Figure 4F, P<0.0001, log-rank test). Similar to wdr-5 mutants, the 194 increase in jhdm-1 mutant lifespan corresponded with a genome-wide increase in H3K9me2 enrichment 195 ( Figure 4E). Mid-gen jhdm-1 mutants had more H3K9me2 than either early-gen jhdm-1 mutants or any 196 generation of wild type ( Figure 4E). The increase in mid-gen jhdm-1 mutants was particularly pronounced 197 when examining RPM coverage ratios between jhdm-1 mutants and wild type in mid-versus early-gen 198 populations ( Figure 4G). 199 75% of peaks called in the jhdm-1 mutant transgenerational experiment were shared with those 200 called in the wdr-5 mutant transgenerational experiment (Figure 3 -figure supplement 1C). This overlap 201 indicated that the H3K9me2 increases in both mutants occurred at similar locations in the genome. To 202 further investigate the location of H3K9me2 peaks, we also examined coverage over all genes as well as 203 over germline genes specifically in jhdm-1 mutants. In early-gen populations, wild type and jhdm-1 have 204 similar levels of H3K9me2 at all genes ( Figure 4H), including germline genes ( Figure 4I). Across generations, 205 wild type experienced a decrease in H3K9me2 enrichment at germline genes and this decrease is 206 dependent upon WDR-5 ( Figure 3E). Since the COMPASS complex should be functional in jhdm-1 mutants, 207 we might expect that H3K9me2 would also exhibit a decrease in germline genes across generations. In long-208 lived mid-gen jhdm-1 mutants, we did indeed observe the same reduced enrichment of H3K9me2 at 209 germline genes ( Figure 4I) as we did at all genes ( Figure 4H), although the reduction was not as large as we 210 observed in their wild-type counterparts ( Figure 4I). This H3K9me2 decrease over genes was particularly 211 notable when compared to the overall accumulation of H3K9me2 at all peaks in mid-gen jhdm-1 mutants 212 ( Figure 4E). 213 214 Inheritance of longevity requires H3K9me2 215 Greer and colleagues found that genetically wild-type descendants of long-lived wdr-5 mutants are 216 as long-lived as their mutant ancestors for up to four generations (Greer et al. 2011). Since H3K9me2 is 217 required for the lifespan extension of wdr-5 mutants, it may also be the transgenerational factor inherited 218 by their wild-type descendants. Using long-lived late-gen wdr-5 mutants, we recapitulated the observation 219 that F3 genetically wild-type descendants of wdr-5 mutants (labeled WT (wdr-5)) were as long-lived as their 220 wdr-5 mutant cousins descended from the same population (labeled wdr-5 (wdr-5)) ( Figure 5A and B) 221 (Greer et al. 2011). As originally reported, we found that the lifespan extension of genetically wild-type 222 descendants reverted by generation F5 ( Figure 5C) (Greer et al. 2011). We next tested whether H3K9me2 is 223 required for the inheritance of longevity by removing MET-2 from otherwise wild-type descendants of long-224 lived wdr-5 mutants ( Figure 5D). We used long-lived late-gen wdr-5 mutants to generate homozygous wdr-225 5 mutants that were also met-2/+ heterozygote mutants (labeled "wdr-5"). F3 met-2 mutant descendants 226 of "wdr-5" mutants (labeled met-2 ("wdr-5")) were significantly shorter-lived than a wild-type control 227 population (17%, P < 0.0001, log-rank test), while their F3 genetically wild-type cousins descended from the 228 same ancestral population (labeled WT ("wdr-5")) were still long-lived ( Figure 5E, P < 0.05, log-rank test). 229 Therefore, MET-2 was required for the inheritance of longevity by descendants of wdr-5 mutants. 230 Eliminating MET-2 abrogated the inheritance of wdr-5 mutant longevity, but this suppression could 231 have been caused either by loss of H3K9me2 or by another factor responding to its absence. met-2 mutants 232 had shorter lifespans than wild type ( Figure 4B). If H3K9me2 is the primary factor mediating inheritance of 233 longevity, a normal chromatin state should be established by restoring H3K9me2 through the 234 reintroduction of MET-2. Therefore, the decreased lifespan of met-2 mutants should not be inherited by 235 wild-type progeny. F3 and F5 genetically wild-type descendants of met-2 mutants had a normal lifespan 236 (Figure 5 -figure supplement 1), consistent with H3K9me2 mediating longevity. 237 Similar to wdr-5 mutants, jhdm-1 mutants had longer lifespans and more H3K9me2 (Figure 4). If 238 elevated H3K9me2 can directly mediate the transgenerational inheritance of longevity, we would also 239 expect the longevity of jhdm-1 mutants to be heritable. We examined the lifespan of genetically wild-type 240 animals descended from long-lived jhdm-1. F3 wild-type animals descended from jhdm-1 mutants (labeled 241 WT (jhdm-1)) had a significant increase in lifespan compared to wild-type controls (26%, P < 0.0001, log-242 rank test), though they were not quite as long-lived as F3 jhdm-1 mutants descended from the same jhdm-1 243 mutant population (labeled jhdm-1 (jhdm-1)) (46% compared to wild-type controls, P < 0.0001, log-rank 244 test) ( Figure 5F, following the same genetic scheme as Figure 5A). By generation F5, the lifespan of 245 genetically wild-type descendants from jhdm-1 mutants more closely resembled that of wild-type controls, 246 but still lived slightly longer ( Figure 5G). Both the inheritance of longevity and its reversion by the fifth 247 generation mirrored what was observed in wdr-5 mutants, indicating that the same mechanism may be 248 responsible for both phenomena. 249

DISCUSSION (an extended form can be found in supplementary materials) 251
In this study, we find that lifespan gradually increases with each generation in populations lacking 252 WDR-5 activity for at least eighteen generations. These data provide a rare example of a complex trait 253 acquired transgenerationally across successive generations. We also found that wild-type lifespan 254 decreases in populations recovering from a thaw, indicating that lifespan can be modulated 255 transgenerationally irrespective of any mutations in chromatin modifiers. If a trait as complex as lifespan 256 can change over time without a mutation, it seems likely that other phenotypes may be epigenetically 257 regulated over generational time. In addition, the transgenerational nature of wild type lifespan could 258 account for some discrepancies in the C. elegans aging field. 259 Across generational time, we found that the relative enrichment of H3K9me2 correlates with 260 longevity, raising the possibility that H3K9me2 is protective for lifespan. This correlation is further 261 supported by our findings that the H3K9 methyltransferase MET-2 is necessary both for a normal lifespan 262 and for the lifespan extension of wdr-5 mutants. Furthermore, animals lacking the putative H3K9 263 demethylase JHDM-1 are long-lived, indicating that elevated H3K9me2 directly increases lifespan. Based on 264 these data, we developed the following model. Normally, WDR-5 functions as a component of COMPASS to 265 add H3K4me during transcription. After thawing or starving, the genome has elevated H3K9me2, which is 266 inheritance of longevity, we find that MET-2 is necessary for the inheritance of wdr-5 mutant longevity. In 280 addition, F3 wild-type descendants of long-lived jhdm-1 mutants are long-lived. Based on these findings, we 281 propose that H3K9me2 also allows for the transgenerational inheritance of longevity. 282 Although this work demonstrates that H3K9me2 functions in the establishment and inheritance of 283 extended lifespan, it does not address the mechanism by which H3K9me2 affects longevity. However, the 284 increased lifespan in COMPASS mutants has been linked to the fatty acid desaturation pathway in the 285 intestine (Han et al. 2017). It is not clear how these metabolic changes correlate with the transgenerational 286 acquisition of longevity in COMPASS mutants or its inheritance by wild-type descendants. Nevertheless, the 287 increase in mono-unsaturated fatty acids in COMPASS mutants is at least partially mediated by 288 downregulating germline target genes, including the S6 kinase rsks-1 (Han et al. 2017). When we examined 289 H3K9me2 at this locus over generational time, we found that late-gen wdr-5 mutants had much higher 290 H3K9me2 enrichment than their wild-type counterparts, matching the overall trend we see at all H3K9me2 291 peaks. Thus, it is possible that the accumulation of mono-unsaturated fatty acids in COMPASS mutants may 292 be caused by higher levels of H3K9me2 in long-lived populations. Further work will be necessary to 293 determine whether this is the case. 294 Overall, this work establishes a role for a heterochromatic histone modification in both the 295 establishment and inheritance of a complex trait. This work also relates two seemingly disparate theories of 296 aging that are widely observed among eukaryotes: the disposable soma theory of aging, which is based on 297 observations that reproductive ability often comes at the expense of lifespan (  Hatched L1s were kept without food for six days before being transferred to plates with OP50. The progeny 327 of these starved L1s were used in the lifespan assay. 328 Lifespan assays. Assays were performed at 20°C on NGM agar plates that did not contain 5-fluoro-2'-329 deoxyuridine (FUdR). On Day 1, young adults (on their first day of egg-laying) were allowed to lay for 4-6 330 hours to hatch a synchronized population for the assay. When progeny were L4s or young adults, 90 331 animals per condition were transferred to new plates, with 30 animals per plate. Animals were transferred 332 every day or every other day during their fertile period (usually the first ten days of adulthood). Plates were 333 scored daily and animals marked as dead if they did not move in response to repeated prodding with a 334 platinum pick. Animals were censored from analysis if they died from ruptured vulvas, matricide ("bag of 335 worms" phenotype), or crawling off the agar. Kaplan-Meier survival curves were generated in GraphPad 336 Prism and significance was calculated using a log rank test ( Progeny count assay. Individual hermaphrodites were cloned as L4s and transferred daily until no longer 342 fertile. Progeny were scored as L4s or young adults. Each experiment started with broods of at least five 343 animals, although broods were censored if mothers died before the end of their laying period. Significance 344 was calculated using an unpaired t test. 345 Protein analysis by immunoblot. To generate protein extract, animals were cultured with OP50 on six to 346 twelve 10-cm NGM agar plates. Mixed-stage populations were collected by washing off plates with PBS, 347 pelleted in 500 ul of PBS, and flash frozen. Frozen pellets were thawed, resuspended in NE2 buffer (250 mM 348 sucrose,10 mM HEPES (pH 7.9), 450 mM NaCl, 2 mM MgCl2, 2 mM CaCl2, 0.1% Triton-X100), and flash 349 frozen. Frozen pellets were then disrupted by a 7-ml Type B glass Dounce homogenizer and allowed to lyse 350 on ice for 15 minutes. Pellets were washed two times in cold PBS, then resuspended in 20 mM Tris-HCl (pH 351 7.9). Extracts were resolved with 12% Mini-PROTEAN TGX Stain-Free Protein Gels (BioRad) and transferred 352 to nitrocellulose membranes. Primary antibodies were: 1:500 H3K9me2 antibody (ab1220, Abcam) and 353 1:5000 actin (MAB1501 (Millipore/Upstate)). Primary antibodies were visualized using 1:3000 Rabbit Anti-354 Mouse IgG H&L (HRP) (ab6728, Abcam) and ECL Pus (Amersham Biosciences). Quantification was 355 performed using a ChemiDoc MP and Image Lab software (BioRad). 356 Chromatin immunoprecipitation. To generate chromatin extract, animals were cultured with OP50 on six 357 to twelve 10-cm NGM agar plates. Mixed-stage populations were collected by washing off plates with PBS, 358 pelleted in 500 ul of PBS, and flash frozen. Frozen pellets were disrupted by a 7-ml Type B glass Dounce 359 homogenizer, fixed for ten minutes with 1% formaldehyde (diluted from 37% (w/v)) at 37°C and quenched 360 with 125 mM glycine. ChIP samples were processed with a Chromatin Immunoprecipitation Assay Kit (EMD 361 Millipore) according to manufacturer's instructions. Samples were sonicated using a Diagenode Bioruptor 362 UCD-200 at 4°C for 15 minutes on high, with a cycle of 45s on and 15s off. 1/20 th of sample volume was 363 taken for input controls. For immunoprecipitation, extracts were incubated overnight at 4°C with 10ul of 364 H3K9me2 antibody (ab1220, Abcam). DNA was extracted by phenol-chloroform and ethanol precipitated. Transgenerational epigenetic inheritance of longevity in Caenorhabditis elegans. Nature, 479 (7373) Towbin, B. D., González-Aguilera, C., Sack, R., Gaidatzis, D., Kalck, V., Meister, P., … Gasser, S. M. (2012). 520 Step-wise methylation of histone H3K9 positions heterochromatin at the nuclear periphery. Cell, 521 150 (5) Webster . Longevity takes many generations to manifest in wdr-5 mutants. Analysis of relative lifespan between wild-type (gray) and wdr-5 mutants (purple) across generational time. P0 wild type were descended from animals recovered from a thaw. P0 wdr-5 mutants were the first homozygous mutants after maintenance as heterozygotes for five generations. For each generation, the x-axis is 40 days. * P < 0.05, **P < 0.01, *** P < 0.001, and **** P < 0.0001 compared to wild-type from the same generation with log-rank test. Median lifespan and statistics are presented in supplementary table 1. Additional replicates shown in Figure 1 -figure supplement 1 and supplementary table 3.

Figure 2. Concurrent and opposite changes in lifespan account for the full wdr-5 mutant lifespan
extension. a, Lifespan of early-, mid-, and late-gen wild-type (yellow, tangerine, and burnt orange, respectively) and wdr-5 mutant populations (lavender, purple, and plum, respectively) descended from animals recovered from a thaw. Data are also shown separated into wild-type (a') and wdr-5 mutant populations (a''). Percentage difference in median lifespan between early-and late-gen is indicated above arrow. P < 0.05, **P < 0.01, and **** P < 0.0001 using log-rank test. Median lifespan and statistics are presented in supplementary table 1. Additional replicates are included in supplementary table 3. Figure 3. Long-lived wdr-5 mutants have more H3K9me2 enrichment than wild type. a, Immunoblot comparing H3K9me2 protein levels in late-gen wild type to late-gen wdr-5 mutants and late-gen met-2 mutants (representative of two independent experiments). Actin is used as a loading control. b, d, e, Metaplots of averaged z-score H3K9me2 ChIP-seq signal across H3K9me2 peaks (b), all genes (d), or germline genes (e) in early-, mid-, and late-gen populations of wild type (yellow, orange, and red, respectively) and wdr-5 mutants (lavender, purple, and plum, respectively). Line shows mean ChIP-seq signal. Data in (b) are also shown separated into wild type (b') and wdr-5 mutants (b''). Plots are centered on peak centers (b, b', b'') or pseudoscaled over genes to 1-kb with 500-bp borders on either side, indicated by vertical gray lines (d, e). c, H3K9me2 ChIP-seq ratios of wdr-5 mutant coverage over wild-type coverage at each H3K9me2 peak (N=10,457). Coverage is normalized to RPM. Thick line shows mean and whiskers show standard deviation. Green dots represent peaks that fall beyond y-axis scale (2 peaks in mid-gen and 33 peaks in late-gen). **** P < 0.0001 with t test. . wdr-5 mutant lifespan extension requires H3K9me2. a, Immunoblot comparing H3K9me2 protein levels in mid-gen mixed-stage wild type to mid-gen jhdm-1 and mid-gen met-2 mutants (representative of two independent experiments). Actin is used as a loading control. b, f, Lifespan of early-gen met-2 mutants (green) and wild type (gray) (b) or mid-gen jhdm-1 mutants (blue) compared to early-(yellow) and late-gen (orange) wild type (f). c, d, Generational analysis comparing relative lifespan in wdr-5 met-2 double mutants (pink) and met-2 single mutants (green) (c) or jhdm-1 mutants (blue) (d) to late-gen wild type (gray). The generation below each assay refers only to mutant populations. For each generation, the x-axis is set at is 40 days. e, h, i, Metaplots of averaged z-score H3K9me2 ChIP-seq signal across H3K9me2 peaks (e), all genes (h), or germline genes (i) in early-and mid-gen populations of wild type (yellow and orange, respectively) and jhdm-1 mutants (blue and navy, respectively). Line shows mean ChIP-seq signal. Plots are either centered on peak centers (e) or pseudoscaled over genes to 1-kb with 500-bp borders on either side, indicated by vertical gray lines (h, i). g, H3K9me2 ChIP-seq ratios of jhdm-1 mutant coverage over wild-type coverage at each H3K9me2 peak. Coverage for each sample is normalized to RPM. Thick line shows mean and whiskers show standard deviation. Green dot represents 65 peaks that lie beyond y-axis scale. *P < 0.05, **P < 0.01, *** P < 0.001, and **** P < 0.0001 compared to wild-type with log-rank test for lifespan assays or with t test for coverage ratios. Median lifespan and statistics are presented in supplementary table 1, with additional replicates included in supplementary table 3.

Extended discussion 1
Lifespan is a complex trait determined by many factors, including overall chromatin state. In this 2 study, we find that lifespan gradually increases with each generation in populations lacking WDR-5 3 activity for at least eighteen generations. These data provide a rare example of a complex trait acquired 4 transgenerationally across successive generations. 5 We were surprised to find that wild-type lifespan decreases over generational time in 6 populations recovering from a thaw. The changes in wild-type lifespan indicate that lifespan can be 7 modulated transgenerationally, irrespective of any mutations in chromatin modifiers. If a trait as 8 complex as lifespan can change over time without a mutation, it seems likely that other phenotypes may 9 be epigenetically regulated over generational time. In addition, the transgenerational nature of wild 10 type lifespan could account for some discrepancies in the C. elegans aging field. We find that after 11 starving or recovering from a thaw, wild-type lifespan decreases for about twelve generations before 12 reaching a steady state. These data, along with other studies, suggest that starving increases lifespan by 13 The gradual appearance of longevity in wdr-5 mutant populations suggests that this factor 30 accumulates over many generations, while two other findings indicate that it resides in the germline. 31 First, heritable factors must be passed through germline tissue, whether they are genetically or 32 epigenetically inherited. Second, a proliferating germline is required for longevity in wdr-5 mutants 33 (Greer et al. 2010). Therefore, in this study we examined whether H3K9me2 might be the heritable 34 factor that confers longevity. H3K9me2 levels decrease as lifespan decreases in wild type, while they 35 remain high in long-lived late-gen wdr-5 mutants compared to wild type. The relative enrichment of 36 H3K9me2 is particularly pronounced at germline-expressed genes, which is counter to the slight overall 37 decrease found in the rest of the genome. Thus, across generational time, the relative enrichment of 38 H3K9me2 correlates with lifespan in both wild-type and wdr-5 mutant populations. 39 The correlation between H3K9me2 and longevity raises the possibility that H3K9me2 is 40 protective for lifespan. This correlation is further supported by our finding that met-2 mutants are short-41 lived. Additionally, we found that MET-2 is necessary for the increased lifespan in wdr-5 mutants, 42 suggesting that H3K9me2 is required for the phenotype. However, multiple pathways are able to 43 shorten lifespan (Kenyon 2010). To test whether higher levels of H3K9me2 can independently extend 44 lifespan, we examined animals lacking JHDM-1, a putative H3K9 demethylase. The absence of JHDM-1 is 45 sufficient to increase lifespan, indicating that elevated H3K9me2 can directly confer longevity. In 46 addition, most H3K9me2 peaks in long-lived jhdm-1 mutants are shared with long-lived wdr-5 mutants, 47 which is consistent with the possibility that lifespan extension is caused by similar means in both 48 mutants. However, longevity appears more quickly in jhdm-1 mutants than in wdr-5 mutants, perhaps 49 because JHDM-1 has a more direct effect on H3K9me2 than COMPASS. Additionally, we noticed that 50 mid-or late-gen jhdm-1 mutants appear healthier than their wdr-5 mutant counterparts, indicating that 51 H3K9me2 may affect health-span in addition to lifespan. 52 The increased lifespan of COMPASS mutants has been linked to changes in fatty acid metabolism 53 and an increase in mono-unsaturated fatty acids (Han et al. 2017). However, it is not clear how these 54 metabolic changes correlate with the transgenerational acquisition of longevity in COMPASS mutants or 55 its inheritance by wild-type descendants. The increase in mono-unsaturated fatty acids in COMPASS 56 mutants is at least partially mediated by downregulating germline target genes, including the S6 kinase 57 rsks-1 (Han et al. 2017). When we examined H3K9me2 at this locus over generational time, we found 58 that late-gen wdr-5 mutants had much higher H3K9me2 enrichment than their wild-type counterparts, 59 matching the overall trend we see at all H3K9me2 peaks. Thus, it is possible that the accumulation of 60 mono-unsaturated fatty acids in COMPASS mutants may be caused by higher levels of H3K9me2 in long-61 lived populations. Further work will be necessary to identify the precise means by which H3K9me2 62 enrichment affects lifespan. 63 Based on our data, we developed the following model. Normally, WDR-5 functions as a 64 component of COMPASS to add H3K4me during transcription. After thawing or starving, the genome has 65 elevated H3K9me2, which is eroded in subsequent generations by the deposition of transcription-66 coupled H3K4me. This H3K9me2 reduction gradually reduces lifespan. In wild type, the continued 67 presence of COMPASS activity in the germline keeps H3K9me2 levels low over genes by maintaining high 68 H3K4me levels. COMPASS's maintenance of H3K4me over expressed genes occurs even when H3K9me2 69 levels are elevated overall in jhdm-1 mutants. In the absence of COMPASS and transcription-coupled 70 H3K4me, H3K9me2 is protected in the genome and can even accumulate at germline genes over 71 generational time. These relatively high levels of H3K9me2 cause a longer lifespan in COMPASS mutants. 72 Thus, we propose that the transgenerational inheritance of H3K9me2 drives the lifespan increase in 73 wdr-5 mutants. 74 Our model may also explain how longevity is transgenerationally inherited in genetically wild-75 type descendants of COMPASS mutants (Greer et al. 2011). Inappropriately high levels of H3K9me2 in 76 long-lived COMPASS mutants could be inherited by their descendants, initially preventing COMPASS 77 from restoring a steady-state chromatin environment. Elevated levels of inherited H3K9me2 could then 78 confer longevity in descendants of COMPASS mutants. After several generations of normal COMPASS 79 activity, H3K4me would then reestablish wild-type levels of H3K9me2, resulting in the reversion to 80 normal lifespan in the fifth generation of descendants. Consistent with our model, we find that MET-2 is 81 necessary for the inheritance of wdr-5 mutant longevity. However, this genetic interaction does not 82 distinguish whether MET-2's effect on inheritance is directly through H3K9me2 or through another 83 factor responding to the absence of H3K9me2. The lifespan extension of jhdm-1 mutants allows for an 84 independent test of whether elevated H3K9me2 mediates inheritance. If H3K9me2 is the mechanism by 85 which descendants of COMPASS inherit their longevity, it should be heritable and confer longevity no 86 matter the genetic background. Similar to what was reported for wdr-5 mutants (Greer et al. 2011), we 87 find that F3 wild-type descendants of long-lived jhdm-1 mutants are long-lived, and longevity reverts in 88 F5 wild-type descendants. Furthermore, if H3K9me2 mediates inheritance directly, we would not expect 89 the short lifespan of met-2 mutants to be inherited when H3K9me2 is restored by MET-2. We show that 90 F3 genetically wild-type descendants of met-2 mutants have normal lifespans, suggesting that H3K9me2 91 is directly involved in inheritance of longevity. Based on these findings, we propose that H3K9me2 also 92 acts as the mechanism for the transgenerational inheritance of longevity. 93 Previously, mutants lacking the H3K4 demethylase SPR-5 were found to acquire increased 94 longevity after six to ten generations (Greer et al. 2015). Thus, we considered the possibility that spr-5 95 and wdr-5 mutants share a common mechanism. However, spr-5 mutants accumulate the active 96 modification H3K4me2 (Katz et al. 2009), which differs from the H3K9me2 accumulation observed in 97 wdr-5 mutants. Additionally, spr-5 mutant longevity is mediated by the known DAF-36/DAF-12 lifespan 98 signaling pathway (Greer et al. 2015), whereas the lifespan extension of COMPASS mutants occurs 99 independent of this pathway. Therefore, it is likely that different mechanisms underlie the 100 transgenerational acquisition of longevity in these mutants. 101 Overall, this work establishes a role for a heterochromatic histone modification in both the 102 establishment and inheritance of a complex trait. This work also relates two seemingly disparate 103 theories of aging that are widely observed among eukaryotes: the disposable soma theory of aging, 104 which is based on observations that reproductive ability often comes at the expense of 105 lifespan (Kirkwood & Holliday 1979), and the heterochromatin loss model of aging, which is based on the 106 observation that heterochromatin declines with age (Tsurumi & Li 2012;Villeponteau 1997). Our model 107 proposes a mechanism that could connect both theories: if reduced heterochromatin is a burden on 108 lifespan, then limiting H3K9me2 through transcription-coupled H3K4me deposition in the germline may 109 represent one cost of maintaining a germline. 110 Figure 1 -figure supplement 1. Longevity in wdr-5 mutants reproducibly takes many generations to manifest. (A)-(G) Analysis of relative lifespan between wild-type (gray) and wdr-5 mutants (purple) in seven independent transgenerational experiments. wdr-5 mutant populations were either reset by being maintained as heterozygotes for five generations following an outcross or by starving, as indicated in graph legend. Replicates (A), (E), and (F) were used for ChIP-seq analysis. For each generation, the xaxis is 40 days. * P < 0.05, **P < 0.01, *** P < 0.001, and **** P < 0.0001 compared to wild-type from the same generation with log-rank test. Mean lifespan and statistics are presented in supplementary table 3.

Figure 1 -figure supplement 2. Longevity is not caused by decreased fecundity or by background mutations. (A)
Comparison of relative lifespan between wild-type (gray) and wdr-5 mutant (purple) populations over generational time. For each generation, the x-axis is set at is 40 days. (B) Difference in progeny number between wild type and wdr-5 mutants for populations shown in (A). Each generation in (B) is shown directly under the corresponding generation in (A). (C) Comparison of lifespan in fed descendants from starved or fed late-gen populations of either wdr-5 mutants (purple and cyan, respectively) or wild type (orange and pink, respectively). (D) Lifespan of early-gen wdr-5 mutants (purple) derived from outcrossing late-gen wdr-5 mutants compared to the corresponding unoutcrossed late-gen wdr-5 mutant population (plum) and mid-gen wild-type animals (gray). * P < 0.06, **P < 0.01 and *** P < 0.001 compared to the wild-type population in the same assay with log-rank test. Progeny data and an additional replicate are presented in supplementary table 2. Median lifespan, statistics, and additional replicates are presented in supplementary table 3. Representative H3K9me2 enrichment profiles in an IGV browser screenshot over a 26-kb region of Chr. III (A) and the rsks-1 locus (B). Average z-scored ChIP-seq tracks for H3K9me2 signal in early-gen wild type (yellow), early-gen wdr-5 mutants (lavender), mid-gen wild type (orange), mid-gen wdr-5 mutants (purple), late-gen wild type (red), and late-gen wdr-5 mutants (plum). All coverage tracks are shown on the same scale, -1.1 to 6.2 (A) or -1.5 to 3.5 (B). Below the tracks, H3K9me2 ChIP-seq peaks are shown in green, and genes are shown in blue. Highlighted region outlined in green shows a peak where mutant / wild type coverage ratios do not change over generational time. Highlighted regions outlined in pink show peaks where the ratio increases in late-gen populations. (C) Overlap between peaks called in transgenerational ChIP-seq experiments of wdr-5 mutants (Figure 3) or jhdm-1 mutants (Figure 4).  -2), gold) compared to met-2 mutants descended from the same parental population (met-2 (met-2), teal) and descendants of wild-type animals (WT (WT), gray). Genetic scheme follows that used in Figure 5A. *P < 0.05 and **** P < 0.0001 with log-rank test. Median lifespan, statistics, and an additional replicate are presented in supplementary table 3. (green) and H3K9me2 (purple) through generational time as lifespan changes in wild type (A) or wdr-5 mutants (B). Color of animal represents lifespan: white indicates short-lived, gray indicates normal, and purple indicates long-lived. In (A), lifespan is compared to P0 wild type. In (B), lifespan is compared to wild type of the same generation. After wdr-5 mutants become long-lived in late-gen populations (B), their genetically wild-type descendants inherit both elevated H3K9me2 and longevity. By the fifth generation, genomic H3K4me and H3K9me2 levels are rebalanced and lifespan reverts to normal. Figures 1-5. Shading indicates groupings of populations assessed in a single lifespan assay. Median lifespan was calculated from Kaplan-Meier survival curves and P values were calculated using a log-rank test * P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. N indicates the number of observed dead animals at the end of the experiment, with the initial number of living animals indicated in parentheses. The difference corresponds to the number of individuals censored for deaths via matricide, vulval rupture, or desiccation from crawling off the plate. Figure panels for specific experiments are indicated in column 8. Replicate number is indicated in column 9.

Supplementary table 1. Summary statistics for lifespan data in
Supplementary table 2. Summary statistics for progeny number and embryonic lethality used in supplementary figure 2. Percent survival was calculated from counting the number of embryos laid and the number of surviving adults. Each experiment started with at least five broods; broods were censored if mother died by matricide or vulval rupture.

Supplementary table 3. Summary statistics for lifespan data in figure supplements and additional
replicates of lifespan assays. Shading indicates groupings of populations assessed in a single lifespan assay. Median lifespan was calculated from Kaplan-Meier survival curves and P values were calculated using a log-rank test * P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. N indicates the number of observed dead animals at the end of the experiment, with the initial number of living animals indicated in parentheses. The difference corresponds to the number of individuals censored for deaths via matricide, vulval rupture, or desiccation from crawling off the plate. Figure panels for specific experiments are indicated in column 8. If data are not represented in a figure, the figure that shows its replicate is indicated. Replicate number is indicated in column 9.    1 -Fig Supp 1c, Fig Supp 2a 2/8 F20 WT 17 75/90 Fig 1 -Fig Supp 1c, Fig Supp 2a 2/8 F20 wdr-5 18 76/90 F20 WT 0.0653 Fig 1 -Fig Supp 1c, Fig Supp 2a 2/8 F24 WT 17 84/90 Fig 1 -Fig Supp 1c, Fig Supp 2a 2/8 F24 wdr-5 20 69/90 F24 WT 0.0037 ** Fig 1 -Fig Supp 1c