Healthy aging interventions reduce non-coding repetitive element transcripts

Transcripts from non-coding repetitive elements (RE) in the genome may be involved in aging. However, they are often ignored in transcriptome studies on healthspan and lifespan, and their role in healthy aging interventions has not been characterized. Here, we analyze RE in RNA-seq datasets from mice subjected to robust healthspan- and lifespan-increasing interventions including calorie restriction, rapamycin, acarbose, 17-α-estradiol, and Protandim. We also examine RE transcripts in long-lived transgenic mice, and in mice subjected to high-fat diet, and we use RNA-seq to investigate the influence of aerobic exercise on RE transcripts with aging in humans. We find that: 1) healthy aging interventions/behaviors globally reduce RE transcripts, whereas aging and age-accelerating treatments increase RE expression; and 2) reduced RE expression with healthy aging interventions is associated with biological/physiological processes mechanistically linked with aging. Thus, RE transcript dysregulation and suppression are likely novel mechanisms underlying aging and healthy aging interventions, respectively.


Healthy aging interventions reduce non-coding repetitive element transcripts
Older age is the greatest risk factor for the development of most chronic diseases (1). Accordingly, 98 recent large-scale 'omics' studies have aimed to characterize novel genes and biological pathways 99 that influence aging, and to identify related interventions (e.g., pharmaceutical compounds, exercise, 100 nutrition) that increase longevity and healthspan (2, 3). Indeed, advances in transcriptomics (e.g., 101 RNA-seq) have led to important insight on many genes and pathways linked with 'the hallmarks of 102 aging' and broader health outcomes (4). However, most of these studies have focused on coding 103 sequences-a small fraction of the genome. Non-coding, repetitive elements (RE, >60% of the 104 genome) have been particularly neglected as 'junk DNA' (5), despite growing evidence that they have 105 many important biological functions (6 To determine if global RE transcript suppression might be a mechanism underlying healthy aging 132 interventions, we first analyzed RE in an RNA-seq dataset on livers from young and old mice and old 133 mice subjected to life-long (24 months) CR (14). We found a small, but significant age-related 134 increase in most major RE transcript types in this dataset, consistent with our previous work (15) and 135 others' (16). However, this effect was significantly attenuated with CR (Fig. 1A). Based on this novel 136 evidence of RE suppression by CR (arguably the strongest health/lifespan-enhancing intervention), 137 we looked to confirm our results in an additional, large dataset including RNA-seq on livers from mice 138 subjected to different durations of CR and pharmacological interventions known to increase 139 health/lifespan (rapamycin, acarbose, 17-⍺-estradiol [17aE], and Protandim), as well as data on 140 transgenic, long-lived mice (2) ( Fig. 1 and Supplementary Data). We found that long-term (8 141 months) CR caused a significant, global reduction in RE transcripts (Fig. 1B). Furthermore, we found 142 that both long-term (8 months) rapamycin and acarbose treatments were associated with a 143 comparable, broad reduction in RE transcripts (Fig. 1B), consistent with the notion that these 144 compounds are 'calorie restriction mimetics' and may act via similar pathways (17). This effect was 145 particularly clear when we examined RE/TE reductions by major sub-type (Fig. 1C). Short-term (2 146 months) interventions with other healthy aging compounds influenced RE transcript levels to various 147 degrees, although reductions were more pronounced with CR and Protandim, which is thought to 148 activate endogenous antioxidant defenses (Fig. 1D). Interestingly, the authors of the original study 149 (2) observed similar variability in gene expression patterns, suggesting time/treatment-specific 150 transcriptome effects. We also found a significant influence of growth hormone receptor knockout 151 (GHRKO, a transgenic longevity model) on the main RE transcript types (Fig. 1E). Moreover, in a 152 separate dataset (18), we found that high-fat diet (HFD, a common 'pro-aging' intervention) 153 significantly increased all major RE/TE (Fig. 1F). Collectively, these results support the idea that 154 global RE transcript levels are linked with healthspan/lifespan, as they are reduced by most "gold 155 standard" anti-aging interventions and increased by adverse, pro-aging treatments. 156 157 Next, we examined similarities in the effects of healthy aging interventions on RE by sub-type/family 158 ( Fig. 2 and Supplementary Data). Again, we observed variable patterns of RE family expression 159 with short-term treatments ( Fig. 2A). With long-term treatment, CR and rapamycin influenced RE/TE 160 families most similarly, and most transcripts were decreased with all treatments (Fig. 2B), and in 161 GHRKO mice (Fig. 2C). We next determined which specific RE transcripts were commonly 162 decreased/increased among all treatments. Despite the variable RE/family expression patterns noted 163 above, short-term treatments modulated many of the same transcripts ( Fig. 2D and Supplementary 164 Data). Long-term treatments also decreased/increased a large number of common transcripts (518 165 and 92, respectively) ( Fig. 2E and Supplementary Data). Consistent with the idea that global RE 166 modulation is linked with healthy aging interventions, we did not notice any particular enrichment for 167 specific RE/TE types in these common transcripts. However, we did note that endogenous retrovirus 168 (ERV) RE transcripts were the most decreased with all long-term treatments and in GHRKO mice. 169 Interestingly, ERVs have been implicated in aging and several diseases of aging including 170 neurodegenerative disorders (19, 20), suggesting that these RE could be an important therapeutic 171 target.

173
Identifying potentially targetable biological mechanisms linking reduced RE expression with healthy 174 aging interventions will require future experiments. However, to provide initial insight, we examined 175 correlations among gene and RE expression patterns in mice subjected to long-term CR, rapamycin 176 and acarbose (treatments that reduced RE transcripts the most). To do this, we conducted a 177 weighted gene correlation network analysis (WGCNA) on both gene and RE transcript counts ( Fig. 3  178 and Supplementary Data). Although gene/RE signatures across interventions were not strikingly 179 similar, we identified one WGCNA module (green) that decreased significantly with all interventions 180 (Fig. 3A). This module contained numerous DNA transposons and several LINE, ERV and LTR 181 transcripts. A gene ontology (GO) analysis of the module also showed significant enrichment for 182 many biological processes, including several linked with aging and disease (Fig. 3B). In fact, the 183 most specific GO terms included protein deacetylation, DNA repair and immune activation/response 184 pathways. The other gene/RE modules that decreased with CR and/or rapamycin were also enriched 185 for specific GO terms including DNA repair, DNA/RNA processing, histone modifications and stress 186 responses (Fig. 3B). These exploratory analyses do not definitively link reduced RE expression with 187 such processes, but they are consistent with current thinking that age-related RE transcript 188 accumulation could cause DNA damage (21) and immune activation/inflammation (22), and that RE 189 dysregulation may be due to age-associated changes in chromatin/histones (23).

191
There is little or no RNA-seq data on true long-term CR or healthy aging compounds in older humans, intervention (~25% of the animal's life). Therefore, as initial proof of concept, we conducted a cross-197 sectional study to determine if long-term exposure to this healthy aging behavior has the potential to 198 reduce RE transcripts (Fig. 4 and Supplementary Data). We performed RNA-seq and gene/RE 199 expression analyses on peripheral blood mononuclear cells (PBMC) from: 1) young and older 200 sedentary adults; and 2) older habitually (≥5 years) exercising adults (Supplementary Table).

201
Consistent with other reports (28), we found that older age was associated with altered PBMC gene 202 expression, but these changes were largely attenuated in exercising older adults (Fig. 4A). 203 Moreover, in support of the idea that healthy aging interventions/behaviors may reduce RE 204 expression in humans, we observed a clear increase in global RE transcript levels in older sedentary 205 adult PBMC, but this effect was strongly attenuated with exercise (Fig. 4B). We also found that 206 maximal aerobic exercise capacity (VO 2 max) was inversely related to a composite count of RE that 207 are significantly increased with aging (Fig. 4C), suggesting that greater aerobic fitness (and perhaps 208 exposure to aerobic exercise) is directly linked with reduced RE expression. Interestingly, VO 2 max is 209 considered a key physiological predictor of longevity in humans (29) to quantify RE transcripts by individual total counts, class, and family. Gene expression counts were 247 extracted from bam alignment files produced during TEtranscripts analyses, and differential 248 expression analyses of both RE and genes were performed using Deseq2 software (36). WGCNA 249 was performed according to standard procedures outlined by the analysis pipeline's authors (37) 250 using normalized gene and RE counts for all samples, and a minimum module size of 300 to capture 251 broader groups of RE that correlated with sample traits (specific interventions). GO analyses of 252 genes in the WGCNA modules were performed using the GOrilla algorithm (38), and specific GO 253 modules were identified as terminal nodes in the directed acyclic graph produced by this program.

255
Human subjects and RNA-seq samples 256 RNA-seq was performed on PBMC from twelve healthy young (18-22 years) and older (62-74 years) 257 adults. Subjects were non-obese, non-smokers and healthy as assessed by medical history, physical 258 examination, blood chemistries and exercise ECG, and small groups were selected to match 259 characteristics as closely as possible. Young (n=5, 2 male) and older (n=5, 2 m) sedentary subjects 260 performed no regular exercise (< 2 days/week, < 30 min/day), whereas older exercising subjects 261 (n=4, 1 m) performed regular vigorous aerobic exercise (≥ 5 days/week, > 45 min/day) for the 262 previous ≥ 5 years. The study conformed to the Declaration of Helsinki; all procedures were 263 approved by the Institutional Review Board of the University of Colorado Boulder, and written 264 informed consent was obtained from all subjects. Maximal oxygen consumption (VO 2 max) was 265 assessed during treadmill exercise as previously described (39) and basic clinical measurements 266 (e.g., blood pressure) were performed using standard techniques. PBMC were isolated from whole 267 blood by traditional Ficoll gradient centrifugation, and RNA-seq and gene expression analyses were 268 performed using standard methods as previously described (15,40