Deep phenotyping and lifetime trajectories reveal limited effects of longevity regulators on the aging process in C57BL/6J mice

Abstract

Current concepts regarding the biology of aging are primarily based on studies aimed at identifying factors regulating lifespan. However, lifespan as a sole proxy measure for aging can be of limited value because it may be restricted by specific pathologies. Here, we employ large-scale phenotyping to analyze hundreds of markers in aging male C57BL/6J mice. For each phenotype, we establish lifetime profiles to determine when age-dependent change is first detectable relative to the young adult baseline. We examine key lifespan regulators (putative anti-aging interventions; PAAIs) for a possible countering of aging. Importantly, unlike most previous studies, we include in our study design young treated groups of animals, subjected to PAAIs prior to the onset of detectable age-dependent phenotypic change. Many PAAI effects influence phenotypes long before the onset of detectable age-dependent change, but, importantly, do not alter the rate of phenotypic change. Hence, these PAAIs have limited effects on aging.

Fig. 1. To what extent can aging, measured as a multidimensional representation of age-dependent phenotypic change, be slowed in mice?

a To estimate aging trajectories for a wide range of age-sensitive phenotypes (ASPs), we examined phenotypes across the lifespan of C57BL/6J mice, including hundreds of phenotypes derived from multi-dimensional deep phenotyping, a range of molecular markers as well as transcriptomic profiles. b We assessed three important pro-longevity interventions for their effects on aging (putative anti-aging interventions; PAAIs). For each PAAI, we generated a young as well as an old cohort of experimental animals and controls, all of which were analyzed concurrently in one single study. For each phenotype in each of these studies, we determined age effects, intervention effects and intervention × age interactions based on the data derived from young and old control as well as experimental animals. These analyses revealed that some ASPs were influenced (countered or accentuated) by the PAAIs, others not. For ASPs countered by PAAIs, we considered the following scenarios: PAAIs could influence ASPs in a way consistent with slowing the rate of age-dependent change in ASPs (rate effect), via age-independent effects on ASPs (baseline effect) or via a combination of rate and baseline effects. To address what the age at first detectable change is for each ASP influenced by an intervention, we intersected data on ASPs from these intervention studies (see panel b) with data from our baseline study (see panel a). We compared effect sizes to examine for each ASP individually whether PAAI effects differed measurably between young and old mice. In addition, we used dimensionality reduction approaches as well as intraclass correlation analyses of intervention effect sizes in young and old animals to determine whether PAAIs overall act on ASPs primarily in a way consistent with slowing their rate of age-dependent change (left panels), via age-independent effects (right panels) or via a combination of rate and baseline effects (middle panels). For further details on our analytical approach, see Supplementary Fig. 1. Created with BioRender.com.

Fig. 2. Multidimensional analyses of age-dependent phenotypic change in C57BL/6J mice.

a–f Deep phenotyping results in wildtype C57BL/6J mice. a Schematic illustration of deep phenotyping study design. Mouse symbols adapted from ref. . b Relative proportion of ASPs among all phenotypes examined. Age at first detectable change (c) and age at full manifestation (d) of ASPs shown as proportion of all ASPs. e Representative examples of ASPs with various ages at first detectable phenotypic change. Data was analyzed using one-way ANOVA with the between-subjects factor age followed by Fisher’s LSD posthoc analyses, where appropriate. f Principal component analysis of deep phenotyping data. g–j Summary of RNA-seq data. g Schematic illustration of RNA-seq study design. Venn diagram shows, for brain (h) and spleen (i), the number of differentially expressed genes (FDR < 0.05) relative to the 3-month old reference group together with the intersection of the corresponding gene sets. j Ingenuity Pathway Analysis shows top canonical pathways, diseases and biological functions as well as predicted upstream regulators of genes differentially expressed in spleen relative to the 3-month old group. Positive z-scores (orange) indicate activating effects, negative z-scores (blue) indicate inhibitory effects on corresponding processes. Pathway analyses of brain data are shown in Supplementary Fig. 3. k–n Summary of molecular analyses designed to study putative driver mechanisms of aging in spleen, lung and brain. k Schematic illustration of study design (for sample size, see Supplementary Data 6). l Proportion of age-sensitive molecular parameters obtained. m Proportion of the different age-at-first-detectable-change categories among all age-sensitive molecular markers in individual tissues. n Representative examples of molecular markers covering the different hallmarks of aging. Line plots (e, n) show means +/− S.D. (individual data points are superimposed; we did not use jittering to separate data points with identical values). Data was analyzed using one-way ANOVA with the between-subjects factor age followed by Fisher’s LSD posthoc analyses, where appropriate. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 relative to 3-month young adult reference group. Sample sizes in each group and experiment are provided within the figure. Source data are provided in Source Data file. Created with BioRender.com.

Fig. 3. ‘Anti-aging’ effects induced by the Ghrhr^lit/lit mutation manifested mostly in young mice (prior to detectable age-dependent phenotypic changes).

a–i Deep phenotyping results in Ghrhr^lit/lit mice. a Schematic illustration of study design. Mouse symbols adapted from ref. . b Principal component analysis of deep phenotyping data. c Venn diagram shows the number of ASPs, genotype-sensitive phenotypes, phenotypes with a genotype × age interaction and their intersection. Sunburst chart shows the number of ASPs either unaltered, counteracted or accentuated by genotype. For ASPs counteracted by genotype, the inner ring shows the proportion of phenotypes with either a genotype main effect, a genotype × age interaction or both. The outer ring shows when changes in the corresponding ASPs were first detected in our baseline study. Line charts show representative examples of phenotypes influenced by age and/or intervention in the different possible ways. Data are presented as z-scores (normalized to the young WT group) and are plotted as mean +/− S.D. (individual data points are superimposed). Two-way ANOVAs with the between-subjects factors age and genotype followed by Fisher’s LSD posthoc analyses, where appropriate, were used for data analysis. *p < 0.05, **p < 0.01, ***p < 0.001 relative to age-matched wildtype littermate controls. Life-time trajectories of the corresponding phenotypes, representing measurements obtained in our baseline study, are shown by the gray-shaded area in the background (bounded by mean +/− S.D.). Effect size of genotype in old (d) or young (e) mice plotted vs. the effect size of age for all ASPs and those intersecting with genotype (via a main effect and/or interaction), respectively. f–i Effect size of genotype in young mice plotted vs. the effect size in old mice for different sets of phenotypes: g ASPs ameliorated by genotype via a main effect and/or an interaction. h ASPs accentuated by genotype. i Phenotypes featuring a genotype main effect and/or a genotype × age interaction but no effect of age. f all phenotypes shown in g–i. ICC = intraclass correlation. Statistical effect size comparisons were performed via two-sided z-tests. Our analyses are based on unadjusted p-values. Source data are provided as a Source Data file.

Fig. 4. PAAIs – standardized coefficients with 95% confidence intervals for age, intervention and intervention × age interaction terms.

Phenotypes analyzed within the Ghrhr^lit/lit, mTOR^KI/KI and IF cohorts are denoted in blue, green and red, respectively. The forest plots show, for all three PAAIs examined in the present paper, standardized coefficients, with 95% confidence intervals, corresponding to the age effects (a, d, g), intervention effects (b, e, h) and intervention × age interactions (c, f, i) derived from linear models applied to all continuously distributed phenotypes included in the present study. Note that we have not included phenotype labels on the y-axes to allow for a compact data presentation. Phenotypes were sorted based on the coefficient of the age effect. For a given intervention, this order is preserved across all panels from left to right (showing age effects, intervention effects and intervention × age interactions) to permit comparison of coefficients across any given phenotype. Sample sizes (number of mice per group) underlying these analyses are detailed, for each phenotype and all PAAIs, in Supplementary Data 8, 10 and 13.

Fig. 5. GHRHR eQTL effects on aging-associated phenotypes in humans.

a–d Phenotyping results in a large deep-phenotyped human cohort that included participants of both genders with an age range from 30–95 years. a Change in phenotype (in standard deviations (SD) from the mean) associated with GHRHR eQTL dosage with the horizontal whiskers indicating the 95% confidence intervals of the mean effect estimate; * denotes p < 0.05 for the linear association between GHRHR eQTL dosage and phenotype The association between eQTL dosage and each phenotypic measure was assessed using multiple linear regression models adjusted for age, sex and population stratification using the first ten genetic principal components. Our analyses are based on unadjusted p-values. Change of platelet count (b), total cholesterol (c) and LDL-cholesterol (d) associated with GHRHR eQTL dosage in 30–49 years old (red line), 50–69 years old (green line) and 70–95 years old humans (blue line); the lines represent the best-fit least squares regression lines with surrounding 95% confidence intervals of the mean indicated in gray. The eQTL dosage was coded as GG = 0, AG = 1, and AA = 2 (GG is associated with lowest expression levels, AA with highest; see Supplementary Fig. 8a). Sample size was 994 human participants per age group. Created with BioRender.com.

Fig. 6. A hypomorphic mTOR mutant allele attenuated ASPs via a mixture of age-independent effects and effects more pronounced in old mice.

a–i Deep phenotyping results in mTOR^KI/KI mice. a Schematic illustration of study design. Mouse symbols adapted from ref. . b Principal component analysis of deep phenotyping data. c Venn diagram shows the number of ASPs, genotype-sensitive phenotypes, phenotypes with a genotype × age interaction and their intersection. Sunburst chart shows the number of ASPs either unaltered, counteracted or accentuated by genotype. For ASPs counteracted by genotype, the inner ring shows the proportion of phenotypes with either a genotype main effect, a genotype × age interaction or both. The outer ring shows when changes in the corresponding ASPs were first detected in our baseline study. Line charts show representative examples of phenotypes influenced by age and/or intervention in the different possible ways. Data are presented as z-scores (normalized to the young WT group) and are plotted as mean +/− S.D. (individual data points are superimposed). Two-way ANOVAs with the between-subjects factors age and genotype followed by Fisher’s LSD posthoc analyses, where appropriate, were used for data analysis. *p < 0.05, **p < 0.01, ***p < 0.001 relative to age-matched wildtype littermate controls. Life-time trajectories of the corresponding phenotypes, representing measurements obtained in our baseline study, are shown by the gray-shaded area in the background (bounded by mean +/− S.D.). Effect size of genotype in old (d) or young (e) mice plotted vs. the effect size of age for all ASPs and those intersecting with genotype (via a main effect and/or interaction), respectively. f–i Effect size of genotype in young mice plotted vs. the effect size in old mice for different sets of phenotypes: g ASPs ameliorated by genotype via a main effect and/or an interaction. h ASPs accentuated by genotype. i Phenotypes featuring a genotype main effect and/or a genotype × age interaction but no effect of age. f All phenotypes shown in g–i. ICC = intraclass correlation. Statistical effect size comparisons were performed via two-sided z-tests. Our analyses are based on unadjusted p-values. Source data are provided as a Source Data file.

Fig. 7. MTOR eQTL effects on aging-associated phenotypes in humans.

a–f Phenotyping results in a human cohort that included participants of both genders with an age range from 30–95 years. a Change in phenotype associated with MTOR eQTL dosage in a large deep-phenotyped human cohort, with the horizontal whiskers indicating the 95% confidence intervals of the mean effect estimate; * denotes p < 0.05 for the linear association between MTOR eQTL dosage and phenotype The association between eQTL dosage and each phenotypic measure was assessed using multiple linear regression models adjusted for age, sex and population stratification using the first ten genetic principal components. Our analyses are based on unadjusted p-values. Change of body fat (b), percentage of body fat (c), body weight (d), plasma creatine concentration (e) and metabolic equivalent hours (f) associated with MTOR eQTL dosage in 30–49 years old (red line), 50–69 years old (green line) and 70–95 years old humans (blue line); the lines represent the best-fit least squares regression lines with surrounding 95% confidence intervals of the mean indicated in gray. The eQTL dosage was coded as GG = 0, CG = 1, and CC = 2 (GG is associated with lowest expression levels, CC with highest; see Supplementary Fig. 8b). Sample size was 997 human participants per age group. Created with BioRender.com.

Fig. 8. ‘Anti-aging’ effects induced by intermittent fasting (IF) often manifest in young mice (prior to detectable age-dependent phenotypic changes).

a–i Deep phenotyping results in IF mice. a Schematic illustration of study design. Mouse symbols adapted from ref. . b Principal component analysis of deep phenotyping data. c Venn diagram shows the number of ASPs, diet-sensitive phenotypes, phenotypes with a diet × age interaction and their intersection. Sunburst chart shows the number of ASPs either unaltered, counteracted or accentuated by diet. For ASPs counteracted by diet, the inner ring shows the proportion of phenotypes with either a diet main effect, a diet × age interaction or both. The outer ring shows when changes in the corresponding ASPs were first detected in our baseline study. Line charts show representative examples of phenotypes influenced by age and/or intervention in the different possible ways. Data are presented as z-scores (normalized to the young WT group) and are plotted as mean +/− S.D. (individual data points are superimposed). Two-way ANOVAs with the between-subjects factors age and diet followed by Fisher’s LSD posthoc analyses, where appropriate, were used for data analysis. *p < 0.05, **p < 0.01, ***p < 0.001 relative to age-matched wildtype littermate controls. Life-time trajectories of the corresponding phenotypes, representing measurements obtained in our baseline study, are shown by the gray-shaded area in the background (bounded by mean +/− S.D.). Effect size of diet in old (d) or young (e) mice plotted vs. the effect size of age for all ASPs and those intersecting with diet (via a main effect and/or interaction), respectively. f–i Effect size of diet in young mice plotted vs. the effect size in old mice for different sets of phenotypes: g ASPs ameliorated by diet via a main effect and/or an interaction. h ASPs accentuated by diet. i, Phenotypes featuring a diet main effect and/or a diet × age interaction but no effect of age. f all phenotypes shown in g–i. ICC = intraclass correlation. Statistical effect size comparisons were performed via two-sided z-tests. Our analyses are based on unadjusted p-values. Source data are provided as a Source Data file.

Fig. 9. ‘Anti-aging’ effects were frequently age-independent in nature.

The schematic illustrates major scenarios by which PAAIs could influence aging phenotypes. First, interventions could have no measurable effect on a set of phenotypes or even accentuate age-dependent phenotypic change. ASPs countered by an intervention could be influenced in ways consistent with a targeting of the mechanisms underlying age-dependent phenotypic change: In this case, PAAI effects should become apparent only after the onset of aging-associated phenotypic change, but not at younger ages (rate effect). PAAI effects at a young age (prior to the age when age-dependent phenotypic change becomes first detectable) indicate that it is not the age-dependent change that is being targeted (baseline effect). Although our studies revealed examples of both rate and baseline effects, many ‘anti-aging’ effects fell into the latter category (age-independent effects that do not provide evidence for a slowed aging rate). Ignoring this distinction would lead to a substantial overestimation of the extent by which PAAIs slow the aging process. Created with BioRender.com.