Abstract

   While previous studies identified common genetic variants associated
   with longevity in centenarians, the role of the rare loss-of-function
   (LOF) mutation burden remains largely unexplored. Here, we investigated
   the burden of rare LOF mutations in Ashkenazi Jewish individuals from
   the Longevity Genes Project and LonGenity study cohorts using
   whole-exome sequencing data. We found that centenarians had a
   significantly lower burden (11-22%) of LOF mutations compared to
   controls. Similar effects were also observed in their offspring.
   Gene-level burden analysis identified 35 genes with depleted LOF
   mutations in centenarians, with 14 of these validated in the UK
   Biobank. Mendelian randomization and multi-omic analyses on these genes
   identified RGP1, PCNX2, and ANO9 as longevity genes with consistent
   causal effects on multiple aging-related traits and altered expression
   during aging. Our findings suggest that a protective genetic
   background, characterized by a reduced burden of damaging variants,
   contributes to exceptional longevity, likely acting in concert with
   specific protective variants to promote healthy aging.

   Subject terms: Genotype, Ageing, Genetic variation
     __________________________________________________________________

   Previous studies have identified common genetic variants linked to
   longevity, but the impact of rare damaging mutations remains unclear.
   Here, the authors show that centenarians carry fewer harmful
   loss-of-function mutations and identify genes that may contribute to
   extreme longevity and healthy aging

Introduction

   Aging is a complex process characterized by an accumulation of
   molecular damage, progressive decline in physiological function,
   increased susceptibility to disease, and, ultimately, higher risk of
   mortality^[42]1. While chronological age is a major risk factor, there
   is remarkable variability in how individuals age, with some
   experiencing severe disability and premature death while others
   maintain good health well into old age^[43]2,[44]3. This heterogeneity
   suggests that aging is a multifactorial process shaped by both genetic
   and environmental factors^[45]4.

   At the extreme end of the lifespan spectrum are centenarians,
   individuals with exceptional longevity who have reached the age of 100
   years or more. Centenarians represent a rare and valuable model of
   successful aging, often displaying delayed onset or escape from major
   age-related diseases such as cardiovascular disease, diabetes, and
   dementia^[46]5,[47]6. Furthermore, many maintain physical and cognitive
   function, as well as independence, well into old age^[48]7.
   Understanding the factors that contribute to their exceptional
   longevity could provide valuable insights into the biology of healthy
   aging and lifespan determination.

   Studies in model organisms have firmly established that lifespan has a
   significant genetic component. Single gene mutations in pathways
   related to insulin/insulin-like growth factor-1 (IGF-1) signaling,
   mechanistic target of rapamycin (mTOR) signaling, and AMP-activated
   protein kinase (AMPK) signaling have been shown to dramatically extend
   lifespan in yeast, worms, flies, and mice^[49]8. Many of these pathways
   are evolutionarily conserved, suggesting they may play a role in human
   aging as well. Indeed, functional variants in the IGF-1 receptor have
   been identified in centenarians, supporting a role for this pathway in
   exceptional longevity^[50]9.

   In humans, genome-wide association studies (GWAS) have identified
   numerous common genetic variants associated with longevity, defined as
   attaining exceptional old age or having long-lived
   parents^[51]10,[52]11. However, these variants explain only a small
   portion of the heritability (12%)^[53]11, suggesting that rare variants
   may also play an important role^[54]12. Rare variants, particularly
   those that lead to loss of gene function (LOF), are of great interest
   in studying human lifespan. LOF variants, including nonsense,
   splice-site, and frameshift mutations, are generally deleterious and
   subject to strong purifying selection^[55]13. An increased burden of
   LOF mutations has been observed in individuals with shorter lifespans
   and shorter period of life people spent free of disease (or
   healthspan), suggesting this may significantly impact human
   health^[56]14. However, LOF variants that confer protective effects,
   such as those in the APOC3 and PCSK9 genes associated with a lower risk
   of cardiovascular disease have also been identified^[57]15,[58]16.

   Despite the growing evidence for the importance of rare variants in
   aging, the overall burden of LOF mutations in exceptionally long-lived
   individuals compared to controls has not been systematically examined.
   A previous study observed no difference in the burden of pathogenic
   variants between centenarians, their offspring, and controls^[59]17.
   However, this study did not specifically focus on LOF variants or
   incorporate key covariates that may introduce batch effects confounding
   the results. Furthermore, the sample size was smaller than in the
   present study, limiting the power to detect significant differences.
   Another study found that the burden of rarest protein-truncating
   variants (PTVs) in two large cohorts was negatively associated with
   human healthspan and lifespan, accounting for 0.4 and 1.3 years of
   their variability, respectively^[60]14.

   In this study, we leveraged whole-exome sequencing data from a large
   cohort of Ashkenazi Jewish centenarians and controls to comprehensively
   compare the burden of rare LOF variants (Fig. [61]1a). By focusing on a
   genetically homogeneous population, we minimized the potential
   confounding effects of population stratification. Importantly, we
   incorporated the dates of recruitment and birth as coefficients in our
   analysis to control for cohort effects and potential secular trends in
   environmental and lifestyle factors that may impact lifespan. Our
   results suggest that centenarians have a lower burden of LOF mutations
   compared to controls. This depletion was observed across multiple
   categories of predicted deleterious variants. Furthermore, we performed
   a genome-wide association study to identify specific genes and pathways
   that were enriched for protective variants in centenarians. Several
   genes reached suggestive significance levels, and pathway analysis
   revealed a depletion of variants in pathways related to hyaluronan
   metabolism, G-protein receptors, post-translational protein
   modification, and mitochondrial translation. Notably, 14 out of 35 of
   these gene associations were validated in an independent cohort from
   the UK Biobank based on parental lifespan-related traits, supporting
   the reproducibility of our findings. Together, these results provide
   new insights into the genetic architecture of human exceptional
   longevity and highlight potential molecular mechanisms that may
   contribute to healthy aging. Further studies will be necessary to
   validate and functionally characterize the roles of these genes and
   pathways in promoting longevity.

Fig. 1. Demographic characteristics and mutation burden of study cohorts.

   [62]Fig. 1
   [63]Open in a new tab

   a Schematic graph of the study design and analysis performed. Created
   in BioRender. Biosciences, R. (2022) BioRender.com/i68q986. b Date of
   birth and date of recruitment for centenarians, offspring, and controls
   from the Longevity Genes Project (LGP) and LonGenity study cohorts.
   Each vertical bar represents an individual, color-coded by their cohort
   status. c Distribution of the cumulative mutation burden across
   different categories of predicted deleterious variants in centenarians,
   offspring, and controls. The categories include pLOF only, pLOF and
   missense, pLOF and predicted deleterious missense (5/5 algorithms
   predict a deleterious variant), and pLOF and predicted deleterious
   missense (at least 1/5 algorithms predict a deleterious variant). Each
   dot represents an individual, with the violin plot showing the
   distribution of mutation counts within each category and cohort. OPEL:
   Offspring of Parents with Exceptional Longevity; OPUS: Offspring of
   Parents with Usual Survival. Source data are provided as a Source Data
   file.

Results

   The whole-exome sequencing data was obtained from 637 centenarians, 917
   offspring of centenarians, and 595 controls from the Longevity Genes
   Project (LGP) and LonGenity study cohorts of Ashkenazi Jewish
   individuals (Table [64]1, Fig. [65]1a, Methods)^[66]18. Based on the
   demographic characteristics, participants were recruited continuously
   over a period of 20 years (2000–2020). However, the recruited
   centenarians were mostly born between 1900 and 1920, while most of the
   offspring and controls were born between 1920 and 1960 (Fig. [67]1b).
   This suggests that the direct comparison of mutation burden between
   centenarians and controls may potentially be confounded by the date of
   recruitment and date of birth.

Table 1.

   Demographic characteristics of the study cohorts
   Cohort Sample size Males Females Mean age at visit (age range, years)
   Standard deviation of age at visit (years)
   Longevity Genes Project Control 224 125 99 70.99 (42.26-93.33) 9.85
   Longevity Genes Project Control (Filtered) 147 84 63 66.27
   (42.26-84.51) 7.96
   Longevity Genes Project Offspring 473 246 227 67.53 (42.88-92.45) 7.98
   Longevity Genes Project Proband (Centenarians) 637 464 173 97.70
   (84.45-110.10) 3.35
   Longevity Genes Project Proband (Centenarians, Filtered) 338 249 89
   99.46 (84.45-110.10) 3.27
   LonGenity Study Offspring of Parents with Exceptional Longevity 444 265
   179 74.20 (61.90-94.08) 6.13
   LonGenity Study Offspring of Parents with Usual Survival 371 196 175
   76.32 (64.67-97.93) 7.02
   LonGenity Study Offspring of Parents with Usual Survival (Filtered) 273
   141 132 73.34 (64.67-87.33) 5.22
   [68]Open in a new tab

   The table presents the sample size, number of males and females, female
   proportion, mean age at visit, and standard deviation of age at
   visit for each cohort in the Longevity Genes Project and LonGenity
   Study.

   We identified loss-of-function (LOF) mutations based on the following
   criteria: alternate allele frequency (AAF) < 1%, Hardy–Weinberg
   equilibrium (HWE) threshold of 10^−15, and variant missingness <10%. We
   classified the variants into different categories based on their
   predicted deleteriousness: pLOF only, pLOF and missense, pLOF and
   predicted deleterious missense (5/5 algorithms predict a deleterious
   variant), and pLOF and predicted deleterious missense (at least 1/5
   algorithms predict a deleterious variant). The deleteriousness of
   missense variants was assessed using five different computational
   methods (Method).

   We counted the cumulative mutation burden in centenarians, their
   offspring, and controls across different categories of predicted
   deleterious variants. Consistent with the potential confounding effects
   of dates of recruitment and birth, we initially observed a similar
   distribution of LOF mutations across the different categories in
   centenarians and controls (Fig. [69]1c). We performed quality control
   and filtering, retaining 338 centenarians with recorded age over 100
   years old and 420 controls with age less than 90 years old
   (Table [70]1, Methods). We observed a similar distribution of raw
   mutation count after filtering (Supplementary Fig. [71]1) We then
   performed the count-based burden test using linear regression models.
   Furthermore, we found that even without adjusting for potential
   confounders, offspring but not centenarians showed a significantly
   lower mutation burden in all pLOF categories (Supplementary
   Fig. [72]2). This is likely due to the smaller batch effect between
   offspring group and control, compared to the centenarian group. We also
   showed that there is no significant difference observed between
   centenarians and their offspring (Supplementary Fig. [73]3).

   To account for these potential confounders, we binned the dates of
   recruitment and birth and added them as coefficients in the burden test
   model. After adjusting for these covariates, we found a consistent and
   significant trend of lower burden of LOF mutations in centenarians and
   their offspring compared to controls across all categories of predicted
   deleterious variants (Fig. [74]2). Notably, the depletion of LOF
   variants was statistically significant for centenarians in all
   categories, including the pLOF-only category (b = −5.5, p = 0.0453).
   The effect sizes for centenarians ranged from −5.5 to −39.6, indicating
   a 11% to 22% reduction in mutation burden compared to controls.
   Furthermore, the offspring of centenarians also exhibited a
   significantly lower mutation burden compared to controls in both the
   LGP (related to LGP centenarians) and LonGenity (unrelated to LGP
   centenarians) cohorts (Fig. [75]2). The effect sizes for offspring were
   smaller than those observed for centenarians, but still significant,
   with p-values ranging from 1.17e-07 to 4.99e-4 in the LGP cohort and
   from 4.52e-4 to 0.021 in the LonGenity cohort. These results suggest
   that the protective effect of a lower LOF mutation burden may be
   inherited by the offspring of centenarians, contributing to their
   increased likelihood of exceptional longevity.

Fig. 2. Burden test results for different categories of predicted deleterious
variants, adjusting for potential confounders.

   [76]Fig. 2
   [77]Open in a new tab

   The points show the effect sizes (beta coefficients) and the error bars
   show the 95% confidence intervals from the linear regression models,
   with the mutation burden as the dependent variable and the cohort
   status (centenarian or offspring, compared to control) as the
   independent variable. The models were adjusted for the binned date of
   recruitment and date of birth as covariates. The p-values for each
   comparison are shown on the right side of the plot. The categories of
   predicted deleterious variants include pLOF only, pLOF and missense,
   pLOF and predicted deleterious missense (5/5 algorithms predict a
   deleterious variant), and pLOF and predicted deleterious missense (at
   least 1/5 algorithms predict a deleterious variant). Source data are
   provided as a Source Data file.

   We also performed a sensitivity analysis by using different covariates,
   including age at recruitment, top 10 genetic principal components,
   numerical date of birth, and date of recruitment, and found consistent
   results for centenarian offspring in the LGP cohort (Supplementary
   Fig. [78]4). Statistical significance for centenarians and the
   LonGenity cohort was sensitive to the choice of covariates, suggesting
   that the genetic associations with longevity are complex and possibly
   influenced by unmeasured factors.

   To identify specific genes and pathways that carry a lower mutation
   burden in centenarians, we performed a gene-level and pathway-level
   burden test. The gene-level analysis identified 35 genes that reached
   the significance level at FDR < 0.05 (Fig. [79]3a). Remarkably, 14 out
   of these 35 genes were validated in an independent study from the UK
   Biobank using parental lifespan-related traits (Fig. [80]3a)^[81]19.
   Note that this is an indirect validation as the genetics of exceptional
   longevity and parental lifespan, while having similarities, may still
   obtain different characteristics. Pathway-level analysis revealed
   processes related to hyaluronan metabolism, Class A/1 (Rhodopsin-like
   receptors), post-translational protein modification, and mitochondrial
   translation reached the significance level at FDR < 0.05 (Fig. [82]3b).
   We observed a mild inflation in our test statistics, with a genomic
   inflation factor (λ) of 1.57. After adjusting for the inflation, the
   top three pathways still reached the suggestive FDR threshold of 0.2.
   These results suggest that the depletion of mutations in these pathways
   may contribute to exceptional longevity.

Fig. 3. Gene-level and pathway-level burden test results.

   [83]Fig. 3
   [84]Open in a new tab

   a Manhattan plot showing the -log10(p-values) for the gene-level burden
   test in centenarians compared to the control group. Each dot represents
   a gene, with the x-axis showing the genomic position and the y-axis
   showing the statistical significance. The red horizontal line indicates
   the P-value threshold at FDR < 0.05 after adjusted for multiple
   comparisons. One-sided tests were performed since we only focused on
   genes with fewer LOF variants in centenarians and the offspring. The
   top genes reaching FDR < 0.05 are labeled, with those validated in the
   UK Biobank cohort marked with an “+”. b Pathway enrichment analysis
   results in centenarians compared to the control group. The scatter plot
   shows the enriched pathways, with the x-axis representing the expected
   -log10(p-values) and the y-axis representing the observed
   -log10(p-values). Two horizontal dashed lines show the P-value
   threshold at FDR < 0.05 and suggestive threshold of FDR < 0.2 after
   adjusted for multiple comparisons. One-sided tests were performed. Top
   pathways are labeled, with pathways passed FDR < 0.05 colored in black
   and pathways passed suggestive threshold of FDR < 0.2 colored in gray.
   We also calculated the genomic inflation factor (λ) of 1.57. Top
   pathways still reach the suggestive threshold of FDR after adjusting
   for inflation. Source data are provided as a Source Data file.

   To further investigate the potential causal effects of the identified
   longevity-associated genes on lifespan-related traits, we performed
   Mendelian Randomization (MR) analyzes using public blood gene
   expression QTL data from eQTLgen and GWAS summary statistics of
   multiple lifespan-related traits (Fig. [85]4a)^[86]20. It is important
   to note that while the MR analysis uses common variants (eQTLs) rather
   than rare coding variants, it can provide complementary evidence about
   a gene’s role in longevity through different mechanisms. MR analysis
   revealed that seven genes had significant causal effects on multiple
   lifespan-related traits, such as frailty index, healthspan, lifespan,
   and extreme longevity (90th and 99th percentiles), and lifespan-GIP1
   (the genetic principal component of healthy longevity, Methods). Among
   them, three genes (RGP1, PCNX2, and ANO9) showed consistent
   pro-longevity effects across the multiple traits tested, supporting
   their potential roles in promoting longevity as suggested by burden
   analysis, while the other four genes showed anti-longevity effects. On
   the other hand, two of the genes (DYNC1H1 and GALNT12) only show a
   significant protective effect on one trait (lifespan and extreme
   longevity at 99th percentile), while PKP4 only shows a significant
   positive effect on healthspan but not in other traits. The other four
   genes (ZNF446, PLA2G4B, EFNA3, and ABCF3) show inconsistent effects on
   lifespan-related traits.

Fig. 4. Mendelian randomization and multi-omic analyzes of
longevity-associated genes.

   [87]Fig. 4
   [88]Open in a new tab

   a Mendelian Randomization (MR) analyzes of blood gene expression on
   aging-related traits. The point shows the estimated causal effect sizes
   (beta) and the error bars show the 95% confidence intervals for each
   gene on various aging-related traits, including aging-GIP1, frailty
   index, healthspan, lifespan, and extreme longevity (90th and 99th
   percentiles). The forest plot is colored based on the significance and
   consistency of the causal effect estimates across different traits.
   Genes with FDR < 0.05 after being adjusted for multiple comparisons are
   colored in orange. Two-sided tests were performed. The blood eQTLs
   summary statistics are obtained from eQTLgen study (n = 31,684). b–e
   Bar plots showing the -log10(P-value) multiplied by the sign of the
   effect size (beta) for each longevity-associated gene. Associations
   with nominal P < 0.05 are colored in yellow, and those with FDR < 0.05
   after being adjusted for multiple comparisons are colored in red.
   Two-sided tests were performed. Data sources include exome-wide
   association analysis with parental lifespan from GeneBass (b),
   age-related changes in promoter DNA methylation using data from 500
   individuals in the MGB biobank (c), age-related changes in blood gene
   expression using data from the transcriptome-wide association study
   (TWAS) for aging by Peters et al. 2015 (d), and age-related changes in
   plasma protein levels using Olink data from 53,015 UK Biobank
   participants (e). Positive values indicate an increase, while negative
   values indicate a decrease with age. f Heatmap showing the significance
   scores of longevity-associated genes across different aging signatures,
   including human aging, rodent aging, and interventions (caloric
   restriction, rapamycin, and growth hormone deficiency). Color intensity
   represents the significance of the association, with red indicating a
   positive significance score and blue a negative significance score.
   Associations that reach nominal significance (P < 0.05) are marked with
   small circles, and those with FDR < 0.05 after being adjusted for
   multiple comparisons are marked with large circles. Two-sided tests
   were performed. Source data are provided as a Source Data file.

   We then profiled the multi-omic associations of the identified
   longevity-associated genes to provide a systematic evaluation of their
   expression and regulation during aging (Fig. [89]4b–e). Comparison with
   exome-wide gene-level associations with parental lifespan obtained from
   GeneBass (Fig. [90]4b)^[91]19, showed that six out of seven causal
   genes were significantly associated with parental lifespan, three genes
   (MLXIP, PCNX2, and DYNC1H1) remain significant after corrected with
   multiple-testing with FDR. Analysis of age-related changes in promoter
   DNA methylation using data from 500 individuals in the Massachusetts
   General Brigham (MGB) biobank (Fig. [92]4c) revealed significant
   changes for most longevity-associated genes, except two (RGP1 and
   BCLAF1). Similarly, age-related changes in blood gene expression
   obtained from the transcriptome-wide association study (TWAS) for aging
   by Peters et al. (Fig. [93]4d) showed significant changes for genes
   such as OPN3, PCNX2, GALNT12, and RGP1^[94]21. Furthermore, age-related
   changes in plasma protein levels using Olink data from 53,015 UK
   Biobank participants (Fig. [95]4e) revealed significant changes for
   proteins encoded by DYNC1H1 and FLT4 genes. The results suggest that
   the expression and regulation of these longevity-associated genes are
   altered during the aging process.

   To gain further insights into the potential relevance of the identified
   longevity-associated genes in aging and interventions, we further
   compared their significance scores across different signatures of aging
   and longevity interventions (Fig. [96]4f)^[97]22,[98]23. The signature
   analysis results in 69 significant associations after adjusting for
   multiple testing of 266 tests using FDR (Fig. [99]4f). It revealed that
   many of these genes (18 out of 21 tested) were also significantly
   associated with aging in humans and rodents, as well as with
   interventions known to extend lifespan, such as caloric restriction
   (ABCF3, CKAP2L, and CEP68), rapamycin treatment (PKP4, CTNND1, and
   RTRAF), growth hormone deficiency (HOGA1, ANKRD33, and MLXIP), as well
   as overall lifespan after intervention (HOGA1). Together, this
   multi-layered evidence supports the potential roles of these genes in
   regulating healthy aging and longevity.

Discussion

   In this study, we have discovered that centenarians, within the large
   cohort we examined, possess a significantly lower burden of predicted
   deleterious LOF variants compared to controls. This finding suggests
   that a protective genetic background, characterized by the depletion of
   damaging coding mutations, contributes to the exceptional longevity of
   centenarians. Notably, we also observed a lower mutation burden in
   centenarian offspring, although the effect was less pronounced. These
   findings support the notion of a heritable component to longevity
   outside of protective and common variants and suggest that the combined
   genetic background, including protective variants and depletion of
   damaging variants, may be transmitted across generations to support
   exceptional longevity.

   Our results are consistent with previous studies that reported an
   increased burden of LOF variants in individuals with shorter lifespans
   and age-related diseases^[100]14,[101]24, and provide further evidence
   for the role of rare coding variants in extreme human longevity. Our
   study extends these findings by demonstrating that the depletion of LOF
   variants in centenarians is not limited to the rarest variants but is
   observed across multiple categories of predicted deleterious variants.
   However, our findings contrast with those of another study that
   observed no difference in the burden of pathogenic variants between
   centenarians, their offspring, and controls^[102]17. This discrepancy
   may be due to differences in study design, such as the focus on LOF
   variants specifically, the larger sample size of our study, and the
   adjustment for potential confounding factors such as date of
   recruitment, age at recruitment, and date of birth. Besides, due to the
   retrospective nature of the centenarian study, the centenarians usually
   have different demographic properties (age, date of birth, and
   potentially other early life exposures) compared to the control group.
   While this can be addressed by including these features as covariates,
   this demographic disparity between centenarians and controls emerges as
   a critical factor limiting the statistical power of centenarian studies
   (Fig. [103]1). In contrast, centenarian offspring, demographically more
   similar to controls, yield stronger statistical evidence, corroborating
   our findings in centenarians. Future prospective studies with improved
   demographic matching are essential to elucidate the role of LOF
   variants in exceptional longevity.

   Our pathway analysis revealed that centenarian exomes are depleted of
   LOF variants in several pathways related to aging and disease,
   including Class A/1 (Rhodopsin-like receptors), hyaluronan metabolism,
   post-translational protein modification, and mitochondrial translation.
   Class A/1 (Rhodopsin-like) receptors are involved in various
   physiological processes and have been implicated in age-related
   diseases, suggesting their potential role in longevity^[104]25.
   Hyaluronan is a key component of the extracellular matrix that has been
   shown to decline with age, and its increase contributes to the
   extension of lifespan^[105]26. Variants that maintain hyaluronan
   homeostasis may, therefore, promote healthy aging in humans.
   Post-translational protein modifications play crucial roles in protein
   function and stability, and their dysregulation has been associated
   with various age-related diseases^[106]1. Mitochondrial translation has
   also been linked to lifespan extension in model organisms^[107]27.

   To complement our analysis of rare LOF variants, we also investigated
   the causal role of identified longevity genes in aging-related traits
   using MR analyzes. This approach allows us to infer potential causal
   relationships between gene expression and phenotypes of interest by
   using eQTLs (common variants that are associated with gene expression)
   as instrumental variables. Our MR analyzes provided evidence for the
   causal effects of several longevity-associated genes, including RGP1,
   PCNX2, and ANO9, on multiple aging-related traits. PCNX2 was identified
   to be associated with longevity in an independent GWAS study^[108]28,
   while ANO9 was associated with various cancers^[109]29. These findings
   suggest that these genes may directly influence the aging process and
   contribute to the extended healthspan and lifespan. The consistent
   causal effect estimates across different aging-related traits further
   support the robustness of these associations. Interestingly, our
   analyzes also revealed genes with more nuanced effects on longevity.
   For instance, DYNC1H1 and GALNT12 showed significant deleterious
   effects on only one trait each (lifespan and extreme longevity at the
   99th percentile, respectively), while PKP4 demonstrated a significant
   positive effect solely on healthspan. This suggests that these genes
   may influence particular aspects of the aging process rather than
   having a broad impact on all longevity-related traits. Moreover, the
   inconsistent effects observed for genes such as ZNF446, PLA2G4B, EFNA3,
   and ABCF3 across different lifespan-related traits underscore the
   complexity of genetic influences on aging and longevity. The multi-omic
   analyzes revealed that the expression and regulation of many
   longevity-associated genes are altered during aging, specifically, 29
   out of 31 for DNA methylation, 4 out of 11 for gene expression, and 2
   out of 2 for plasma protein (Fig. [110]4). Follow-up studies are needed
   to elucidate the specific mechanisms by which these genes and their
   encoded proteins contribute to healthy aging and longevity.

   Future studies could also explore the relationship between the burden
   of deleterious germline mutations and the rate of biological aging in
   centenarians and the general population. Epigenetic clocks, which
   measure biological age based on DNA methylation patterns, have emerged
   as a promising tool for assessing the pace of aging^[111]30,[112]31.
   Previous studies have shown that centenarians exhibit slower epigenetic
   aging rates compared to the general population^[113]32. Integrating
   rare variant burden data with epigenetic clock measures could provide
   novel insights into the interplay between genetic and epigenetic
   factors in shaping the rate of aging and exceptional longevity,
   especially with current standardized tools like ClockBase and
   Biolearn^[114]33,[115]34, as well as advanced aging clocks, including
   GrimAge2^[116]35, DunedinPace^[117]36, and causality-enriched
   clocks^[118]37. Such studies may uncover whether the reduced burden of
   harmful mutations observed in centenarians contributes to their slower
   biological aging rates.

   Our study also has several limitations. First, while we adjusted for
   several important covariates, there may be other confounding factors
   that were not accounted for, such as environmental exposures and
   lifestyle factors. Second, our study focused on a specific population
   (Ashkenazi Jews), although validation analysis in UK biobank suggests
   that the result may be generalizable to other ethnic groups. Future
   studies in diverse populations will be necessary to confirm the
   generalizability of our findings. Third, the validation analysis is
   based on parental lifespan traits in the UK biobank. Although previous
   studies on common variants show a substantial similarity between
   parental lifespan and exceptional longevity (r[g] = 0.81)^[119]38, it
   is unclear how similar the rare genetic variants contribute to these
   two traits. Future validation and meta-analysis with other centenarian
   cohorts may help strengthen the robustness of our findings. Fourth, our
   study relied on computational predictions of variant deleteriousness,
   which may not always reflect the true biological impact of a variant.
   Functional studies will be necessary to validate the causal roles of
   the identified variants and genes in longevity.

   It is important to acknowledge that some LOF and missense variants can
   be protective, as demonstrated by previous studies^[120]39–[121]41.
   However, our hypothesis is that the overall probability of LOF variants
   being protective is lower than the probability of them being
   deleterious. This is because damaging a component in a complex system
   is more likely to have a detrimental effect than a protective
   one^[122]42. Additionally, there is a selection bias, as highly
   damaging mutations are under-represented in the population, while
   highly protective mutations are preserved^[123]43. These factors may
   explain the small effect sizes observed in our study. It should also be
   noted that we did not identify any protective LOF variants (i.e.,
   enrichment of LOF variants in centenarians) as demonstrated in previous
   study^[124]18, because we used a one-tailed test, focusing only on the
   depletion of LOF variants.

   In conclusion, our study provides new insights into the genetic
   architecture of human exceptional longevity, exemplified by individuals
   who live to 100 years or beyond, highlighting the importance of rare
   LOF variants and identifying novel genes and pathways that may promote
   healthy aging. We demonstrate that centenarians have a lower burden of
   predicted deleterious LOF variants compared to controls and that this
   protective genetic background may be transmitted across generations.
   Our findings also underscore the complex interplay between genetic
   variation, environmental factors, and age-related diseases in shaping
   human lifespan. Further studies in diverse populations and integrating
   multiple omics data will be necessary to fully elucidate the mechanisms
   underlying exceptional longevity and develop targeted interventions to
   promote healthy aging. Nonetheless, our results represent an important
   step towards understanding the genetic basis of human longevity and
   provide a foundation for future studies in this field.

Methods

Study population and data collection

   The study population was derived from two ongoing studies of aging and
   longevity in the Ashkenazi Jewish population: the cross-sectional
   Longevity Genes Project (LGP) and the longitudinal LonGenity
   study^[125]18. The LGP cohort consisted of 637 individuals with
   exceptional longevity, 473 offspring of long-lived individuals, and 224
   controls, while the LonGenity cohort included 444 offspring of
   centenarians and 371 controls. All participants provided written
   informed consent, and the study was approved by the Institutional
   Review Board at Albert Einstein College of Medicine.

   For the analysis, we applied filtering criteria to ensure the inclusion
   of appropriate individuals in each group. In the centenarian group, we
   removed individuals with a  death or dropout record before 100 years,
   retaining 338 exceptionally long-lived centenarians. Similarly, in the
   control group, we removed individuals without death or dropout record
   before 90 years, resulting in 147 individuals from the LGP cohort and
   273 individuals from the LonGenity cohort being included in the
   analysis (Table [126]1).

Whole-exome sequencing

   DNA samples from all participants were subjected to whole-exome
   sequencing using the Illumina HiSeq 2000 platform at the Regeneron
   Genetics Center^[127]17. The sequencing reads were aligned to the human
   reference genome (hg38) using the Burrows-Wheeler Aligner (BWA-mem
   v0.7.17)^[128]44, and duplicate reads were removed using Picard tools
   (version 1.96, [129]http://broadinstitute.github.io/picard/). Variant
   calling was performed using the Genome Analysis Toolkit (GATK
   v3.7)^[130]45.

Quality control and variant annotation

   After genomic principal component analysis (PCA), four individuals with
   non-European ancestry were excluded from the study. Quality control
   filtering was applied to remove potentially false-positive variants and
   genotype calls. Variants were filtered based on the following criteria:
   alternate allele frequency (AAF) < 1% in the Ashkenazi Jewish
   population, Hardy–Weinberg equilibrium (HWE) P-value > 10^−15, and
   variant missingness <10%, as suggested by a previous study^[131]46.
   After QC filtering, autosomal-only variants with a minimum allele count
   (MAC) of 1 were divided into sets for centenarians, offspring, and
   controls for downstream analysis.

   Loss-of-function (LOF) variants were defined as nonsense, splice-site,
   or frameshift mutations. Missense variants were classified as (1)
   possible deleterious missense mutation if they were predicted to be
   damaging by at least 1 out of 5 algorithms (SIFT^[132]47,
   Polyphen2_HDIV^[133]48, Polyphen2_HVAR^[134]48, LRT^[135]49, and
   MutationTaster^[136]50) or (2) deleterious missense mutation if all
   five algorithms predicted them to be damaging. SIFT (v6.2.1),
   Polyphen2_HDIV (v2.2.2), Polyphen2_HVAR (v2.2.2), LRT (v2016), and
   MutationTaster (v2021) were used in this analysis.

Burden test analysis

   Prior to the burden test, we removed the individual in the extreme
   longevity group with a lifespan or last reported age less than 100
   years old. Therefore, only the 338 centenarians are kept. Similarly,
   individuals in the control group with last reported age larger than 90
   years old were also removed, with the remaining 147 individuals from
   LGP and 273 individuals for lonGenity (Table [137]1).

   Descriptive statistics were used to summarize the demographic
   characteristics of the study population. The cumulative mutation burden
   for each individual was calculated as the total number of
   population-level LOF (pLOF) and predicted deleterious missense
   variants. Mutation burden is calculated based on different categories
   of predicted deleterious variants (pLOF only, pLOF and deleterious
   missense [5/5 algorithms], and pLOF and possible deleterious missense
   [≥1/5 algorithms], pLOF and all missense).

   Count-based burden tests were performed using linear models with binned
   covariates to account for potential confounding factors, such as date
   of recruitment, date of birth, gender, age at visit, and top four
   genomic principal components^[138]51. The cumulative mutation burden
   was used as the dependent variable, and the independent variables
   included centenarian status (or offspring status), binned date of
   recruitment, and binned date of birth. Sensitivity analyzes were
   conducted by including additional covariates, such as age at
   recruitment, top 10 genetic principal components, numerical date of
   birth (i.e., number of days since 1900-01-01), and date of recruitment.

Gene-level and pathway-level burden analysis

   Gene-level and pathway-level burden tests were performed using linear
   models, with the cumulative mutation burden in each gene or pathway as
   the dependent variable and centenarian status as the independent
   variable. Only genes containing at least five pLOF variants across the
   cohort were included. In total, 4925 unique genes were tested, and the
   significance threshold for gene-level tests was set at FDR < 0.05.
   Significant gene-level associations were replicated using summary
   statistics from a gene-based association study of paternal or maternal
   lifespan in the GeneBass from UK biobank^[139]19. The significance
   threshold for replication was set at P < 0.05.

Mendelian randomization

   To investigate the causal relationships between gene expression and
   aging-related traits, we performed Mendelian Randomization (MR)
   analyzes using blood cis-eQTL data from eQTLgen, which includes 31,684
   blood samples from 37 studies^[140]20. The outcome traits included
   aging-GIP1, frailty index, healthspan, lifespan, and extreme longevity
   (90th and 99th percentiles).

   The parental lifespan GWAS was used as a proxy for individual lifespan
   and included 512,047 mothers and 500,193 fathers of European
   ancestry^[141]11. The extreme longevity GWAS included 11,262 European
   subjects with a lifespan above the 90th percentile and 25,483 controls
   below the 60th percentile age^[142]10. Healthspan, defined as the age
   of the first incidence of major age-related diseases or death, was
   analyzed using a GWAS of 300,447 UK Biobank participants aged
   37–73^[143]52. The frailty index GWAS included 164,610 UK Biobank
   participants aged 60–70 and 10,616 Swedish TwinGene participants aged
   41–87^[144]53. Aging-GIP1, the first genetic principal component of six
   human aging traits, captures both length of life and well-being
   indices^[145]54.

   We performed cis-Mendelian Randomization following the approach
   described by Ying et al^[146]37. Genetic variants strongly associated
   with whole blood gene expression levels (FDR < 0.05) were selected as
   instrumental variables for the MR analysis. To minimize pleiotropic
   effects, only cis-eQTLs (located within 2 MB of target genes) were
   used, and LD clumping was applied to remove eQTLs with strong LD
   (r^2 > 0.3). We employed three MR methods based on the number of
   available eQTLs: Wald ratio for a single eQTL, generalized inverse
   variance weighted (gIVW) for at least two eQTLs, and generalized
   MR-Egger regression (gEgger) for at least three eQTLs^[147]55. The
   gEgger method is robust to directional pleiotropy, we therefore
   reported the P value from gEgger if pleiotropy is detected by gEgger
   intercept.

Multi-omic analysis of the identified longevity-associated genes

   To systematically evaluate the expression and regulation of the
   identified longevity-associated genes during aging, we profiled their
   multi-omic associations using various datasets. We obtained the
   exome-wide gene association with parental lifespan using summary
   statistics from GeneBass^[148]19. Blood gene expression changes with
   age were obtained from the transcriptome-wide association study (TWAS)
   for aging by Peters et al^[149]21.

   Age-related changes in promoter DNA methylation were assessed using
   data from 500 individuals in the Mass General Brigham (MGB) Biobank,
   which is also described in this study^[150]56. DNA methylation profiles
   were generated using the Illumina Infinium MethylationEPIC v2.0 array,
   which covers over 935,000 CpG sites enriched for regulatory
   regions^[151]56. The cohort comprised subjects of diverse ages, roughly
   balanced between male and female, and generally representative of the
   racial/ethnic distribution of the local area. For each CpG site
   associated with our identified longevity-associated genes, we performed
   a linear regression to predict the methylation beta value using age,
   where the regression coefficient and p-value are calculated. The CpG
   with the strongest association with age is used to represent the
   result.

   Age-related changes in plasma protein levels were investigated using
   Olink proteomics data from 53,015 UK Biobank participants (UK Biobank
   Record Table [152]1072). Only two of our identified
   longevity-associated genes are presented in the Olink panel. We then
   performed a linear regression to predict the protein level using age,
   where the regression coefficient and p-value are calculated. FDR was
   applied to adjust for multiple testing of all 471 sites tested.

   We performed FDR to adjust for multiple tests in each omic layer.

Longevity signature analysis

   To further explore the potential relevance of the identified
   longevity-associated genes in aging and interventions, we compared
   their significance scores across different signatures of aging and
   longevity interventions using the GENtervention database^[153]57. For
   transcriptomic signatures of lifespan-extending interventions, we
   selected the ones reflecting the most established longevity
   interventions that were identified based on gene expression data from
   at least 3 independent sources, as described in Tyshkovskiy et al.
   2019^[154]23. The signatures included human aging and rodent aging, and
   interventions (caloric restriction, rapamycin treatment, and growth
   hormone deficiency). We also include signatures of lifespan across
   interventions based on a larger set of longevity and
   lifespan-shortening interventions^[155]22. The significance scores were
   calculated as the -log10(P-value) multiplied by the sign of the effect
   size (beta) for each gene in each signature. Nominal significance was
   set at P < 0.05. Hierarchical clustering with Euclidean distance was
   performed for the genes based on significance score.

Statistics & reproducibility

   The study included a total of 2149 participants: 338 centenarians (aged
   100 or older), 917 offspring of long-lived individuals, and 894
   controls. Detailed age and sex/gender breakdowns for each group are
   provided in Table [156]1. Sex and gender were considered in the study
   design and determined based on self-report at the time of recruitment.
   All participants provided written informed consent as stated in the
   “Study population and data collection” section. Participants were not
   compensated for their involvement in the study. No statistical method
   was used to predetermine the sample size. Data exclusion criteria are
   detailed in the “Study population and data collection” section. No
   other data were excluded from the analyzes. Statistical analyzes
   primarily employed linear models for burden tests and Mendelian
   Randomization, with adjustments for potential confounding factors as
   described in the “Burden test analysis” and “Mendelian Randomization”
   sections. Multiple testing corrections were applied using FDR. The
   experiments were not randomized, and the investigators were not blinded
   to allocation during experiments and outcome assessment, as this was an
   observational genetic study. Reproducibility was addressed through
   replication in independent datasets (UK Biobank).

Reporting summary

   Further information on research design is available in the [157]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [158]Supplementary Information^ (304.9KB, pdf)
   [159]Peer Review File^ (2.7MB, pdf)
   [160]41467_2024_52967_MOESM3_ESM.pdf^ (83KB, pdf)

   Description of Additional Supplementary Files
   [161]Supplementary Data 1^ (643.5KB, csv)
   [162]Supplementary Data 2^ (4MB, csv)
   [163]Reporting Summary^ (1.1MB, pdf)

Source data

   [164]Source Data^ (1.5MB, zip)

Acknowledgements