Genetic architecture of human aging and longevity: Insights from genome-wide association studies

Abstract

Aging represents a fundamental evolutionary feature shared across all living organisms, intrinsically coupled with development and lifespan. It is orchestrated by a complex polygenic architecture involving numerous small-effect variants distributed across diverse biological pathways, giving rise to striking interindividual variation in aging trajectories and lifespan. Over the past decade and a half, genome-wide association studies (GWAS) have uncovered multiple loci associated with lifespan, healthspan, exceptional longevity, and aging, converging on key biological processes such as lipid metabolism, inflammation, insulin/IGF signaling, and DNA repair. These discoveries have illuminated conserved molecular networks underlying the regulation of aging and longevity. Nevertheless, the identified variants collectively account for only a modest fraction of heritability, underscoring that aging and longevity arise from the cumulative and coordinated actions of myriad common alleles within complex biological networks. In this minireview, we synthesize major genetic insights from GWAS of aging and longevity, delineate recurrent pathways and molecular themes, and discuss how these findings refine our understanding of the genomic foundations of lifespan variation. We further highlight outstanding challenges, including phenotypic heterogeneity, ancestry-specific effects, and the limited predictive power of current models, and propose conceptual directions for future research aimed at establishing a more comprehensive and mechanistic framework for the genetic architecture of human aging and healthy longevity.

INTRODUCTION

Aging and human longevity represent complex biological traits that emerge from the cumulative interplay of genetic, environmental, and stochastic factors throughout the lifespan (1, 2). Although environmental influences and lifestyle factors exert strong effects (3), family and twin studies have long suggested a substantial genetic component to lifespan (4, 5), with heritability estimates ranging from approximately 10% to 30% (6-8). Given that aging is an inevitable biological process in all humans, elucidating its genetic determinants has become a central challenge in human genomics. Understanding its underlying architecture is expected to provide key insights into the biological mechanisms that modulate the pace of aging and resilience against age-related diseases.

Early candidate gene studies identified APOE and FOXO3A as consistently associated with longevity across multiple populations (9, 10), but these approaches were limited by small sample sizes and hypothesis-driven bias. The advent of genome-wide association studies (GWASs) enabled unbiased, large-scale analyses across diverse phenotypes. These include lifespan, typically defined as the total length of survival as a continuous trait (11-14); parental lifespan, a proxy phenotype that leverages offspring genotypes to increase statistical power (15-18); healthspan, which captures the duration of life free from major chronic diseases and functional decline (19, 20); and exceptional longevity, which focuses on individuals at the extreme tail of the survival distribution (21-23).

The first GWAS of exceptional longevity, published in Science in 2010 (24), was later retracted due to quality control issues and republished in a corrected form in PLOS One (25). Together, these studies over the past 15 years have identified numerous loci linked to lifespan variation, converging on key biological pathways such as lipid metabolism (26, 27), inflammation and immune regulation (28-30), insulin/IGF signaling (31, 32), DNA repair (33, 34), and mitochondrial homeostasis (35, 36). These pathways closely mirror aging-related regulatory networks observed in model organisms (37, 38), suggesting that human aging is governed by evolutionarily conserved molecular circuits rather than by a small number of trait-specific genes.

Despite these advances, the effect sizes of individual variants remain small, and the proportion of heritability explained by known loci is limited (39). This supports a highly polygenic model, in which aging and longevity arise from the cumulative action of hundreds to thousands of small-effect variants. Moreover, heterogeneity across study cohorts, ancestry backgrounds, and phenotype definitions complicates cross-study comparisons (40, 41). Some genetic signals overlap with major age-related diseases such as cardiovascular disease and dementia (42, 43), whereas others appear to contribute specifically to lifespan extension (44, 45), implying that disease susceptibility and aging resilience may be shaped by partly distinct genetic mechanisms.

In this minireview, we synthesize key findings from GWASs of aging and longevity and propose an updated perspective on the genetic basis of human lifespan variation. We highlight central features of the aging genetic architecture, including polygenicity, pathway convergence, and evolutionary constraint, and discuss future directions involving the expansion of population diversity, refined phenotype definitions, and deep exploration of rare variants and extreme longevity cohorts. Through this synthesis, we aim to delineate how accumulating GWAS evidence is reshaping our understanding of the genetic foundations of human aging and how these insights may inform precision prediction and intervention strategies for healthy longevity.

GENETIC INSIGHTS FROM GWAS OF AGING AND LONGEVITY

We comprehensively reviewed 29 published studies covering six phenotypes related to aging and longevity—aging, longevity, parental longevity, lifespan, parental lifespan, and healthspan—representing 45 independent GWAS analyses in total (Table 1). The number of studies (25) per phenotype was as follows: aging (n = 2) (12, 46, 47), longevity (n = 17) (9, 18, 21-23, 25, 41, 48-55), parental longevity (n = 2) (15, 16), lifespan (n = 6) (11, 13, 14, 40), parental lifespan (n = 15) (15-18, 56, 57), and healthspan (n = 3) (19, 20). From each study, we compiled loci and genes reported at genome-wide significant (P < 5 × 10⁻⁵) or suggestive (P < 5 × 10⁻⁸) significance levels for subsequent analysis (gene/variant counts: aging 29/60, longevity 70/167, parental longevity 8/82, lifespan 45/60, parental lifespan 54/281, healthspan 7/21).

Table 1.

15 Years of GWAS on aging and longevity

Study	Ancestry/ Population	Traits	Age criteria	Sample size		Number of variant	Number of GWAS		Note
				Stage 1 (Discovery)	Stage 2 (Replication)		Significant variants (meta)	Suggestive variants
Newman, 2010 (48)	European	Longevity	age ≥ 90 years	1,836 cases; 1,955 controls	2,594 cases; 3,431 controls	2,287,520	0 (273)	7	Meta-analysis based on 4 cohort, meta P < 1 × 10−4
Yashin, 2010 (11)	European	Lifespan	-	1,173 individuals	-	550,000	0	169	Suggestive threshold P ≤ 1 × 10−6
Deelen, 2011 (49)	European	Longevity	age ≥ 90 years	403 cases; 1,670 conrols	4,149 cases; 7,582 controls	516,721	0	62	Suggestive threshold P < 1 × 10−4
Malovini, 2011 (50)	Italian	Longevity	age ≥ 90 years	410 cases; 553 controls	-	298,715	0	67	Suggestive threshold P < 1 × 10−4
Nebel, 2011 (51)	German	Longevity	age ≥ 94 years	763 cases; 1,085 controls	-	664,472	1	≥ 16
Walter, 2011 (12)	European	Aging (time to death or censored)	-	25,007 individuals	1295 individuals	∼2,500,000	0	14
		Aging (age with no disease or death)		16,995 individuals	-		0	8
		Aging (merged)		-	-		0	3
Sebastiani, 2012 (25)	European	Longevity	age ≥ 95 years	801 cases; 914 controls	253 cases; 341 controls 60 cases; 2,863 controls	243,980	0	1
Deelen, 2014 (52)	European	Longevity	age ≥ 85 years	7,729 cases; 16,121 controls	13,060 cases; 61,156 controls	2,480,356	≥ 1	≥ 9
			age ≥ 90 years	5,406 cases; 15,112 controls	7,330 cases; 61,156 controls	2,470,825	≥ 1	≥ 12
Broer, 2015 (9)	European	Longevity	age ≥ 90 years	6,036 cases; 3,757 controls	-	∼2,500,000	0	7
Yashin, 2015 (40)	European	Lifespan	-	679 individuals	-	550,000	6	7	Only female
Zeng, 2016 (21)	Han Chinese	Longevity	age ≥ 100 years	2,178 cases; 2,299 controls	5,406 cases; 15,112 controls 1,030 caese; 368 controls	818,084	2	11
Pilling, 2016 (15)	Brithish	Parental lifespan	-	75,244 individuals	-	9,658,292	0	-
		Parental lifespan	-	63,775 individuals (paternal)			36	-
		Parental lifespan	-	52,776 individuals (maternal)			1	-
		Parental longevity	age ≥ 95 years	1,339 cases; 44,288 controls			2	-	Mother ≥ 98 years, fathers ≥ 95 years
Flachsbart, 2016 (53)	German	Longevity	age ≥ 90 years	1,458 cases; 6,368 controls	1,750 cases; 2,551 controls	142,136	≥1	84	Significant threshold P < 6.15 × 10−7
									Suggestive threshold P < 5 × 10−4
Tanaka, 2016 (57)	Brithish	Parental lifespan	-	5,716 individuals	1,951 individuals	∼2,500,000	0 (1)	-
Sebastiani, 2017 (54)	American	Longevity	age ≥ 95 years	2,070 cases; 6,259 controls	-	∼6,000,000	37	≥ 14
Joshi, 2017 (56)	European	Parental lifespan	-	586,626 individuals	20,518 individuals	Meta analysis	≥ 4	-
Pilling, 2017 (16)	European	Parental lifespan (combined parental attained age, Martingale residuals)	-	389,166 individuals	12,940 individuals 8,952 individuals	11,516,125	≥ 10	-
		Parental lifespan (mother's attained age)	-	412,937 individuals			3	-
		Parental lifespan (father's attained age)	-	415,311 individuals			8	-
		Parental longevity (both parents in top 10%)	age ≥ 87 years	7,182 cases; 79,767 controls			≥ 6	≥ 7
		Parental lifespan (combined parental age at death)	-	208,118 individuals			-	-
		Parental lifespan (mother's age at death)	-	246,941 individuals			3	-
		Parental lifespan (father's age at death)	-	317,652 individuals			8	-
Zeng, 2018 (22)	Han Chinese	Longevity	age ≥ 100 years	2,178 cases; 2,299 controls	NECS and IDEAL	∼5,600,000	0	≥ 11 in male ≥ 11 in female
Timmers, 2019 (17)	European	Parental lifespan	-	1,012,240 individuals	LifeGen	∼9,000,000	≥ 12	-	Significant threshold P < 2.5 × 10−8
Zenin, 2019 (19)	European	Healthspan		300,477 individuals	93,313 individuals	92,693,895	394	-
Deelen, 2019 (41)	European	Longevity	age ≥ 90 th percentile	11,262 cases; 25,483 controls	2,557 cases; 4,987 controls	Meta analysis	≥ 2	-
			age ≥ 99 th percentile	3,484 cases; 25,483 controls	1,911 cases; 4,987 controls		≥ 1	-
Wright, 2019 (18)	European	Longevity	age ≥ 91 years	12,663 cases; 469,403 controls	389,166 individuals, 658,000 individuals	536,775	≥ 1	-
			age ≥ 86 years (5%)	25,553 cases; 436,513 controls			0	-
		Parental lifespan	-	133,203 individuals (maternal, 1886-1918)		540,852	0	-
				167,179 individuals (paternal, 1886-1918)		541,017	≥ 2	-
				270,548 individuals (maternal, 1886-1940)		541,493	≥ 1	-
				309,383 individuals (paternal, 1886-1940)		541,614	≥ 4	-
Timmers, 2020 (46)	European	Aging (Longevity, Parental lifespan, Healthspan)	age ≥ 90 th percentile	1,349,462 individuals	-	7,320,282	≥ 24	-
Liu, 2021 (55)	Chinese	Longevity	age ≥ 90 years	5,767 cases; 5,278 controls	-	meta analysis	≥ 3	-
Gurinovich, 2021 (23)	European	Longevity	age ≥ 100 years	1,320 cases; 2,899 controls	312 cases; 638 controls	9,039,731	33	219	Sugesstivie threshold P < 5 × 10−6
Saul, 2022 (20)	Finnish	Healthspan	age ≥ 75 years	750 cases; 1,502 controls	-	∼9,600,000	0	≥ 6
			age ≥ 75 years	750 cases; 2,663 controls	-	∼9,600,000	0	≥ 6
Akiyama, 2023 (13)	Japanese	Lifespan	-	137,693 individuals	-	6,108,833	≥ 1	-
Rosoff, 2023 (47)	European	Aging	-	∼1,900,000 individuals	-	6,793,898	52	-
Park, 2024 (14)	European	Lifespan	-	393,833 individuals	-	6,127,227	≥ 2	-

Fig. 1 illustrates the non-linear relationship between sample size (log₁₀-transformed) and the number of genome-wide significant variants across 45 GWASs of aging- and longevity-related traits. The association remains weak at smaller sample sizes (log₁₀ n < 5) but rises sharply beyond approximately 200,000-300,000 participants (log₁₀ n ≈ 5.3-5.5), indicating a threshold effect in the detection of polygenic signals. This suggests that only large-scale GWASs have sufficient power to reveal the diffuse genetic architecture underlying lifespan variation.

Fig. 1.

Genome-wide discovery scaling across 45 GWASs of aging and longevity. Sample size (log₁₀-transformed, x-axis) is plotted against the number of genome-wide significant variants (y-axis) across 45 GWASs representing six phenotypes: aging, longevity, parental longevity, lifespan, parental lifespan, and healthspan. Point size indicates the –log₁₀ p-value and color the effect size (β) of the most significant variant. Shapes denote traits (circle = aging, triangle = longevity, inverted triangle = parental longevity, square = lifespan, open square = parental lifespan, diamond = healthspan). The red line represents a fitted trend with 95% confidence band.

Each point represents a single GWAS, with point color and size denoting the effect size (β) and significance (–log₁₀ p) of the lead variant, respectively. Early studies (log₁₀ n ≈ 3-4) identified a handful of large-effect loci such as APOE, TOMM40, and FOXO3 (9, 10, 24), whereas recent meta-analyses using resources like the UK Biobank (17, 19, 20, 56) uncovered numerous small-effect variants dispersed throughout the genome. The curvature of the regression line captures this shift—from sparse, candidate-gene discoveries to the exponential rise in detected loci once the polygenic signal surpasses the necessary threshold.

This pattern likely arises from the combined influence of statistical power, phenotype definition, cohort-specific factors, and the inherent phenotyping uncertainty of traits such as longevity and aging. These phenotypes are often indirectly defined (e.g., survival status, parental lifespan, or retrospective age cutoffs), leading to noise and effect-size attenuation. Consequently, robust discovery requires either vastly increased sample sizes to overcome phenotype imprecision or refined, quantitatively defined phenotyping strategies that capture biological aging more precisely. This interplay between statistical power and phenotypic accuracy underscores that progress in aging genomics depends not only on scaling datasets but also on redefining the traits themselves toward measurable, reproducible indicators of aging biology.

Functional characterization of aging and longevity genes

As shown in Fig. 2A, multiple GWAS conducted over the past 15 years have repeatedly identified specific genetic loci associated with aging and longevity. Among them, APOE represents the most consistently replicated signal, reported in 22 independent studies, underscoring its pivotal role in human lifespan variation (58) and Alzheimer’s disease as well (59). LPA (12 replications) and CHRNA5/CHRNA3 (8 each) follow, reflecting the influence of lipid transport and smoking-related behavioral pathways on survival (60). The repeated detection of ATXN2, SH2B3, HLA-DQA1, FURIN, and CDKN2B-AS1 highlights the importance of immune regulation, vascular remodeling, and cellular signaling in longevity. The reproducibility of these loci across ancestries and phenotype definitions supports the existence of a core genetic framework governing human lifespan.

Fig. 2.

Genetic and functional architecture of aging and longevity. (A) Replication frequency of reported genes across 33 genome-wide association studies (GWASs) for aging and longevity. Bars represent the top 15 replicated genes, ranked by the number of independent studies in which they were reported. (B) Gene-level intersections among six phenotypes (aging, longevity, parental longevity, lifespan, parental lifespan, and healthspan) visualized using an UpSet plot. Bars show the number of shared or phenotype-specific genes. (C) Gene–disease interaction network based on DisGeNET annotations. Nodes represent genes and disease or trait terms; node size indicates the number of associated genes. (D) Gene set enrichment analysis (GSEA) of the union set of genes reported in GWASs. Enriched GO terms (adjusted P < 0.05, ≥ 2 genes) are clustered by functional similarity.

Complementing these findings, Fig. 2B depicts gene-level intersections among six GWAS phenotypes—aging, longevity, parental longevity, lifespan, parental lifespan, and healthspan—using an Upset plot. Longevity (70 genes) and parental lifespan (54 genes) exhibited the largest numbers of significant loci, reflecting the statistical power of large-scale cohort analyses. Lifespan (45 genes) also contributed substantially, whereas aging (29 genes), parental longevity (8 genes), and healthspan (7 genes) were represented by smaller sets. The largest overlap occurred between longevity and parental lifespan (65 genes), followed by lifespan–parental lifespan (44 genes) and longevity–lifespan (34 genes). These intersections indicate that diverse lifespan phenotypes share common biological pathways—particularly those involving metabolism, inflammation, and cellular homeostasis. Notably, the overlapping genes include APOE, LPA, LDLR, EPHX2 (lipid metabolism), FOXO3, IGF1R, INSR (insulin/IGF signaling), and IL6R, HLA-DQA1, HLA-DRB1 (immune regulation).

A smaller intersection of 13 genes shared among aging–longevity–parental lifespan likely represents intrinsic biological aging processes independent of disease mediation, encompassing WRN, SIRT6, and CDKN2A/B, which are involved in DNA repair, stress resistance, and anti-senescence mechanisms. Only a handful of genes (1-2) were phenotype-specific, indicating that most lifespan-associated genes exhibit pleiotropic effects across multiple phenotypes. Together, the replication patterns in Fig. 2A and the intersection structures in Fig. 2B underscore a shared polygenic architecture underpinning human aging, with longevity, lifespan, and healthspan representing distinct manifestations along the same genetic continuum.

Fig. 2C, based on a DisGeNet-derived interaction network, visualizes how these genes cluster around two major biological domains. The first comprises lipid metabolism and cardiovascular risk factors (e.g., serum cholesterol and LDL cholesterol measurements), which serve as key determinants of metabolic homeostasis and healthy aging (27, 61). APOE and LPA likely modulate cholesterol balance, thereby buffering against metabolic and cardiovascular diseases or delaying their onset. The second cluster involves environmental and behavioral risk factors, notably smoking (62, 63), linked through CHRNA5 and CHRNA3, which mediate nicotine response via neuronal receptor signaling and the brain’s reward circuitry (60, 64). These interactions suggest that genetic and environmental factors converge through shared neurobehavioral pathways influencing lifespan.

Finally, Fig. 2D presents the results of gene set enrichment analysis (GSEA), revealing 111 significantly enriched terms (adjusted P < 0.05, ≥ 2 genes per term). These pathways converge on four major biological axes:

1.Neurotransmission and neurodegeneration (acetylcholine-/glutamate-gated signaling), linking synaptic integrity to aging and highlighting the vulnerability of cholinergic and glutamatergic systems in neurodegenerative diseases such as Alzheimer’s and Parkinson’s (64, 65).

2.Immunosenescence and inflammaging, involving IL6, CD4＋, and CD8＋ T-cell pathways that drive age-related loss of immune homeostasis and chronic inflammation (66, 67).

3.Metabolic and vascular aging, including lipoprotein metabolism and vasoconstriction pathways governed by APOE and LPA, central to cardiovascular resilience (58, 68).

4.DNA damage repair and genome maintenance, highlighting apoptotic DNA processing as a mechanism linking genomic integrity to longevity (34).

In summary, aging and longevity genes organize along two major biological axes:

(1)a metabolic–vascular axis, exemplified by APOE/LPA and lipid metabolism pathways; and

(2)a neuro-immune axis, involving CHRNA family and neurotransmitter signaling, which mediate resilience through neural and immune homeostasis. Together, these converging pathways form the shared polygenic foundation that shapes both lifespan and healthspan.

Genetic correlations and shared functional architecture across aging-related phenotypes

Across six aging-related phenotypes, gene-level overlap was generally low, with most loci exhibiting phenotype specificity. The largest intersection—between aging and parental lifespan—included only 13 shared genes, while the majority were unique to each phenotype (longevity 65/70, lifespan 44/45, parental lifespan 34/54). This indicates a high degree of genetic heterogeneity in how aging and longevity are defined and mapped at the genomic level. Nevertheless, despite this heterogeneity, we observed extensive functional convergence across phenotypes, revealing common biological axes underpinning aging and lifespan regulation.

Fig. 3 presents the results of gene set enrichment analysis (GSEA) conducted for each phenotype, clustering 1,018 enriched GO terms to visualize shared and distinct functional domains. Five major clusters were identified, reflecting the interplay of immune, metabolic, neural, and cellular maintenance processes in aging and longevity.

Fig. 3.

Functional overlap of aging- and longevity-associated genes across six traits. Network visualization of enriched biological processes (GO:BP, nominal P < 0.01) identified in GWASs for six traits: aging, longevity, parental longevity, lifespan, parental lifespan, and healthspan. Nodes represent GO biological process terms, colored by trait; node size reflects enrichment significance (−log10 p). Edges indicate shared genes between GO terms, showing the degree of functional overlap among traits.

Cluster 1 – Immune response and immuno-metabolic regulation (shared by aging and parental lifespan, 83 GO terms)

This cluster includes T cell differentiation involved in immune response, antigen processing and presentation, triglyceride metabolic process, and acetylcholine receptor signaling. Together, these pathways indicate that aging and longevity are orchestrated by cross-system regulation among the immune, metabolic, and neuronal networks. Impaired T-cell differentiation and antigen presentation promote immunosenescence (69) and chronic inflammation (70), which in turn disrupt lipid homeostasis (67, 71). Conversely, lipid imbalance can exacerbate inflammatory signaling, while chronic immune-metabolic stress may damage cholinergic neurotransmission and accelerate cognitive decline (65, 72). This cluster thus represents a multisystemic aging axis where breakdown of immune–metabolic–neural feedback loops drives systemic aging.

Cluster 2 – Neural homeostasis and neuroprotection (shared by lifespan and parental lifespan, 10 GO terms)

Terms such as dendritic spine maintenance, neurotransmitter-gated ion channel clustering, and regulation of amyloid-beta clearance highlight mechanisms preserving synaptic structure and plasticity (73). Balanced cholesterol metabolism and efficient amyloid-beta clearance protect neurons from metabolic and proteotoxic stress. Conversely, cholesterol dysregulation or amyloid accumulation with age induces synaptic toxicity and neuroinflammation (74, 75). This cluster emphasizes synaptic integrity and metabolic resilience as central defenses against brain aging.

Cluster 3 – T-cell–centered immunometabolic regulation (shared by longevity and parental lifespan, 10 GO terms)

This cluster encompasses T cell lineage commitment, hepatic immune response, and negative regulation of cholesterol/sterol/phosphatidylcholine metabolic process. It captures the immuno-metabolic cross-regulation governing T-cell activation. Regulation of T-cell differentiation and proliferation is essential to suppress chronic inflammation (76), and T-cell expansion depends on lipid and cholesterol supply for membrane synthesis (77). The inhibitory control of cholesterol transport observed here likely represents a metabolic buffering mechanism limiting T-cell overactivation and mitigating inflammaging.

Cluster 4 – Inflammation resolution and apoptotic regulation (shared by healthspan, longevity, and parental lifespan, 6 GO terms)

This was the only cluster involving healthspan, including apoptotic DNA fragmentation, inflammatory cell apoptotic process, neutrophil homeostasis, and positive regulation of myeloid cell apoptosis. These pathways mediate the resolution phase of inflammation by promoting apoptosis of activated neutrophils and myeloid cells, preventing chronic inflammation and maintaining immune equilibrium (78).

Cluster 5 – Lipid and cholesterol metabolism (shared by five phenotypes, 91 GO terms)

This cluster includes cholesterol metabolic process, VLDL particle clearance, lipid transport, sterol metabolic process, and triglyceride-rich lipoprotein clearance. Lipid metabolism is fundamental to membrane integrity, steroid synthesis, and energy balance (26, 27). Efficient clearance of VLDL and triglyceride-rich particles prevents metabolic dysfunction and systemic inflammation (79, 80), supporting whole-body metabolic stability and healthy aging.

Collectively, these findings demonstrate that while genetic signals of aging and lifespan are phenotypically heterogeneous, their functional architectures converge onto shared biological axes—immune, metabolic, neural, and cellular maintenance systems. These conserved axes reflect different manifestations of the same underlying biological continuum of human aging.

Future directions: from polygenic associations to regulatory and dynamic models of aging

Genome-wide association studies (GWASs) have played a central role in delineating the polygenic architecture of aging and longevity. However, their explanatory scope remains conceptually and methodologically limited in clarifying when and through which biological pathways inherited variants exert their effects across the lifelong aging trajectory. The predominance of non-coding signals further indicates that most longevity-associated variants act through gene regulatory mechanisms rather than protein-coding changes (81, 82). Bridging this gap requires a multidimensional shift in aging genomics.

First, greater precision in phenotype definition must be coupled with a broader exploration of the human genetic variation spectrum. Aging is inherently difficult to operationalize at the individual level, and phenotypic imprecision remains a major constraint on discovery power. Future studies should extend beyond common variants to interrogate rare and structural variants enriched in extreme-longevity cohorts. In parallel, overcoming the current Eurocentric bias will require cross-ancestry meta-GWASs and trans-ethnic fine-mapping. Expanding ancestral diversity is not merely an equity issue but a scientific necessity, as demonstrated across multiple disease-genomics studies showing that population diversity enables the disentanglement of shared biological mechanisms from ancestry-specific effects (83, 84). Leveraging both extremity and diversity may represent an efficient strategy for achieving globally generalizable aging models.

Second, integrating epigenomic layers is essential for resolving the regulatory cascade underlying non-coding genetic associations. DNA methylation–based epigenetic clocks have become foundational tools for quantifying biological aging rather than chronological time (85, 86). Methylation quantitative trait locus (mQTL) analyses provide a critical framework for linking genetic variants to DNA methylation changes (87, 88). Importantly, variants that influence the rate of epigenetic drift—the stochastic erosion of epigenetic regulation—may explain additional aging-related phenotypic variance beyond what static GWASs can capture (89).

Third, establishing causality will require functional validation and single-cell–resolution approaches. CRISPR-based perturbation screens offer powerful means to directly test candidate regulatory loci (90), while systems-level dynamical modeling of intracellular molecular processes (91), when combined with single-cell multi-omics profiling, enables precise dissection of cell-type–specific aging mechanisms (92). Integrating genomic, epigenomic, and longitudinal phenotypic data will facilitate modeling aging as a dynamic physiological process, rather than a static accumulation of risk (93).

Fourth, aging emerges from continuous interactions between genetic susceptibility and environmental exposure. Future studies should systematically assess gene–environment (G × E) interactions, as well as sex-specific and age-dependent effects. Understanding how genetic risk manifests across environmental contexts—such as diet, smoking, and socioeconomic conditions—is critical for explaining inter-individual differences in resilience and vulnerability.

Finally, the field must transition toward clinically meaningful prediction through multi-modal integration. Combining digital biomarkers with multi-organ phenotypic data and leveraging AI-driven risk models will allow aging research to move beyond descriptive associations toward personalized, predictive, and actionable strategies for healthy longevity.

DISCUSSION

This review synthesized 45 GWASs of aging and longevity conducted over the past 15 years, encompassing six phenotypes (aging, longevity, lifespan, parental lifespan, parental longevity, and healthspan) and more than one million individuals worldwide. Although the genetic architecture appears heterogeneous across traits and ancestries, the underlying functional themes reveal a remarkably coherent biological structure that connects metabolism, inflammation, neural maintenance, and cellular resilience as the core determinants of human longevity.

Figs. 1-3 collectively summarize this trajectory. Fig. 1 shows a nonlinear transition point in the number of significant variants as sample size increases, consistent with a highly polygenic architecture driven by thousands of small-effect alleles rather than a few major genes. This apparent threshold effect may partly reflect phenotypic noise in longevity-related traits, highlighting that future GWAS progress will rely on both scaling sample sizes and improving phenotype precision. Fig. 2A highlights replicated signals—APOE, LPA, CHRNA5, FOXO3, SH2B3, CDKN2B-AS1—that form a core genetic framework spanning metabolic, immune, and cell-signaling pathways. Fig. 2B demonstrates that despite low gene-level overlap among phenotypes, the shared genes cluster within a limited set of biological axes: lipid metabolism, insulin/IGF signaling, and inflammatory control. Finally, Fig. 3 reveals via gene-set enrichment analysis a higher-order functional convertgence in immune–metabolic–neuronal regulation, inflammation resolution, and DNA repair pathways. Together, these findings illustrate that while aging and longevity genes are scattered across the genome, their functions are deeply entwined within a small number of conserved biological networks.

Before further interpreting these convergent pattern, ancestral diversity and portability represent pressing challenges. Nearly all included studies are Eurocentric; less than five involve East Asian, South Asian, or African populations. Population-specific LD patterns and allele frequencies can alter effect sizes and even directions for key loci such as APOE and FOXO3. Polygenic scores trained on European datasets show sharp declines in predictive accuracy in non-European cohorts (94). Future progress demands cross-ancestry meta-GWASs, trans-ethnic fine-mapping, and multi-cohort integration (e.g., KoGES, BBJ, All of Us) to disentangle shared from population-specific mechanisms and to enhance the global portability of aging models.

Several interpretative themes emerge from this integrative analysis. First, phenotype definition matters. The large number of phenotype-specific genes—65 unique to longevity, 44 to lifespan, and 34 to parental lifespan (Fig. 2B)—underscores that the operational definition of “aging” strongly influences genetic discoveries. Quantitative traits such as parental lifespan show greater statistical power than binary long-lived vs control designs (17, 95) (Fig. 1), highlighting the need for precisely modeled phenotypes tailored to specific biological questions.

Second, healthspan remains an underexplored frontier. Despite its growing clinical relevance, only two healthspan GWASs exist (19, 20), yet they exhibit distinct functional signatures related to inflammation resolution, apoptotic control, and genome stability (Fig. 3, Cluster 4). These findings shift the focus from mere survival to the preservation of physiological function, emphasizing that healthy longevity is a product not only of disease avoidance but also of sustained cellular integrity.

Third, methodological limitations must be recognized. Survival and longevity analyses are susceptible to competing risks (96), cohort effects (97), and survivor bias (98); biobank-based analyses face participation bias (99); binary phenotypes risk misclassification; and age- or sex-dependent effects, as well as gene-environment interactions (e.g., smoking, diet, socioeconomic status), remain poorly captured. Rigorous phenotype harmonization, standardized QC pipelines, and cross-study replication are needed to improve the robustness of future analyses.

Fourth, the present synthesis supports a dual-pathway model of aging genetics. One axis comprises disease-mediated pathways (17)—lipid metabolism, inflammation, and behavioral risk factors—represented by genes such as APOE, LPA, IL6R, and CHRNA5. The other axis reflects intrinsic resilience pathways involving cellular stress response, DNA repair, and synaptic maintenance (2)—typified by FOXO3, SIRT6, WRN, and CDKN2A/B. Together these pathways define human longevity as an interplay between disease avoidance and homeostatic maintenance.

In summary, the genetic architecture of human aging is best understood as a highly polygenic but functionally constrained system, in which myriad small-effect variants fine-tune a limited set of evolutionarily conserved networks that regulate metabolism, immune balance, neural resilience, and cellular maintenance. Expanding ancestral representation, harmonizing phenotypes, and bridging genomic discovery to functional mechanisms will be essential steps toward a globally inclusive and mechanistically grounded framework for precision aging and healthy longevity.