Life-course influence of birthweight and subsequent pathways on healthy aging: a Mendelian randomization study

Abstract

Background

Birthweight readily measurable marker of fetal growth that may influence health across the lifespan. We aimed to investigate the potential causal association between birthweight and healthy aging and to identify the mediating roles of subsequent socioeconomic, behavioral, functional, and disease-related factors to inform life-course strategies to promote healthy aging and reduce health inequities.

Methods

We performed two-sample Mendelian randomization analyses in European-ancestry participants to estimate the effect of birthweight (n = 298,142–423,683) on two robust, composite healthy aging phenotypes (genetically independent phenotype of aging (aging-GIP) and multivariate aging-related genetic factor (mvAge)) and six individual aging phenotypes, including healthspan, resilience, parental lifespan, self-rated health, phenotypic age deceleration, and 90^th percentile self-longevity (n = 34,710–1,958,774), and screened for 100 candidate mediators (n = 14,267–1,812,017) using a two-step mediation analysis.

Results

Genetically determined each 1-SD higher birthweight was associated with higher aging-GIP (β [95% CI] in different models ranging from 0.131 [0.066–0.196] to 0.162 [0.089–0.235] SDs) and mvAge (0.036 [0.010–0.063] to 0.045 [0.024–0.067]), independent of later-life obesity indicators; also with more interpretable benefits, including 12%–16% higher odds of longer healthspan, a 0.079–0.089 SD improvement in resilience, and a 1.22–1.74 year increase in parental lifespan. Of 100 candidates, 26 and 25 mediated the effect of birthweight on aging-GIP and mvAge, respectively, including socioeconomic indicators (education, household income, occupational attainment; individual mediation proportion: 12.72%–27.79%); behaviors (e.g., cheese intake, age at first sex; 10.38%–29.56%); physical functions (e.g., blood pressure, grip strength; 7.57%–42.65%); and cardiometabolic diseases (e.g., type 2 diabetes, cardiovascular diseases; 25.02%–70.11%).

Conclusions

Higher birthweight within the normal range directly promotes healthy aging, mediated by multifaceted modifiable factors. Our findings advocate adopting a life-course approach to foster healthy aging, starting with optimal birthweight and extending to interventions that enhance socioeconomic status, promote healthy behaviors, strengthen physical functions, and prevent cardiometabolic diseases.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-026-04737-8.

Background

Promoting healthy aging necessitates prevention strategies from a lifecycle lens [1], as the widening gap between lifespan and healthspan creates both socioeconomic strains and a mounting burden of age-related diseases [1–4]. Birthweight, a readily measurable marker of fetal growth, reflects multifaceted public health factors such as maternal nutrition and prenatal healthcare, of great public health significance [5–7]. Compelling observational and genetic evidence has associated lower birthweight with accelerated aging and higher risks of type 2 diabetes, cardiovascular disease (CVD), and mortality [8–12]. However, meticulous evidence on the life-course impact of birthweight on healthy aging and how birthweight influences subsequent socioeconomic, behavioral, and functional factors and diseases across the life-course is limited.

Healthy aging is inherently multidimensional and cannot be adequately captured by a single clinical endpoint [13–16]. To address this complexity, recent genome-wide association studies (GWAS) have developed composite healthy aging metrics, including the genetically independent phenotype of aging (aging-GIP) and the multivariate aging-related genetic factor (mvAge) [13, 15]. By integrating common genetic markers across different aging domains, these phenotypes capture the overarching genetic susceptibility to multidimensional aging, reflecting both longevity and quality of life [13, 15], and facilitate research linking birthweight and healthy aging.

Several uncertainties remain to be elucidated. First, obesity is intricately intertwined with birthweight and aging trajectories [17, 18], but whether birthweight exerts an impact on healthy aging independent of later-life obesity is unknown. Second, although emerging evidence links birthweight with later-life socioeconomic status and behaviors [9, 19], the multifaceted mediating pathways in the effect of birthweight on healthy aging are unexplored, which is imperative for informing life-course health management. Thus far, conventional cohort studies face challenges in mapping the holistic effect chain of birthweight on healthy aging outcomes due to the need for relatively long follow-up periods, sufficient sample size, and comprehensive phenotypic data for extensive evaluation, which limits the depth and breadth of evidence available. Alternatively, the Mendelian randomization (MR) approach offers a solution by using genetic variants that are naturally randomly assigned at conception as instrumental variables (IVs) to establish causality, which can reduce susceptibility to confounding and reverse causality [20], especially when the exposure of interest (e.g., birthweight) is not practically or ethically amenable to randomization.

In this two-step MR study, we systematically evaluated the life-course influence of birthweight on healthy aging. Specifically, we (i) examined the total effects of genetically predicted birthweight on two robust, composite healthy aging phenotypes (aging-GIP and mvAge) and six individual aging-related phenotypes; (ii) assessed whether these associations were independent of later-life obesity; (iii) elucidated potential mediating pathways by screening 100 candidate mediators spanning socioeconomic, behavioral, physical, and disease domains; (iv) explored the origins of these associations in an exploratory analysis utilizing fetal- and maternal-specific birthweight genetic instruments; and (v) evaluated the robustness of findings through replication in independent GWAS datasets.

Methods

Study design

This study consisted of two analytic phases (Fig. 1, Additional file 1: Figs. S1–3). In Phase 1, we assessed the associations between genetically predicted birthweight and healthy aging phenotypes using univariable MR (UVMR) and clarified whether the effects were independent of childhood body mass index (BMI), adulthood BMI, or adulthood waist circumference (WC) using multivariable MR (MVMR). In Phase 2, we screened for 100 candidate mediators in the associations of birthweight with aging-GIP and mvAge, and quantified individual mediating proportions using the two-step MR. Our reporting strictly adhered to the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization (STROBE-MR) guidelines (Additional file 2: Table S1) [21]. This study was conducted using publicly accessible, summary-level GWAS data from European-descent individuals. Ethical approval and informed consent can be found in the corresponding GWAS publications referenced in this work [13–16, 22–38].

Fig. 1.

Overview of the study design. This MR study consisted of two analytic phases. In Phase 1, we assessed the associations between genetically predicted birthweight and healthy aging phenotypes using UVMR and clarified whether the effects were independent of childhood BMI, adulthood BMI, or adulthood WC using MVMR. In Phase 2, we screened for 100 candidate mediators including socioeconomic indicators, behavioral factors, physical function-related traits, and diseases in the associations of birthweight with aging-GIP and mvAge, and quantified individual mediating proportions using the two-step MR. The detailed technical pipeline and data usage are further elaborated in Figs. S1–2. Abbreviations: aging-GIP, genetically independent phenotype of aging; BMI, body mass index; EGG, Early Growth Genetics; GIANT, Genetic Investigation of ANthropometric Traits; IV, instrumental variables; IVW, inverse-variance weighted; MR, Mendelian randomization; MR-PRESSO, MR pleiotropy residual sum and outlier; mvAge, multivariate aging-related genetic factor; MV-IVW, multivariable inverse-variance weighted; MVMR, multivariable Mendelian randomization; PhenoAgeDecel, phenotypic age deceleration; UVMR, univariable Mendelian randomization; WC, waist circumference

Data sources and genetic instruments

Genetic instruments were selected for birthweight (n = 298,142–423,683); covariates (childhood BMI, adulthood BMI, and adulthood WC, n = 39,620–681,275); and 100 candidate mediators including three socioeconomic indicators, 25 behavioral factors, 33 physical function-related traits, and 39 common diseases (n = 14,267–1,812,017). The outcome included two robust, composite metrics of healthy aging (aging-GIP, n = 154,478; mvAge, n = 1,958,774) and six individual aging phenotypes (healthspan, resilience, self-rated health, parental lifespan, PhenoAge deceleration, and 90^th percentile self-longevity phenotype; n = 34,710–1,012,240). The aging-GIP was established on the basis of GWASs of typical aging phenotypes (i.e., the first five individual aging phenotypes) and could capture the shared genetic architecture across core aspects of aging, consolidate diverse genetic signals, and enhance the identification of biological mechanisms underlying aging processes [13]. The mvAge was constructed by integrating GWASs of multidimensional aging phenotypes (the last five individual aging phenotypes) using genomic structural equation modeling to derive a common genetic factor and could capture broader biological pathways underlying healthy aging [15]. Detailed information on all GWAS datasets, phenotype definitions, instrument selection, and candidate mediator selection is provided in Table 1, Additional file 2: Tables S2–3 [39–75], and Additional file 3: Methods S1–2.

Table 1.

Overview of the GWAS data used in the study

Phenotype	Sample size or case/control	Ancestry	Unit	Consortium or cohort study	PMID
Exposure (main analysis)
Birthweight	298,142	European	1-SD (about 484 g)	EGG	31043758
Exposure (exploratory analysis)
Birthweight: separate fetal- and maternal-specific genetic effects	406,063	European	1-SD (about 484 g)	EGG	31043758
Exposure (replication analysis)
Birthweight	423,683	European	1-SD (about 492 g)	Icelandic birth register, EGG, and UK Biobank	34282336
Birthweight: separate fetal- and maternal-specific genetic effects	104,920	European
Covariate
Childhood BMI	39,620	European	1-SD	EGG	33045005
BMI	681,275	European	1-SD	GIANT	30124842
WC	232,101	European	1-SD	GIANT	25673412
Outcome
Composite healthy aging phenotype
Aging-GIP	154,478	European	1-SD	Meta	37118362
mvAge	1,958,774	European	/	Meta	37550455
Individual aging phenotype
Healthspan	300,447	European	Log-transformed hazard ratio	UK Biobank	30729179
Resilience (the inverse of the frailty)	175,226	European	1-SD	UK Biobank, TwinGene	34431594
Parental lifespan	1,012,240	European	10 years	UK Biobank, LifeGen	30642433
Self-rated health	359,681	European	1-point in the score	UK Biobank	27864402
PhenoAge deceleration	34,710	European	1-year	Meta	34187551
90th percentile self-longevity	11,262/25,483	European	log-odds	Meta	31413261
Qualified mediator (n = 26)a
Socioeconomic indicator
Education	766,345	European	1-SD	SSGAC	30038396
Household income	397,751	European	1-SD	UK Biobank	29846171
Occupational attainment	248,847	European	1-point in skill level	UK Biobank	34613391
Behavioral factor
Leisure screen time	Up to 526,725	European	1-SD	Meta	36071172
Cheese intake	451,486	European	1-SD	UK Biobank	29846171
Age at first sex	397,338	European	Year	Meta	34211149
Age at first live birth	542,901	European	Year	Meta	34211149
Physical function
Grip strength	461,026	European	1-SD	UK Biobank	29846171
SBP	757,601	European	mmHg	Meta	30224653
Fasting glucose	200,622	European	mmol/L	MAGIC	34059833
2hGlu	63,396	European	log-transformed mmol/L	MAGIC	34059833
Total cholesterol	1,320,016	European	1-SD	Meta	34887591
Triglycerides	1,320,016	European	1-SD	Meta	34887591
LDL-C	1,320,016	European	1-SD	Meta	34887591
nonHDL-C	1,320,016	European	1-SD	Meta	34887591
ApoB	24,925	European	1-SD	Meta	27005778
UACR	547,361	European	1-SD	CKDGen	31511532
ALT	752,839	European	1-SD	Meta	33972514
SHBG	370,125	European	1-SD	UK Biobank	32042192
Disease
Type 2 diabetes	242,283/1,569,734	European	log-odds	DIAGRAM	38374256
Hypertension	129,909/354,689	European	log-odds	UK Biobank	33959723
CHD	60,801/123,504	77% European	log-odds	CARDIoGRAMplusC4D	26343387
CAD	122,733/424,528	European	log-odds	Meta	29212778
MI	61,000/577,000	European	log-odds	Meta	33532862
Coronary atherosclerosis	23,363/187,840	European	log-odds	FinnGen	36653562
Angina pectoris	18,168/187,840	European	log-odds	FinnGen	36653562

Abbreviations: 2hGlu 2 h-glucose post-challenge, Aging-GIP genetically independent phenotype of aging, ALT alanine aminotransferase, ApoB apolipoprotein B, BMI body mass index, CAD coronary artery disease, CARDIoGRAMplusC4D Coronary Artery Disease Genome Wide Replication and Meta-analysis plus the Coronary Artery Disease Genetics, CHD coronary heart disease, CKDGen Chronic Kidney Disease Genetics Consortium, DIAGRAM DIAbetes Genetics Replication And Meta-analysis, EGG Early Growth Genetics, GIANT Genetic Investigation of ANthropometric Traits, GWAS genome-wide association study, LDL-C low-density lipoprotein cholesterol, MAGIC Meta-Analysis of Glucose and Insulin-related Traits Consortium, MI myocardial infarction, mvAge multivariate aging-related genetic factor, nonHDL-C non-high-density lipoprotein cholesterol, PhenoAge phenotypic age, SBP systolic blood pressure, SD standard deviation, SHBG sex hormone-binding globulin, SSGAC Social Science Genetic Association Consortium, UACR urine albumin-to-creatinine ratio, WC waist circumference

^aOnly the qualified mediators are shown in Table 1. Information on the excluded candidates can be found in Table S2

Statistical analysis

UVMR and MVMR analyses

We performed UVMR to assess the total effects of genetically predicted birthweight on healthy aging phenotypes. MVMR was further applied to estimate the direct effect of birthweight after adjusting for childhood BMI, adulthood BMI, or adulthood WC, determining whether birthweight impacts healthy aging independently of later-life obesity. We employed the inverse-variance weighted (IVW) method for maximum efficiency [76], as the primary analysis in UVMR and MVMR. The IVW method produces the most precise, unbiased, and efficient causal estimates, given that the IVs meet the MR assumptions [77]. The random-effect IVW method adjusts for heterogeneity of IVs, and the multivariable inverse-variance weighted (MV-IVW) method accounts for correlations between multiple exposures [78, 79]. The causal estimates were provided as odds ratios (ORs) with 95% confidence intervals (CIs) for binary outcomes and β coefficients with 95% CIs for continuous outcomes.

Exploratory and replication analyses

To investigate the origins of the association between birthweight and healthy aging, we conducted exploratory UVMR analyses using fetal-only, fetal-effect (adjusted for maternal genotype), maternal-only, and maternal-effect (adjusted for fetal genotype) genetic instruments for birthweight (Additional file 3: Method S1).

To ensure the robustness of our findings, we then performed a replication analysis, repeating the primary and exploratory analyses using an independent set of genetic instruments from a large birthweight GWAS of 423,683 European-ancestry individuals (Additional file 3: Method S1) [23, 24].

Mediation MR analysis

To identify potential mediating pathways through which birthweight influences healthy aging, we employed two-step MR analyses focusing on aging-GIP and mvAge. These analyses were hypothesis-driven and aimed to conduct exploratory research. Each mediator was analyzed separately in a one-by-one framework. The first step estimated the causal effect (β1) of birthweight on each mediator using UVMR. The second step assessed the causal effect (β2) of each mediator on each healthy aging outcome with adjustment for birthweight using MVMR, contingent on the mediator being causally linked to the outcome in UVMR. When evidence indicated that birthweight influenced the mediator, which in turn affected the outcome, we calculated the mediation proportion by dividing the mediation effect (β1 × β2) by the total effect of birthweight on the outcome. The 95% CIs for the mediation proportions were determined using the Delta method [80], with negative mediation proportions set at a minimum threshold of 0%. The mediation proportions cannot be summed, because they are not mutually exclusive and may be influenced by correlations among mediators.

MR sensitivity analyses

For UVMR analyses, we performed the weighted median, simple mode, weighted mode, MR-Egger, and MR pleiotropy residual sum and outlier (MR-PRESSO) methods as sensitivity analyses to evaluate the robustness of the IVW estimates under different assumptions (Additional file 3:Method S3) [76, 79, 81, 82]. Where there was sample overlap between exposure and outcome datasets, we assessed the type 1 error rates based on the calculator at https://sb452.shinyapps.io/overlap/ [83], and conducted the MRlap analysis to validate the causal estimates, as the MRlap is robust to biases introduced by sample overlap, winner’s curse, and weak instruments [84]. For MVMR analyses, we employed the multivariable MR-Egger (MVMR-Egger), MVMR-median, and multivariable MR-Lasso (MVMR-Lasso) methods to validate the robustness of the MV-IVW results (Additional file 3: Method S3) [85]. We calculated the F-statistics to evaluate the validity of the IVs, Cochran’s Q statistic to assess the heterogeneity, and the intercept of MR Egger to test for the pleiotropy of the IVW estimates. The Benjamini–Hochberg method was used for false discovery rate (FDR) correction to account for multiple testing.

All MR analyses were conducted with the R packages TwoSampleMR, MendelianRandomization, MRPRESSO, MVMR, and MRlap in R software (version 4.3.1; R Foundation for Statistical Computing, Vienna, Austria). All statistical tests were two-sided. A P value < 0.05 indicates statistical significance. The IVW estimates with P < 0.05 and FDR-adjusted P value < 0.05 and supported by at least one sensitivity analysis were considered causal evidence, whereas the IVW estimates with P < 0.05 but FDR-adjusted P value ≥ 0.05 or not supported by sensitivity analysis were considered suggestive causal evidence.

Results

Total effect of birthweight on healthy aging phenotypes

In UVMR, genetically determined each 1-standard deviation (SD) higher birthweight was associated with higher aging-GIP (IVW-estimated β, 0.086 SDs; 95% CI, 0.027 to 0.145) and mvAge (0.034; 0.013 to 0.054) after FDR corrections for multiple comparisons (Fig. 2a, Additional file 2: Table S4). Consistently, positive associations were observed between each 1-SD higher birthweight and better resilience (β, 0.055 SDs; 95% CI, 0.012 to 0.098), higher odds of healthspan (odds ratio (OR), 1.12; 95% CI, 1.04 to 1.21), and longer parental lifespan (β, 0.846 years; 95% CI, 0.281 to 1.421). These associations, except for parental lifespan, were supported by two UVMR sensitivity analyses (Additional file 2: Table S4). In exploratory analyses, genetically determined fetal-effect (or fetal-only) birthweight was positively associated with aging-GIP, mvAge, and healthspan, whereas maternal-effect (or maternal-only) birthweight showed no significant associations with healthy aging phenotypes after FDR correction (Additional file 2: Table S4). There was potential heterogeneity between IVs, but the instrumental validity test suggested sufficient instrument strength (F-statistics ≥ 24), and no horizontal pleiotropy was detected (all P_intercept ≥ 0.05; Additional file 2: Table S5). The type 1 error rates due to partial sample overlap between birthweight and healthy aging phenotypes were well controlled under 0.05 (Additional file 2: Table S6), and the associations were well-validated by the MRlap method (Additional file 2: Table S7).

Fig. 2.

UVMR and MVMR estimates for the causal, independent effects of birthweight on healthy aging. a UVMR establishes the associations between genetically predicted birthweight and healthy aging phenotypes. The effect sizes for each healthy aging phenotype were associated with the genetically predicted each 1-SD higher birthweight. MR estimates were derived from the IVW method in UVMR. All significant associations persisted with FDR adjustment for multiple comparisons (Table S4). b MVMR tests if the identified significant associations are independent of later-life obesity. The effect sizes for each healthy aging phenotype were associated with genetically predicted each 1-SD higher birthweight, with adjustment for childhood BMI, adulthood BMI, or adulthood WC. MR estimates were derived from the MV-IVW method in MVMR (Table S10). The squares represent β coefficients or ORs, and the error bars represent 95% CIs. Abbreviations: aging-GIP, genetically independent phenotype of aging; BMI, body mass index; CI, confidence interval; FDR, false discovery rate; MR, Mendelian randomization; MV-IVW, multivariable inverse variance weighted; mvAge, multivariate aging-related genetic factor; MVMR, multivariable Mendelian randomization; OR, odds ratio; PhenoAge, phenotypic age; SD, standard deviation; UVMR, univariable Mendelian randomization; WC, waist circumference

Replication analysis confirmed the primary findings, showing that genetically determined birthweight was positively associated with aging‑GIP and mvAge, as well as healthspan, resilience, and parental lifespan (Additional file 2: Table S8, Table S9). Similarly, as for exploratory analyses, associations with better healthy aging outcomes were observed for fetal-effect birthweight but not for maternal-effect birthweight, suggesting the association might be predominantly driven by fetal genetic effects. These results were robustly supported by at least one sensitivity analysis, with no evidence of weak instruments (F-statistics ≥ 41) or horizontal pleiotropy (all P_intercept ≥ 0.05).

Independent effect of birthweight on healthy aging phenotypes adjusting for later-life obesity indicators

To assess whether genetically predicted birthweight influences healthy aging independently of later-life obesity, we performed MVMR analyses adjusting in turn for childhood BMI, adulthood BMI, and adulthood WC. The positive associations of genetically predicted birthweight with aging-GIP and mvAge, as well as healthspan, resilience, and parental lifespan, persisted with adjustment for childhood BMI, adulthood BMI, and adulthood WC, respectively (Fig. 2b). These MV-IVW estimates were well-validated by at least one MVMR sensitivity analysis (Additional file 2: Table S10), and replication analysis yielded highly consistent results (Additional file 2: Table S11), further corroborating the effect of genetically predicted birthweight on healthy aging independent of later-life obesity.

Two-step MR identifying qualified mediators between birthweight and healthy aging phenotypes

In total, 26 and 25 candidates met all screening criteria and were qualified as mediators between birthweight and aging-GIP and mvAge, respectively (Fig. 3). From birthweight to mediators, strong UVMR evidence supported associations between genetically predicted birthweight and these mediators (FDR-adjusted P < 0.05 and confirmed by at least one UVMR sensitivity analysis; Fig. 4, Additional file 2: Tables S12–S13). The type 1 error rates due to partial sample overlap were well controlled under 0.05 (Additional file 2: Table S6), and the associations were validated by the MRlap method (Additional file 2: Table S7).

Fig. 3.

Selection process for mediators in the associations between birthweight and composite healthy aging phenotypes. Four criteria were applied to screen for the mediators: (1) the mediator should be causally associated with the healthy aging outcome; (2) birthweight should be causally associated with the mediator; (3) the mediator should have a direct causal effect on the healthy aging outcome independently of birthweight; and (4) the total effect of birthweight on the healthy aging outcome and the mediating effect of the mediator should be in the same direction. In total, 26 and 25 mediators qualified for all criteria for aging-GIP and mvAge, respectively. Abbreviations: 2hGlu, 2h-glucose post-challenge; ADHD, attention deficit hyperactivity disorder; aging-GIP, genetically independent phenotype of aging; ALP, alkaline phosphatase; ALT, alanine aminotransferase; ApoA-I, apolipoprotein A-I; ApoB, apolipoprotein B; CAD, coronary artery disease; CHD, coronary heart disease; COPD, chronic obstructive pulmonary disease; CRP, C-reactive protein; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; GGT, gammaglutamyl transferase; HbA1c, glycated hemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; IFC, insulin fold change; LDL-C, low-density lipoprotein cholesterol; MI, myocardial infarction; mvAge, multivariate aging-related genetic factor; MVPA, moderate to vigorous physical activity; nonHDL-C, non-high-density lipoprotein cholesterol; SBP, systolic blood pressure; SES, socioeconomic status; SHBG, sex hormone-binding globulin; UACR, urine albumin-to-creatinine ratio

Fig. 4.

Mediating role of each mediator in the associations between birthweight and composite healthy aging phenotypes. a–b Two-step MR was used to evaluate the mediating role of each mediator in the associations of birthweight with aging-GIP (a) and mvAge (b). Left: UVMR estimates for the effect of genetically predicted birthweight on each mediator (β1). Middle: MVMR estimates for the effect of each mediator on each healthy aging outcome with adjustment for birthweight (β2). Right: the mediation proportion of each mediator in the association between birthweight and each healthy aging outcome. MR estimates were derived from the IVW method in UVMR and the MV-IVW method in MVMR. The 95% CIs for the mediation proportions were determined using the Delta method, with negative mediation proportions set at a minimum threshold of 0%. The data are presented as β coefficients or proportions, with corresponding 95% CIs. The squares represent β coefficients, and the bars represent proportions, with the error bars indicating 95% CIs. All statistical tests were two-sided. P < 0.05 was considered significant. Abbreviations: 2hGlu, 2h-glucose post-challenge; aging-GIP, genetically independent phenotype of aging; ALT, alanine aminotransferase; ApoB, apolipoprotein B; CAD, coronary artery disease; CHD, coronary heart disease; CI, confidence interval; IVW, inverse variance weighted; LDL-C, low-density lipoprotein cholesterol; MI, myocardial infarction; MR, Mendelian randomization; mvAge, multivariate agingrelated genetic factor; MV-IVW, multivariable inverse variance weighted; MVMR, multivariable Mendelian randomization; nonHDL-C, non-high-density lipoprotein cholesterol; SBP, systolic blood pressure; SHBG, sex hormone-binding globulin; UACR, urine albumin-to-creatinine ratio; UVMR, univariable Mendelian randomization

From mediators to healthy aging phenotypes, strong UVMR evidence supported associations of the 26 qualified mediators with aging-GIP and the 25 mediators with mvAge (FDR-adjusted P < 0.05 and confirmed by at least one UVMR sensitivity analysis; Additional file 2: Table S14, Table S15), which were also validated by the MRlap method (Additional file 2: Table S7). MVMR further supported direct effects of these mediators on healthy aging phenotypes independent of birthweight (confirmed by at least one MVMR sensitivity analysis; Additional file 2: Table S16).

Mediating effects of mediators in the associations between birthweight and healthy aging phenotypes

Among physical functions and diseases, type 2 diabetes showed the highest individual mediation proportion (70.11%) for the effect of birthweight on aging-GIP, followed by myocardial infarction ((MI), 57.78%); coronary artery disease ((CAD), 47.48%); hypertension (45.30%); coronary heart disease ((CHD), 44.77%); systolic blood pressure ((SBP), 42.65%); grip strength (36.72%); coronary atherosclerosis (29.02%); angina pectoris (25.02%); 2 h-glucose post-challenge ((2hGlu), 24.94%); triglycerides (24.18%); apolipoprotein B ((ApoB), 23.41%); and non-high-density lipoprotein cholesterol ((nonHDL-C), 20.35%), etc. (Fig. 4a). Behavioral mediators between birthweight and aging-GIP ranked by individual mediation proportion included cheese intake (29.56%), age at first sex (25.36%), age at first live birth (23.69%), and leisure screen time (17.35%). Socioeconomic mediators, including educational attainment, household income, and occupational attainment, each mediated 20.90% to 27.79% of the total effect.

For mvAge, the mediator profiles and rankings were largely similar to those observed for aging‑GIP (Fig. 4b). Type 2 diabetes remained the strongest mediator (mediation proportion: 55.17%), followed by CVDs (28.41% to 46.03%) and physical functions (7.57% to 27.12%). Behavioral and socioeconomic factors retained similar rankings, though with slightly attenuated proportions.

Discussion

This study elucidated the effects of birthweight on healthy aging phenotypes and the mediating pathways involved. We found that genetically determined higher birthweight within the normal range was associated with a higher likelihood of healthy aging, as represented by aging-GIP and mvAge, as well as healthspan, resilience, and parental lifespan, and was consistently independent of childhood BMI, adulthood BMI, and adulthood WC. Out of 100 candidates, 26 and 25 mediators were identified to explain the effect of birthweight on aging-GIP and mvAge, respectively, encompassing socioeconomic indicators (i.e., educational attainment, household income, and occupational attainment); behaviors (i.e., cheese intake, age at first sex, age at first live birth, and leisure screen time); physical functions (e.g., glycemic traits, SBP, grip strength, lipids and lipoproteins, urinary albumin-to-creatinine ratio (UACR), alanine aminotransferase (ALT), and sex hormone-binding globulin (SHBG)); and cardiometabolic diseases (e.g., type 2 diabetes and CVDs), implying underlying mediating pathways and preventive targets. Our findings support the role of birthweight in shaping healthy aging, and enlighten socioeconomic, behavioral, and therapeutic interventive targets for bridging aging disparities attributable to lower birthweight.

Previous cohort studies on birthweight and aging have primarily focused on mortality rather than healthy aging phenotypes, reporting null [11, 86], inverse linear [12], or U-shaped relationships [87]. Differences in study population characteristics, sample size, and potential confounders might partly explain these conflicting observations. In this study, we found that independent of later-life obesity indicators, genetically determined each 1-SD (≈484 g) increase in birthweight was significantly associated with healthy aging, reflected by composite phenotypes including higher aging-GIP and mvAge. These associations were further corroborated by individual, interpretable phenotypes, including 12%–16% higher likelihood of longer healthspan remaining free from seven major age-related diseases or death, a 0.079–0.089 SD improvement in resilience (equivalent to approximately 0.3 fewer health deficits), and an increase of 1.22–1.74 years in parental lifespan. These findings largely isolate the potential impact of later-life obesity, indicating that healthy aging is rooted in healthy birth [8, 88]. Our exploratory analyses also suggest that fetal genetic effects, rather than maternal genotype (as a proxy for the intrauterine environment), may primarily drive the genetically determined birthweight’s influence on healthy aging. Two previous mother–offspring pairs (and father-offspring pairs) MR studies also reported null associations of maternal single nucleotide polymorphisms (SNPs) influencing offspring birthweight with offspring cardiometabolic outcomes, implying a limited role of maternal intrauterine influences as captured by maternal genetics [22, 89]. As maternal genetic instruments may imperfectly capture intrauterine exposures (e.g., non-genetic maternal factors like smoking and diet) [22], our results do not exclude the Developmental Origins of Health and Disease (DOHaD) processes driven by environmental factors beyond those explained by maternal genetics.

Given that birthweight is determined early in life, we further explored modifiable mediators in subsequent life trajectories to elucidate underlying mechanisms and identify specific preventive targets among individuals with a suboptimal birthweight. Of the initial 100 candidates, we revealed that several socioeconomic indicators such as educational attainment, household income, and occupational attainment, as well as behaviors including dietary intakes of cheese, age at first sex, age at first live birth, and leisure screen time, each mediated a substantial proportion (about 10% to 30%) of the total effect of birthweight on healthy aging. Importantly, these socioeconomic and behavioral mediators should be interpreted within a life-course framework rather than as direct biological consequences of birthweight. Although such traits are strongly shaped by family and social environments, accumulating evidence indicates that fetal growth restriction, proxied by lower birthweight, can exert lasting influences on brain structure and function, neurodevelopment, physical capacity, and overall health capital [90–92]. These early-life biological constraints may subsequently shape educational trajectories, occupational attainment, and health-related behaviors through health-related selection and capacity-dependent pathways. Although socioeconomic and behavioral factors seem to play a less direct role than physical functional factors in the pathway from birthweight to healthy aging, their influence might be more widespread and should not be overlooked [93]. Previous evidence has documented that most of these identified socioeconomic (e.g., education, income) and behavioral mediators (e.g., dietary habits) are important determinants of multiple health and longevity outcomes [18, 50, 94]. Notably, although cheese intake was identified as a potential mediator, it likely reflects broader dietary patterns and should not be interpreted as a direct dietary recommendation. Taken together, our findings highlight the public health implications of birthweight on healthy aging through later-life socioeconomic and behavioral factors, offering interventive opportunities to reshape aging trajectories for individuals with lower birthweight.

In addition, we revealed that of multiple physical functional factors and diseases mediating the effect of birthweight on healthy aging, cardiometabolic diseases such as type 2 diabetes and different types of CVDs manifested more pronounced mediation proportions (approximately 25% to 70%). These findings were supported by compelling evidence that restricted fetal growth might lead to alternations in insulin signaling pathways, vascular adaptations, and epigenetic modifications, resulting in disrupted glucose metabolism and CVDs in later life [6, 7, 95]. Additional physical function-related mediators including grip strength, lipids and lipoproteins, and liver and kidney indicators exerted an individual mediation proportion of less than 37%, most of which are well-documented contributors to aging [18, 50, 94]. These mediators help guide tailored prevention strategies for individuals with suboptimal birthweight later in life, focusing on managing cardiometabolic risk factors and enhancing physical functions (e.g., grip strength) to promote healthy aging. Notably, given that various mediators may influence healthy aging through overlapping or correlated biological pathways, we assessed each mediator separately in practical scenarios. While this approach improves interpretability and methodological stability, the resulting mediation proportions cannot be summed and should be interpreted with caution.

This study highlighted the importance of birthweight and its subsequent life-course pathways in shaping healthy aging, with imperative public health and clinical implications. We outlined mediators that span the entire lifespan, encompassing early-life factors like age at first sex, midlife physical functions, and late-life cardiometabolic diseases, alongside lifelong behaviors and socioeconomic factors. The temporal breadth and diversity of mediators reflect the complexity of how birthweight influences healthy longevity throughout the human lifespan and underscore the multifaceted nature of this relationship, aligning with the World Health Organization life-course framework [1, 96]. This conveys an encouraging public health implication: low birthweight, as a marker of suboptimal early-life development, does not inevitably lead to an adverse aging trajectory. Instead, modifying key mediators in later life can help mitigate early-life disadvantages, offering opportunities to promote healthy aging equity.

The main strengths of this study included the rigorously designed MR analytical framework, large sample size (over one million), a comprehensive and representative array of aging phenotypes, meticulous and extensive considerations of candidate mediators, and the robust associations inferred from the main analysis supported by various sensitivity analyses and replication analyses. However, several limitations should be acknowledged. First, common genetic variants explain a smaller proportion of inter-individual variation in mortality compared to the exposome [97]. Although multiple sensitivity analyses and complementary tests yielded consistent results, indicating no evidence of weak-instrument bias and suggesting that horizontal pleiotropy and sample overlap were unlikely to substantially influence the causal estimates, the MR assumptions cannot be fully verified [20]. Accordingly, the causal associations should be interpreted with caution, and our MR estimates should be viewed as providing a stable, genetically informed layer of evidence that complements observational evidence. Second, although the composite healthy aging phenotypes (aging-GIP and mvAge) capture the broad genetic liability across various aging-related traits that include binary and continuous units, their abstract nature makes the interpretation of MR estimates challenging. Nonetheless, we also presented corroborative evidence from individual aging phenotypes (e.g., healthspan and parental lifespan) that more directly map to clinical endpoints, which may facilitate the clinical interpretation of MR estimates that reflect lifelong estimates [20]. Third, the obesity indicator as a covariate in the MVMR analyses was treated as a linear continuous exposure. Therefore, the resulting estimates reflect average linear effects of genetic liability to obesity. Although observational evidence reports non-linear associations between BMI and mortality [98], the potential non-linear effect of the obesity indicator was not taken into account in our analyses. Fourth, the potential exposure-mediator or mediator-mediator interactions cannot be modeled in the present two-step MR setting. Nevertheless, the MR approaches use genetic variants fixed at conception and naturally randomly assigned to individuals as proxies for the exposure and the mediator [20], which largely alleviates the potential bias from these interactions, and we assess the mediation proportion of each mediator individually in practical scenarios. Fifth, although we included common and important socioeconomic indicators, behaviors, physical functions, and diseases to facilitate clinical practice, this study cannot fully capture all mediation pathways, such as non-heritable factors (e.g., pension resources allocation influenced by national policies). Sixth, to limit potential influence from population stratification, our analyses were restricted to individuals of European ancestry, and the generalization of our findings to other ethnic groups warrants further investigation.

Conclusions

In conclusion, this study provides genetic evidence that birthweight is associated with healthy aging phenotypes, and that part of this association appears to operate through modifiable socioeconomic, behavioral, functional, and cardiometabolic pathways. Our findings support the importance of adopting a life-course approach to foster healthy aging by optimizing birthweight and targeting subsequent modifiable mediators, including improving socioeconomic status, promoting healthy behaviors, enhancing physical functions, and preventing cardiometabolic diseases.

Authors’ contributions

LK, TW, RL, and YB contributed to the conception and design of the study. LK performed statistical analysis and drafted the manuscript. TW critically revised the manuscript and checked the statistical analysis. TW, YB, WW, and GN obtained funding. TW, LK, RL, YB, CD, CY, DL, MC, JZ, MX, YX, ML, ZZ, JL, YC, GN, and WW contributed to the investigation, writing—review and editing, and the final approval of the version to be published. TW is the guarantor of this work and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Authors’ Twitter handles

Tiange Wang’s X: @tiangewang1031.

Funding

This work was supported by the grants from the National Natural Science Foundation of China (82570955 and 82370820), the Noncommunicable Chronic Diseases-National Science and Technology Major Project (2023ZD0508603), the National Key R&D Program of China (2023YFC2506700), the “Shu-Guang Scholar Programme” from Shanghai Municipal Education Commission, and the Innovative Research Team of High-level Local Universities in Shanghai.

Data availability

All GWAS summary statistics analyzed in this study are publicly available as shown in Table 1 and Additional file 2:Table S2 for download by qualified researchers. The GWAS data for birthweight can be obtained from the EGG Consortium (http://egg-consortium.org/), for healthy aging phenotypes can be retrieved or requested from the study authors at 10.7488/ds/2972 (the aging-GIP), https://zenodo.org/record/7926323 (mvAge), 10.6084/m9.figshare.9204998.v3 (frailty index), http://www.nealelab.is/uk-biobank (self-rated health), 10.5281/zenodo.1302861 (healthspan), 10.7488/ds/2463 (parental lifespan), https://datashare.ed.ac.uk/handle/10283/3645 (PhenoAge), and https://www.longevitygenomics.org/downloads (longevity). All data generated in the current study could be obtained from the Supplementary Materials.

Declarations

Ethics approval and consent to participate

All the summary-level GWAS data used in the analyses are publicly available, and therefore ethical approval was not imperative for this study. Ethical approval for the GWASs can be found in the corresponding GWAS publications cited in the manuscript.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.