WHAT IS ALREADY KNOWN ON THIS TOPIC

It is well established that a shorter lifespan or premature death could be ascribed to modifiable lifestyle factors, specifically tobacco use, alcohol consumption, diet quality, and physical activity. A health-conscious lifestyle might have great potential to assuage the genetic susceptibility towards a shorter lifespan.

There has been no investigation to probe the joint effects of lifestyle factors and genetic determinants on human lifespan.

The extent to which a healthy lifestyle could counterbalance the high genetic risk remains elusive.

WHAT THIS STUDY ADDS

A high genetic risk corresponded to a 21% increased risk of death compared with a low genetic risk independent of lifestyle factors.

Genetic and lifestyle factors manifested independent associations with lifespan.

Adherence to healthy lifestyles could largely attenuate the genetic risk of shorter lifespan or premature death.

HOW MIGHT THIS STUDY AFFECT RESEARCH, PRACTICE OR POLICY

This study elucidates the pivotal role of a healthy lifestyle in mitigating the impact of genetic factors on lifespan reduction.

Given that our analysis was confined to white-European ancestry, the generalisability of our findings should be further evaluated in more diverse populations.

Public health policies for improving healthy lifestyles would serve as potent complements to conventional healthcare and mitigate the influence of genetic factors on human lifespan.

Introduction

Human lifespan is modulated by a combination of genetic and non-genetic factors including lifestyle behaviours.1 The heritability of lifespan has been estimated to be around 16% according to a study with sufficient global scope conducted on large genealogical trees.2 Apolipoprotein E (APOE) has been perceived as a longevity gene, which was identified as the top associated locus at genome-wide significance and consistently replicated in several studies.3–7 Other genetic loci such as CHRNA 3/5, LPA, CDKN2B-AS1 and LDLR are additionally identified to be associated with lifespan from recent genome-wide association study (GWAS) meta-analyses.7–9 Although single genetic variant accounts for only a small fraction of the variability of human lifespan, the polygenic risk score (PRS) combining multiple loci together provides a measurement of the predisposition for longer lifespan and more potential clinical utility.8

It is well established that shorter lifespan or premature death could be attributable to modifiable lifestyle factors, in particular tobacco use, alcohol consumption, diet quality, and physical activity.10–16 A healthy lifestyle may be able to attenuate the genetic risk of shorter lifespan. Researchers revealed strong negative correlations between body fat, smoking, and susceptibility to coronary artery disease and longer lifespan.16 However, there is no study to examine the joint effects of lifestyle factors and genetic determinants on the human lifespan. The extent to which individuals with high genetic risk can be offset by a healthy lifestyle remains elusive.

In this study, we incorporated data from three large population-based cohorts (LifeGen, US National Health and Nutrition Examination Survey (NHANES), and UK Biobank) to create a polygenic risk score to capture the genetic susceptibility associated with human lifespan, to assess the influence of common lifestyle factors (ie, smoking, alcohol consumption, diet, physical activity, body shape, and sleep duration), and to investigate the joint effects across genetic and lifestyle factors on human lifespan.

MethodsStudy design and data sources

This study adopted a multi-staged design by incorporating data from the LifeGen,16 US NHANES,17 and UK Biobank cohorts.18 The GWAS statistics derived from the LifeGen cohort, which included 26 independent European-heritage population cohorts with data on 606 059 parental lifespans,16 were used for the construction of lifespan PRS. The healthy lifestyle scores (HLS) associated with human lifespan were generated using data from the 2005–2018 NHANES. The established PRS and HLS were then adopted in the independent UK Biobank cohort to assess the joint effects across genetic and lifestyle factors. Details on the study populations and data preparation are described in the online supplemental methods and our previous studies.8 Local institutional review board ethics approval was not necessary for this study. All participants provided informed consent at the baseline assessment.

Polygenic risk score of lifespan

Independent genetic variants, captured by the LifeGen GWAS to be associated with human lifespan (p<5e-08) without linkage disequilibrium (r2<0.001), were used for the construction of PRS (online supplemental table 1).8 A polygenic risk score for lifespan was constructed for all individuals in the UK Biobank by multiplying the number of lifespan-decreasing alleles for each single nucleotide polymorphism (SNP) by its effect size on lifespan and then summing up this weighted score for all used SNPs. This PRS was then used to categorise the UK Biobank participants into long (<lowest quintile), intermediate (quintiles 2 to 4), and short (>highest quintile) groups to present the genetically determined human lifespan. The genotyped APOE SNPs rs429358 and rs7412, which possessed the largest genetic effect on lifespan, were further stratified to examine how the APOE ε4 status, polygenic risk, and lifestyles interplay together to influence the human lifespan in the UK Biobank participants.

Healthy lifestyle score

We adopted six common lifestyle factors associated with lifespan according to previous evidence,19 20 that is, smoking, alcohol consumption, physical activity, body shape, sleep duration, and diet. The definitions of lifestyle factors for NHANES and UK Biobank cohorts are shown with full details in online supplemental methods and table 2. Effect estimates (β coefficient) derived from NHANES were used for the construction of a weighted and standardised HLS in the UK Biobank participants. Briefly, we performed the Cox proportional hazards regression model in US NHANES to obtain β coefficients of each lifestyle factor with adjustment for other lifestyle factors and available covariates. Then, a weighted and standardised HLS was constructed in UK Biobank based on the β coefficient of each lifestyle factor derived from US NHANES. This score was then used to categorise UK Biobank participants into unfavourable (<lowest quintile), intermediate (quintiles 2–4), and favourable (>highest quintiles) groups.

Lifespan ascertainment

Lifespan was defined as the date of death minus the date of birth or the sum of age at baseline and follow-up time. In US NHANES, death certificate records were linked by the National Center for Health Statistics through the National Death Index to 31 December 2019. In our analysis, data for the lifespan of survivors in UK Biobank were censored on 31 December 2021. Death event was ascertained using the International Classification of Diseases, Tenth Revision (ICD-10) coding system and obtained from data field 40 000 and 40 001. Deaths due to accidents and injuries or COVID-19 were excluded.

Covariates

Information on covariates, including age (continuous in years), sex (men and women), education attainment (college or university degree and above, and high school and below), and socioeconomic status, were collected in the baseline questionnaire. The Townsend deprivation index as a complex indicator of socioeconomic status was constructed using the method mentioned online (https://biobank.ndph.ox.ac.uk/showcase/label.cgi?id=76). The Charlson Comorbidity Index (CCI) was defined using the method developed by Quan et al (based on ICD-10 and enhance ICD-9-CM; online supplemental table 3).21 Missing data were coded as a missing indicator category for categorical variables,22 using sex-specific means to impute the missing value for continuous variables.

Statistical analysis

Baseline characteristics of included participants were described across their survivorship as frequency (n) and proportion (%) for categorical variables and mean (±SD) for normally distributed continuous variables. Multivariable logistic regression analysis was used to assess the associations between the polygenic risk score and individual lifestyle factors. We applied the Cox proportional hazard regression model regressed against lifespan and surviving (alive or dead) status to examine the associations of genetic risk categories, lifestyle categories, and genetic risk and lifestyle combined categories. The model was fully adjusted for the covariates mentioned above as well as the first 20 principal components of ancestry.23 The interactions between the PRS and lifestyle factors were tested using a multiplicative interaction model. The calculation of life expectancy and its confidence interval was carried out for individuals with different genetic and lifestyle risk categories using flexible parametric survival models with age as timescale.24 The proportionality of hazards assumption was assessed using the Schoenfeld residuals method.

Secondary analysis was performed to derive the ‘optimal lifestyle combination’, in which we eliminated each lifestyle factor to reconstruct the weighted lifestyle score and rank the importance of the lifestyle according to the size of the coefficient. Several sensitivity analyses were also conducted, including: (1) analysis using the genetic risk quintiles instead of categories; (2) analysis using the number of healthy lifestyle factors instead of categories; (3) analysis using an unweighted lifestyle score; (4) analysis excluding participants with incomplete covariate data (n=2644). To examine the consistency of the association in subpopulations, we conducted stratification analyses by age (≥60 and <60 years), sex (female and male), education attainment (≥college/university and <college/university), and the tertiles of the Townsend deprivation index (from low to high, T1–3). We also stratified the analysis on the associations of the healthy lifestyle categories with death risk by genetic risk. We additionally adjusted self-reported family history of cancer, cardiovascular disease (CVD) or diabetes and depression symptoms, assessed using a two-item depression scale (PHQ-2).25

All p values were two-sided, and p<0.05 was considered statistically significant. All statistical analyses were performed using the R version 4.2.0.

Results

After excluding individuals who had no genetic data, failed to pass genetic quality control, died of COVID-19, injury, or accidental causes, or had missing data for lifestyle factors, 353 742 European heritage participants from the UK Biobank were included in the main analysis (table 1; online supplemental figure 1). Baseline characteristics of participants are demonstrated by the vital status (dead or alive) in table 1. Over a median follow-up of 12.86 years (IQR 12.14–13.55 years), 24 239 deaths were identified among eligible participants from the UK Biobank. Using the GWAS summary statistics from the LifeGen cohort, we obtained 19 independent SNPs to construct the lifespan PRS among independent UK Biobank participants (online supplemental table 1). The PRS was normally distributed (online supplementary fig 2online supplemental figure 2 2) and was not associated with any lifestyle factor other than healthy diet (OR 1.02, 95% CI 1.01 to 1.03) (online supplemental table 4). To generate the HLS associated with lifespan, we assessed each of the six common lifestyle factors using data from the US NHANES, which included 19 484 eligible adult participants and 1599 death events during a median follow-up of 6.92 years (IQR 3.83–10.42 years). The demographic characteristics of eligible participants are presented in online supplemental table 5. The associations of individual lifestyle factors with lifespan and their weights used for the construction of HLS are presented in online supplemental table 6, in which unfavourable lifestyle was in general associated with reduced length of lifespan for each of the component lifestyle factors.

Table 1

Baseline characteristics of eligible participants in UK Biobank

In the analysis of PRS, the risk of death increased across genetic risk categories (long to short) in a linear way (ptrend<0.001) (table 2). Compared to individuals in the genetic category of long lifespan, those in the genetic category of short lifespan had a higher hazard ratio of death (HR 1.21, 95% CI 1.16 to 1.26) (table 2). The associations remained significant after additional adjustment for lifestyle factors. The same pattern of associations was observed in the analysis using the PRS as a continuous variable, instead of categories (online supplemental table 7). The cumulative death rate during the follow-up was also higher in the high genetic risk group compared with the low genetic risk group (log rank p<0.001) (online supplemental figure 3).

Table 2

Risk of death according to genetic risk and lifestyle categories in UK Biobank

In analysis of HLS, the risk of death increased across lifestyle categories (favourable to unfavourable) in a dose-response manner (ptrend<0.001) (table 2). The HR of death for individuals in the unfavourable category was 1.78 (95% CI 1.71 to 1.85), compared with those in the favourable category. The associations did not change in sensitivity analysis with further adjustment for genetic risk (table 2) and in the analysis using unweighted HLS (online supplemental table 8). The cumulative death rate of participants during the follow-up was higher in the group with an unfavourable lifestyle compared with the group with a favourable lifestyle (log rank p<0.001) (online supplemental figure 3).

In the analysis of joint categories for genetic and lifestyle risk, the HR of death showed an increasing trend with elevated PRS and HLS (figure 1). Especially compared to individuals with genetic propensity for a long lifespan (low PRS) and a favourable lifestyle (high HLS), those with a genetic propensity for a short lifespan (high PRS) and an unfavourable lifestyle (low HLS) had 104% higher rates of death (HR 2.04, 95% CI 1.87 to 2.22, p<0.001). In contrast, individuals with a genetic propensity for a short lifespan (high PRS) but a favourable lifestyle (high HLS) (HR 1.26, 95% CI 1.14 to 1.39) had 54% lower rates of death than those with a genetic propensity for a short lifespan (high PRS) and an unfavourable lifestyle (low HLS) (HR 1.80, 95% CI 1.64 to 1.96).

Figure 1

Risk of death by joint categorisation for genetic risk and healthy lifestyle score in UK Biobank. Adjusted for age, age-square, sex, socioeconomic status, education, Charlson Comorbidity Index, and first 20 principal components of ancestry.

Strata analysis confirmed that an unfavourable lifestyle (low HLS) was associated with a higher risk of death across all genetic groups (table 3). We did not detect any multiplicative interaction between the PRS and the HLS (pinteraction=0.10). There was no statistically significant interaction between a healthy lifestyle and APOE ε4 (pinteraction=0.25). Also, the results remained consistent with the main analysis when stratifying by APOE ε4 carrier status (online supplemental table 9). The observed associations remained statistically significant in a series of sensitivity analyses: (1) using the unweighted HLS (online supplemental table 10); (2) excluding participants with missing data on covariates (online supplemental table 10); (3) additionally adjusted for self-reported family history of cancer, CVD or diabetes and depression symptoms (online supplemental table 11); and (4) stratified by age, sex, education attainment, and Townsend deprivation index (online supplemental table 12).

Table 3

Association of risk of death with lifestyle categories by genetic risk level

As for the secondary analysis, we additionally assessed the joint impact of genetic and lifestyle risk on the life expectancy of UK Biobank participants. The life expectancy at 40 years was 52.52 (95% CI 52.00 to 53.01) years for participants with a genetic propensity for long lifespan (low PRS) and a favourable lifestyle (high HLS), and was 45.83 (95% CI 45.35 to 46.32) years for participants with a genetic propensity for short lifespan (high PRS) and an unfavourable lifestyle (low HLS), with a mean difference of 6.69 (95% CI 5.98 to 7.39) years in lifespan (online supplemental table 13). An unfavourable lifestyle has a strong effect on the years of life lost regardless of which lifespan group an individual is at, and the subgroup of long lifespan and unfavourable lifestyle has the most years of life lost (figure 2). Among individuals with a genetic propensity for short lifespan (high PRS), those with a favourable lifestyle (high HLS) would have 5.22 (95% CI 5.18 to 5.24) years longer of lifespan than those with an unfavourable lifestyle (low HLS) (online supplemental table 13). Given that the largest proportion of participants had four healthy lifestyle factors (28.91%), the ‘optimal lifestyle combination’ for a prolonged lifespan were derived to be never smoking, regular physical activity, adequate sleep duration, and healthy diet, according to the rank of the size of the effect estimates (online supplemental table 14).

Figure 2

Years of life lost of other subgroups versus long lifespan and favourable lifestyle group by joint categorisation for genetic risk and healthy lifestyle score in UK Biobank.

Discussion

In this study, we comprehensively investigated the associations between genetic risk and lifestyle risk factors regressed against lifespan in 353 742 participants of the UK Biobank cohort. Our results indicated that a high genetic risk was associated with a 21% increased risk of death compared with a low genetic risk, independent of lifestyle factors. In contrast, an unfavourable lifestyle was associated with an approximately 78% increased risk of death compared with a favourable lifestyle within and across genetic risk categories. Furthermore, the genetic risk of a shorter lifespan or premature death might be offset by a favourable lifestyle by approximately 62%. Participants with a genetic predisposition to a short lifespan and an unfavourable lifestyle had a 2.04 times higher death risk compared with those with a genetic predisposition to a long lifespan and a favourable lifestyle. Our study also indicated that adherence to healthy lifestyles could substantially attenuate the loss of lifespan for individuals with genetic susceptibility to a shorter lifespan. In addition, we constituted the optimal lifestyle combination containing four lifestyle factors, including no current smoking, regular physical activity, adequate sleep duration, and a healthy diet, to bring better benefits for prolonging human lifespan.

To our knowledge, our study is the first to investigate the joint association of genetic risk and lifestyle factors with human lifespan. Previous epidemiological studies have established the critical role of healthy lifestyles in prolonging lifespan. Adherence to a healthy lifestyle is associated with a gain in life expectancy of 8.8 years (for men) and 8.1 years (for women) at age 30 in China, 12.2 years (for men) and 14.0 years (for women) at age 50 in the USA, and 17 years (for men) and 13.9 years (for women) at age 40 in Germany.26–28 Also, the life expectancy free of diabetes, cardiovascular diseases, and cancer at age 50 was 34.4 years for women and 31.1 years for men who adopted a healthy lifestyle.29 A study of middle-aged adults based on data from the UK Biobank found that engaging in a healthier lifestyle was associated with up to 6.3 years longer life for men and 7.6 years for women, regardless of the presence of multimorbidity.19 Zhang et al revealed that both in American and British adults, a minor portion of the disparity in health outcomes resulting from socioeconomic factors was mediated by unhealthy lifestyles.30 Another study in UK Biobank found that unfavourable lifestyles portended a higher risk of all-cause mortality and CVD mortality, independent of the genetic risk score constructed by 300 CVD-related SNPs.31 The general conclusions of these studies are consistent with our study. However, those studies only considered four or five healthy lifestyles and did not incorporate genetic factors for lifespan. Also, there have been studies constructing PRSs to evaluate the association with human lifespan.32 33 Researchers have explored the relationship between some of the lifestyles (eg, regular exercise, smoking status, etc) and lifespan individually. The joint relationship of multiple lifestyles with PRS and lifespan has not been discussed. In our study, we found that a healthy lifestyle could lower overall risk within and between genetic risk groups, and the genetic predisposition to a shorter lifespan can be substantially compensated by having a healthy lifestyle. Participants with high genetic risk could prolong approximately 5.22 years of life expectancy at age 40 with a favourable lifestyle. Given that lifestyle behavioural habits are usually developed before middle age, taking effective public health interventions is quite crucial for those at high genetic risk to extend their lifespan before the formation of a fixed lifestyle.

Strengths and limitations

The major strengths of our study include the prospective design, a large sample size from two well-established cohorts from the USA and the UK, and the availability of genotype and lifestyle information that enabled us to examine their joint effect comprehensively. We constructed the healthy lifestyle score derived from NHANES and applied it in UK Biobank to avoid the inflation of the weight. Also, we leveraged genetic loci associated with lifespan from LifeGen’s GWAS independent of UK Biobank to avoid overfitting.8 Notably, the adoption of GWAS data distinct from UK Biobank participants might mitigate the inherent heterogeneity in PRS computation. In addition, to enhance the robustness of the results, we included comprehensive sensitivity analyses by incorporating variables such as family histories of non-communicable diseases, known to exert significant influence on longevity, and the evaluation of depression symptoms. Furthermore, most people have shown poor adherence to healthy lifestyles in modern society. Evidence from the Nurses’ Health Study and the Health Professionals Follow-up Study showed that <2% of participants had five or more healthy lifestyle factors simultaneously,28 and only 6.2% of the UK Biobank population had six healthy lifestyle factors in our analysis. Therefore, our study brought up the concept of ‘optimal lifestyle combination’ for the first time. The combination of the listed four lifestyles could convey better benefits for a longer lifespan than any other combination of four healthy lifestyle factors, offering people health recommendations with strong practical implications.

Our study also has several limitations. First, there is still abundant room for further progress in determining the genetic variants associated with lifespan. While the polygenic risk score has included 19 validated SNPs, it explains only limited proportions of the genetic risk of a shorter lifespan. Second, the life expectancy at birth in the UK has approached 79.0 years for males and 82.9 years for females from 2018 to 2020.34 Nevertheless, the follow-up period of UK Biobank in our study is confined to about 12.86 years, and the longest lifespan observed is 87 years, which may engender an underestimate of our findings. Third, it is inevitable that self-reported lifestyle factors could lead to an incorrect assessment of healthy lifestyle scores. Fourth, lifestyle factors were measured only once at baseline. However, people may have made changes in their lifestyle during the follow-up for many reasons, such as the diagnosis of diseases, thus affecting risk estimates. Fifth, there were varied definitions of healthy lifestyle factors in the two cohorts we employed. Thus, we defined healthy lifestyle factors based on previous studies. Furthermore, previous research has shown significant variability in lifestyle choices across different age groups, especially among the young and elderly subgroups.27 Therefore, the age disparity between participants in NHANES and UK Biobank cohorts introduces complexity. In light of this, weighted lifestyle scores may not be able to capture effectively the nuanced impact of lifestyle choices within the UK Biobank cohort. Finally, previous research has suggested that the UK Biobank cohort is not fully representative of the general UK population given the ‘healthy volunteer’ selection bias and low participation rate.35 Also, our analysis was limited to white-European ancestry, making the findings less generalisable to the general population and other ethnic groups. The generalisability of our findings should be further evaluated in more diverse populations.

Conclusions

Our study reveals that genetic and lifestyle factors were independently associated with lifespan. Adherence to healthy lifestyles could significantly offset the genetic risk of a shorter lifespan or premature death. Accordingly, lifespan could be further extended with public intervention for healthy lifestyles across entire populations. Successes in several regions have set good examples that healthy lifestyle promotion policies would contribute significantly to increased life expectancy and reduced mortality.36–39 Public health policies for improving healthy lifestyles would be a potent complement to standard healthcare and diminish the impact of genetic factors on human lifespan.

Data availability statement

Data are available in a public, open access repository. NHANES data are available at http://www.cdc.gov/nchs/nhis/index.htm. UK Biobank study was under Application Number 66354. The UK Biobank is an open access resource and bona fide researchers can apply to use the UK Biobank dataset by registering and applying at http://ukbiobank.ac.uk/register-apply/. Further information is available from the corresponding author upon request.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

The study protocol for the US NHANES was approved by the US NHANES institutional review board and National Center for Health Statistics Research ethics review board. The North West Multi-Centre Research Ethics Committee approved the collection and use of UK Biobank data. All participants provided written informed consent. Institutional review board approval was waived for this analysis because of the publicly available and de-identified data.

Acknowledgments

Thanks are due to the technical support provided by the Core Facilities, Zhejiang University Medical Center/Liangzhu Laboratory.