THE CAUSAL RELATIONSHIP BETWEEN GENETICALLY PREDICTED BIOLOGICAL AGING, ALZHEIMER’S DISEASE AND COGNITIVE FUNCTION: A MENDELIAN RANDOMISATION STUDY

 

Y. Hao¹, W. Tian¹, B. Xie¹, X. Fu¹, S. Wang², Y. Yang¹

 

1. Department of Neurology and Neuroscience Center, The First Hospital of Jilin University, Jilin 130021, China; 2. Jilin Province Branch of ChinaUnicom, Jilin 130000, China.

Corresponding Author: Yu Yang, Department of Neurology and Neuroscience Center, The First Hospital of Jilin University, Jilin 130021, China. yang_yu@jlu.edu.cn

J Prev Alz Dis 2024;
Published online July 2, 2024, http://dx.doi.org/10.14283/jpad.2024.128

 

---

Abstract

Aging is one of the most important risk factors for Alzheimer’s disease (AD). Biological aging is a better indicator of the body’s functional state than age (chronological aging). Leukocyte telomere length (LTL) and epigenetic clocks constructed from DNA methylation patterns have emerged as reliable markers of biological aging. Recent studies have shown that it may be possible to slow down or even reverse biological aging, offering promising prospects for treating AD. Several observational studies have reported an association between biological aging, AD, and cognitive function, but the causality behind this association and the effects of different biological aging markers on AD risk and cognitive function remain unclear. Therefore, we explored the causal relationship between them by Mendelian randomization (MR) study. Inverse-variance weighted (IVW) method is the most dominant analytical method in MR studies, which is a weighted average of estimates from different genotype combinations, and this weighted average provides an overall estimate of the causal effect. The results of the IVW analyses showed that HannumAge acceleration and LTL shortening were able to increase the risk of late-onset AD (LOAD), but not early-onset AD (EOAD). Excellent prospective memory and fluid intelligence are potentially protective against GrimAge acceleration. GrimAge acceleration and HorvathAge acceleration increase the risk of LOAD through effects on LTL. Our findings provide important insights into the role of biological aging in the pathogenesis of AD, while also highlighting the interplay of different biological aging markers and their complexity in different AD subtypes.

Key words: Epigenetic age acceleration, leukocyte telomere length, Alzheimer’s disease, mendelian randomization, cognitive function.

---

 

Introduction

AD is the most common type of dementia, accounting for approximately 60%-80% of dementias (1). The typical pathological features of AD include extracellular amyloid-β (Aβ) plaques accumulated by Aβ and intracellular neurofibrillary tangles formed by phosphorylated tau (2). The clinical features are mainly progressive cognitive decline (3). It is estimated that there are currently more than 50 million people with dementia worldwide; however, the number of people with dementia is expected to triple by 2050 with the accelerated aging of the global population (4).AD is not a single disease, but rather a complex spectrum of disorders that includes different subtypes. The debate over the classification of AD into EOAD and LOAD according to age (5) has improved our understanding of the relationship between aging and AD, which has great clinical significance. EOAD usually develops at an early age (around 40-65 years old), with a clear family history of inheritance; whereas, LOAD usually develops above the age of 65 years, and the role of aging in LOAD, in addition to the genetic factors, should not be underestimated.
Unlike actual age (Chronological Age), which is simply calculated based on the date of birth, biological age reflects the true health status and degree of aging (physiological aging) of an individual. LTL and the epigenetic clock are currently the most reliable markers for assessing biological aging. The epigenetic clock is calculated by combining the DNA methylation status of selected cytosine-phosphoguanine (CpG) sites to estimate biological age (6). The first generation of epigenetic clocks, including HannumAge and HorvathAge, were developed using specific CpG loci that are strongly correlated with chronological age (7, 8). The second generation of clocks, including PhenoAge and GrimAge, were trained from a combination of CpGs loci in blood samples and age-related clinical markers (9, 10). Individuals with accelerated epigenetic age (EAA) represents that the predicted epigenetic age is greater than their actual age, suggesting a higher degree of aging compared to their peers (11). EAA is a better predictor of various health conditions and mortality risks than chronological age or epigenetic age (11). Several observational studies have shown that EAA is associated with a range of aging-related diseases, including AD, cancer, and heart disease. In AD, the epigenetic clock and age acceleration in the prefrontal cortex have also been associated with and amyloid load, neuroinflammation, and cognitive function (12, 13). However, the relationship between biological aging and AD is currently controversial (14-16). Inconsistency in findings may be due to selection bias, confounders, and reverse causation in observational studies.
MR study is an effective method used for causal inference, which uses genetic variation as an instrumental variable (IV) to derive causal relationships between exposures and outcomes, and is effective in avoiding confounding bias in traditional epidemiological studies (17). Genome-wide association studies (GWAS) have identified a large number of genetic variants associated with human diseases or phenotypes. These genetic variants (static genetic variants) are often fixed between individuals in a population and influence the performance of the relevant traits. In MR studies, genetic variants, usually single nucleotide polymorphisms (SNPs), are used as IVs for exposure factors (18). The basic principle of MR studies is Mendel’s law of inheritance (19), which states that when DNA is passed from parent to offspring at gamete formation, alleles segregate independently, thereby minimizing confounding factors and ruling out reverse causation. In the present study, we first used a bidirectional MR analysis to explain the causal relationship between biological aging (EAA and LTL), AD (LOAD and EOAD), and cognitive function. Secondly, a two-step MR approach was used to explore whether it could have a causal effect on late-onset AD via LTL (mediated mechanism) (Figure 1). Although the exposure factors in this MR study (EAA and LTL) were variable epigenetic variants, the SNPs used to represent the exposure factors were static genetic variants. Meanwhile, a series of screening metrics were used to ensure that these SNPs were strongly correlated with the exposure factors, thus satisfying the « relevance assumption» of the MR study.

MR analyses depend on three core assumptions; (Ⅰ) relevance assumption, (Ⅱ) independence assumption and (Ⅲ) exclusion assumption. (A)Bidirectional MR Study. The green represented the forward MR analyses, with biological aging as exposure and AD and cognitive function as the outcome. The red represented the reverse MR analyses, with AD as exposure and biological aging as the outcome. (B) Two-step MR Study with LTL as mediator. The total effect was decomposed into (Ⅰ) direct effect and (Ⅱ) indirect effect. β1 is the effect of EAA on LTL. β2 is the effect of LTL on LOAD. β0 is the total effect of EAA on LOAD. (Ⅰ) Indirect effect=β1×β2; the percentage of indirect effect=(β1×β2)/β0; (Ⅱ) Direct effect=β0-β1×β2; the percentage of direct effect=(β0-β1×β2)/β0.

 

Methods

Data sources

For detailed information on the GWAS summary data utilized in our study, please refer to Supplementary Table 1. It is important to note that all populations included in our study are of European ancestry, aiming to minimize potential biases arising from demographic heterogeneity.
The GWAS summary statistics of EAA was derived from the study by McCartney et al (20). In this study, four EAA datasets were obtained based on 28 cohort studies of 34,710 European ancestry researchers: GrimAge acceleration (GrimAA), PhenoAge acceleration (PhenoAA), HannumAA (HannumAA) and HorvathAge acceleration (HorvathAA). Cohort summaries of the EAA dataset can be found in Supplementary Table 2, which contains cohorts excluding the FinnGen and UK biobank. Details of the study design and experimental methods can be found in McCartney et al.
The GWAS summary statistics of LTL was derived from a recently published article, including 472,174 individuals of European ancestry in the UK Biobank and 20,134,421 SNPs. Age, sex, and ethnic group were adjusted in this study (21).
Alzheimer’s disease GWAS summary statistics from FinnGen, categorised into EOAD (N = 215,472) and LOAD (N = 217,541).
The cognitive function consists of three cognitive domains, namely prospective memory, fluid intelligence and reaction time. We assessed the above cognitive domains using four measures of cognitive function from the UK Biobank. Specifically, prospective memory was assessed by prospective memory test results (N = 152,605), and number of correct matches in round (N = 462,302); fluid intelligence was assessed using fluid intelligence score (N = 149,051); and reaction time was assessed using mean time to correctly identify matches (N = 459,523) (22). The Integrative Epidemiology Unit (IEU) Open GWAS Project(https://gwas.mrcieu.ac.uk/) contains Open GWAS data from multiple consortias (eg, UK biobank, FinnGen) or published articles. It includes cognitive function-related data from the UK Biobank mentioned above, so we directly used the GWAS ID provided by IEU for the MR analysis. Specific GWAS ID can be found in Supplementary Table 1.

Instrumental variables selection

To obtain reliable genetic IVs, MR analyses must fulfill three core assumptions (Figure 1). Details of the definitions, limitations and specific responses to the three core assumptions of MR are provided in Supplementary document 1.
We selected SNPs that were strongly associated with exposure with p-value less than 5×10-8. To exclude SNPs that were in strong linkage disequilibrium (LD), we performed the clumping procedure with R2 < 0.001 and a window size = 10000 kb (23). When some exposures (EOAD, PM1 and PM2) had few or even no SNPs available at the genome-wide significance threshold, we relaxed the significance threshold for these exposures (p < 5 × 10-7/5 × 10-6, R2< 0.01, window size = 5000 kb) (Supplementary Table 3). We calculated the F statistics for each SNP by the following equation: F=β2/Se2 (β: exposure β; Se: exposure se) (24). IVs with F statistics of less than ten were considered weak instruments and would be excluded for MR analysis (25).

Statistical analysis

In this study, we applied multiple approaches, including IVW, MR-Egger, weighted median, and weighted mode methods, to estimate the causal effects of exposures on outcomes. The IVW method was the predominant analysis method and was mainly used to assess causality. Its estimate can be regarded as the slope of a weighted linear regression of the IVs in the outcome on the exposure factor, and the intercept term is considered to be zero (26, 27). IVW provides accurate estimates when the selected SNPs are all valid IVs. IVW method is classified into fixed-effects and random-effects models based on the consistency of the differences in the SNP impacts, with the multiplicative random model being suitable for heterogeneous data (28). The MR-Egger method similarly uses the slope as a causal-effects estimate, but it takes into account the presence of an intercept term that can be used to assess multiplicity among IVs (29). The weighted mode method clusters SNPs into subsets based on the similarity of causal effects, which in turn estimates the causal effect of the subset with the largest number of SNPs (30). The weighted median method provides consistent estimates of the causal effect when the invalid IVs are as high as half of them (31). A Benjamini–Hochberg false discovery rate (FDR) was used to correct for multiple comparisons. IVW P values below 0.05 but not survived the FDR correction were considered as suggestive of a potential association.
MR-Pleiotropy Residual Sum and Outlier (MR-PRESSO), Cochran’s Q test, and MR-Egger intercept test were used for sensitivity analysis to evaluate the robustness of significant results (29, 32). MR-PRESSO method was used to detect outliers (33). If the outliers were detected, they would be removed, and we would repeat the MR causal analysis. Cochran’s Q test and MR-Egger intercept was used for evaluating directional pleiotropy and heterogeneity, respectively34. P-values below 0.05 were indicators of pleiotropy and heterogeneity. The flow chart of the MR study design is shown in Figure 2.

LD, linkage disequilibrium; MR, Mendelian randomization; SNP, single nucleotide polymorphisms; IVW_fe, IVW fixed-effect model; IVW_mre, IVW multiplicative random model.

 

All analysis were carried out using packages “TwoSampleMR” (https://github.com/MRCIEU/TwoSampleMR) in R version 4.1.2. The packages‘forestploter’ (https://cran.r-project.org/web/packages/forestploter/index.html) was used for statistical analysis, data output and visualization. P values less than 0.05 were considered statistically significant.

 

Results

The number of SNPs used for this MR analysis ranged from 4 to 115 after cascade selection of Figure 2 (Supplementary Table 4). The F-statistics of all these SNPs were above the threshold of 10, indicating no weak instrumental bias.

Bidirectional MR analysis between biological aging and Alzheimer’s disease

In the forward MR analysis with biological aging as “exposure” and AD as “outcome”, IVW MR analysis with a fixed-effects model showed that genetically predicted HannumAA and LTL were significantly associated with LOAD (Figure 3A, B). HannumAA had a positive correlation with LOAD (OR = 1.101, 95% CI=1.017 to 1.193, p = 0.017). LTL had a negative correlation with LOAD (OR = 0.809, 95% CI=0.686 to 0.954, p = 0.011). Meanwhile, the results remained statistically significant after the FDR (p = 0.043) (Supplementary Table 7). By sensitivity analysis, there was no significant heterogeneity (HannumAA, IVW Q_pval=0.071; LTL, IVW Q_pval=0.233) and pleiotropy (HannumAA, intercept pval=0.921; LTL, intercept pval=0.275) (Supplementary Table 6).
In the reverse MR analysis with AD as “exposure” and biological aging as “outcome”, we did not find strong evidence for a corelation between AD and biological aging (Figure 3C, D).

The standard line is the red dashed line. (A) The forward MR study with EAA as “exposure” and AD as “outcome”. (B) The forward MR with LTL as “exposure” and AD as “outcome”. (C) The reverse MR with AD as “exposure” and EAA as “outcome”. (D) The reverse MR study with AD as “exposure” and LTL as “outcome”. nsnp, number of single nucleotide polymorphisms; Beta, risk index; OR, odds ratio; 95% CI, 95% confidence interval; FDR, false discovery rate; IVW_fe, IVW fixed-effect model.

 

Bidirectional MR analysis between epigenetic age acceleration and cognitive function

In the forward MR analysis with epigenetic age acceleration as “exposure” and cognitive function as “outcome”, we did not find strong evidence of causality for EAA on cognitive function (Figure 4A).

The standard line is the red dashed line. (A) The Forward MR with EAA as “exposure” and cognitive function as “outcome”. (B) The reverse MR with cognitive function as “exposure” and EAA as “outcome”. nsnp, number of single nucleotide polymorphisms; Beta, risk index; OR: odds ratio; 95% CI, 95% confidence interval; FDR, false discovery rate; IVW_fe, IVW fixed-effect model; IVW_mre, IVW multiplicative random model. The standard line is the red dashed line. PM1, prospective memory (prospective memory test results); PM2, prospective memory (number of correct matches in round); FI, fluid intelligence (fluid intelligence score); RT, reaction time (mean time to correctly identify matches).

 

In the reverse MR analysis with cognitive function as “exposure” and EAA as “outcome”, IVW MR analysis with a fixed-effects model released that genetically predicted prospective memory and fluid intelligence were associated with GrimAge acceleration (PM2, pval=0.044; FI, pval=0.034) (Figure 4B). However, the above results did not stand up to FDR correction (p = 0.155) (Figure 4B). Heterogeneity and pleiotropy were not detected in the sensitivity analyses (PM2, IVW Q_pva l= 0.4003; intercept pval = 0.637; FI, IVW, Q_pval = 0.565; intercept pval = 0.794) (Supplementary Table 6).

 

Two step MR analysis with leukocyte telomere length as a mediator

For the resultant features produced by the bidirectional MR analysis, we pondered whether EAA could indirectly affect LOAD through LTL? Therefore, we used a two-step MR analysis to further investigate the causal relationship between EAA, LTL and LOAD. The results showed a negative correlation between GrimAA, HorvathAA and LTL (GrimAA, beta =-0.009, 95% CI=-0.014 to -0.003, p = 0.003; HorvathAA, beta = -0.009, 95% CI=-0.014 to -0.003, p = 0.002) (Supplementary Table 5). It was still significant after the FDR multiple comparisons (GrimAA, p=0.008; HorvathAA, p=0.008) (Supplementary Table 7). And fortunately, there was no heterogeneity (GrimAA, IVW Q_pval=0.203; HorvathAA, IVW Q_pval=0.103) and pleiotropy (GrimAA, intercept pval=0.892; HorvathAA, intercept pval=0.877) (Supplementary Table 6), indicating robust results. As the result of negative correlation existed between LTL and LOAD (Figure 3B), while no correlation existed between GrimAA, HorvathAA and LOAD (GrimAA, beta =-0.030, 95% CI=-0.138 to 0.078, p = 0.590; HorvathAA, beta = -0.013, 95% CI=-0.050 to 0.025, p = 0.507) (Supplementary Table 5). This suggests that the effect of GrimAA and HorvathAA on LOAD is exclusively mediated by LTL. Further we calculated the direct and indirect effects played by LTL as a mediator using the product method. The results showed that the direct effect (GrimAA, direct effect=-0.028, direct effect%= 93.753; HorvathAA, direct effect=-0.011, direct effect%= 85.561) of LTL on LOAD was greater than the indirect effect (GrimAA, indirect effect= 0.002, indirect effect%= 6.247; HorvathAA, indirect effect=-0.002, indirect effect%= 14.439) (Figure 5).

It demonstrates that GrimAA and HorvathAA increase the risk of LOAD exclusively by shortening LTL. The green arrow represents the relationship between cognitive function and EAA. This implies that better cognitive function (prospective memory and fluid intelligence) may slow EAA. The orange arrow represents the relationship between EAA and LOAD. It indicates that HannumAA can directly increase the risk of LOAD, not through LTL. p is the p-value after FDR multiple comparisons.

 

Discussion

In this study, we utilized MR methods to investigate the potential causal relationship between biological aging (EAA and LTL), AD and cognitive function. Our results provide key insights to deepen our understanding of AD pathogenesis, while also highlighting the complexity of different subtypes of AD.
Firstly, we analyzed the causal relationship between biological aging and AD through a comprehensive bidirectional MR study. By employing a bidirectional analysis approach, we were able to distinguish between upstream and downstream factors in the disease pathway, eliminating the possibility of reverse causation 35. The forward MR findings showed that EAA (HannumAA) can increase the risk of LOAD but not EOAD. This finding highlights the possible differences in etiological mechanisms of different AD subtypes. EOAD is usually governed by genetic factors, specific gene mutations, and AD-related pathology (36), whereas LOAD may be more susceptible to biological aging (37). Therefore, it is important to take into account the differences between these two AD subtypes when developing therapeutic and preventive strategies and not to ignore the effects of aging in LOAD. Notably, in our study, we did not observe significant evidence for an effect of the other three EAA (GrimAA, PhenoAA, HorvathAA) on LOAD. Of the four clocks, HannumAg may more directly reflect the biology of CpGs sites in blood, whereas HorvathAge, GrimAge, and PhenoAge may reflect more of a combined effect of tissues or multiple biological systems, including metabolism, inflammation, and so on. This suggests that DNAm sites in blood better explain the role of the epigenetic clock in LOAD. Perhaps we can find blood-based DNAm markers for treatment and prediction of LOAD (38).
There is inconsistency in the results of previous studies on the relationship between LTL and AD (39-42). Our study confirms and extends previous research on the ability of LTL shortening to increase the risk of AD. And our study also found that the effects of GrimAA and HorvathAA on LOAD may be solely attributable to LTL. This implies that LTL plays an important role in the etiological mechanism of LOAD (43). Shortening of LTL may lead to impaired cellular function and promote the development of LOAD (44), but the biological mechanisms need to be further explored. In-depth studies in this area will help develop more targeted therapeutic strategies.
Secondly, we similarly assessed the causal relationship between EAA and cognitive function by comprehensive bidirectional MR analysis. In the forward MR analysis, we failed to find a causal relationship between genetically predicted EAA and cognitive function. This suggests that the relationship between EAA and LOAD may be mediated by influencing specific pathophysiological processes, such as the accumulation of β-amyloid and tau proteins in the brain, which do not necessarily manifest directly in cognitive function changes at an early stage.The pathological progression of LOAD usually begins long before cognitive function declines markedly, and thus EAA may reflect more of these early pathological changes. In addition to this, prospective memory, fluid intelligence, and reaction time in this study represent only some of the domains of cognitive impairment in patients with Alzheimer’s disease and do not fully represent cognitive function. Therefore, more cognitive domains are needed to validate the relationship between EAA and cognitive function in the future. In the reverse MR study, we found that better cognitive functioning (prospective memory and fluid intelligence) was negatively correlated with EAA. This implies that cognitive training may have a protective effect against EAA. Future RCTs need to be designed to directly validate the effect of cognitive training on EAA.
Although our study provided valuable insights, there are some limitations to consider. Firstly, telomere length was measured in leukocytes, and although studies have shown that telomere length in leukocytes can be a valid proxy for telomere length in other tissue types, measuring telomere length in brain cells may be more helpful in linking telomere length to AD. Secondly, as the GWAS data on LTL and cognitive function were from the same population, this was not in accordance with the two-sample MR analysis principle. Therefore, we did not further explore the causal relationship between LTL and cognitive function. Thirdly, although we found that EAA (HannumAA) can increase the risk of LOAD, which is more predictive, actual age or biological age, in assessing the risk of LOAD needs to be observed in long-term longitudinal studies. Finally, the participants in this study were of European ancestry, which minimizes population stratification bias but limits the generalizability of our findings to other ethnicities.
Despite some limitations, the strengths of this study cannot be ignored. Our study was conducted using a bidirectional MR analysis, which uses germline genetic variation as an IV for exposure to study causal inference of the effect of exposure on outcome, thus minimizing the effects of confounders and reverse causality. Secondly, we used data on biological ageing (epigenetic age acceleration and LTL) that are currently the largest and most up-to-date GWAS database available. Thirdly, we made a distinction between different subtypes of AD that were differentially analyzed, which helped to reveal different etiological mechanisms and potential therapeutic strategies. This differential analysis can provide individualized therapeutic recommendations for clinical practice to better meet the needs of different AD patients. Lastly, all participants in the GWAS studies were predominantly of European origin and all studies had genomic controls, suggesting that population stratification and genomic inflation are unlikely to influence our results.

 

Conclusion

In conclusion, our findings indicated that HannumAge acceleration and LTL shortening can increase LOAD risk, but not EOAD. GrimAge acceleration and HorvathAge acceleration can increase LOAD risk by affecting LTL. Excellent prospective memory and fluid intelligence had a potentially protective effect on GrimAge acceleration. Our findings provide important evidence for the role of biological aging in the pathogenesis of AD, and also highlight the interplay of different biological aging metrics and their complexity in different AD subtypes. In the future, we need to further explore the functional mechanisms between biological aging and AD to lay a solid foundation for anti-aging therapy in AD.

 

Author contributions: Conceptualization, Y.Y., Y.H. and W.T.; methodology, Y.Y., Y.H. and W.T.; software, Y.H. and S.W.; validation, X.F. and B.X.; formal analysis, Y.H. and W.T.; data curation, W.T. X.F. and B.X.; writing—original draft preparation, Y.H.; writing—review and editing, Y.Y. and Y.H.; visualization, W.T. X.F. and B.X.; supervision, Y.Y.; All authors have read and agreed to the published version of the manuscript.

Funding: This study was supported by Jilin Medical and Health Talents Special Project (JLSWSRCZX2023-50) and Major Project on Brain Science and Brain-like Research (2022ZD0211605).

Acknowledgements: We want to acknowledge the IEU open GWAS project(https://gwas.mrcieu.ac.uk/) provide the original datasets of the current study. We want to acknowledge the participants in the UK biobank and FinnGen study and investigators that made all GWAS summary statistics publicly available.

Ethics approval and consent to participate: Not applicable.

Consent for publication: Not applicable.

Competing interests: The authors declare no competing interests.

 

SUPPLEMENTARY MATERIAL1

 

SUPPLEMENTARY MATERIAL2

 

References

1. Jasbi P, Shi X, Chu P, et al. Metabolic Profiling of Neocortical Tissue Discriminates Alzheimer’s Disease from Mild Cognitive Impairment, High Pathology Controls, and Normal Controls. Journal of proteome research 2021;20(9):4303-17. doi: 10.1021/acs.jproteome.1c00290 [published Online First: 2021/08/07]
2. Fani L, Ahmad S, Ikram MK, et al. Immunity and amyloid beta, total tau and neurofilament light chain: Findings from a community-based cohort study. Alzheimer’s & dementia : the journal of the Alzheimer’s Association 2021;17(3):446-56. doi: 10.1002/alz.12212 [published Online First: 2020/11/21]
3. Apostolova LG. Alzheimer Disease. Continuum (Minneapolis, Minn) 2016;22(2 Dementia):419-34. doi: 10.1212/con.0000000000000307 [published Online First: 2016/04/05]
4. Irwin MR, Vitiello MV. Implications of sleep disturbance and inflammation for Alzheimer’s disease dementia. The Lancet Neurology 2019;18(3):296-306. doi: 10.1016/s1474-4422(18)30450-2 [published Online First: 2019/01/22]
5. Mantyh WG, Cochran JN, Taylor JW, et al. Early-onset Alzheimer’s disease explained by polygenic risk of late-onset disease? Alzheimer’s & dementia (Amsterdam, Netherlands) 2023;15(4):e12482. doi: 10.1002/dad2.12482 [published Online First: 2023/10/02]
6. Jylhävä J, Pedersen NL, Hägg S. Biological Age Predictors. EBioMedicine 2017;21:29-36. doi: 10.1016/j.ebiom.2017.03.046 [published Online First: 2017/04/12]
7. Horvath S. DNA methylation age of human tissues and cell types. Genome biology 2013;14(10):R115. doi: 10.1186/gb-2013-14-10-r115 [published Online First: 2013/10/22]
8. Hannum G, Guinney J, Zhao L, et al. Genome-wide methylation profiles reveal quantitative views of human aging rates. Molecular cell 2013;49(2):359-67. doi: 10.1016/j.molcel.2012.10.016 [published Online First: 2012/11/28]
9. Levine ME, Lu AT, Quach A, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging 2018;10(4):573-91. doi: 10.18632/aging.101414 [published Online First: 2018/04/21]
10. Lu AT, Quach A, Wilson JG, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging 2019;11(2):303-27. doi: 10.18632/aging.101684 [published Online First: 2019/01/23]
11. Horvath S, Raj K. DNA methylation-based biomarkers and the epigenetic clock theory of ageing. Nature reviews Genetics 2018;19(6):371-84. doi: 10.1038/s41576-018-0004-3 [published Online First: 2018/04/13]
12. Levine ME, Lu AT, Bennett DA, et al. Epigenetic age of the pre-frontal cortex is associated with neuritic plaques, amyloid load, and Alzheimer’s disease related cognitive functioning. Aging 2015;7(12):1198-211. doi: 10.18632/aging.100864 [published Online First: 2015/12/20]
13. Grodstein F, Lemos B, Yu L, et al. The association of epigenetic clocks in brain tissue with brain pathologies and common aging phenotypes. Neurobiology of disease 2021;157:105428. doi: 10.1016/j.nbd.2021.105428 [published Online First: 2021/06/22]
14. Fransquet PD, Lacaze P, Saffery R, et al. Accelerated Epigenetic Aging in Peripheral Blood does not Predict Dementia Risk. Current Alzheimer research 2021;18(5):443-51. doi: 10.2174/1567205018666210823100721 [published Online First: 2021/08/26]
15. Degerman S, Josefsson M, Nordin Adolfsson A, et al. Maintained memory in aging is associated with young epigenetic age. Neurobiology of aging 2017;55:167-71. doi: 10.1016/j.neurobiolaging.2017.02.009 [published Online First: 2017/03/16]
16. Hillary RF, Stevenson AJ. An epigenetic predictor of death captures multi-modal measures of brain health. Molecular psychiatry 2021;26(8):3806-16. doi: 10.1038/s41380-019-0616-9
17. Davies NM, Holmes MV, Davey Smith G. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. BMJ (Clinical research ed) 2018;362:k601. doi: 10.1136/bmj.k601 [published Online First: 2018/07/14]
18. Bennett DA, Holmes MV. Mendelian randomisation in cardiovascular research: an introduction for clinicians. Heart (British Cardiac Society) 2017;103(18):1400-07. doi: 10.1136/heartjnl-2016-310605 [published Online First: 2017/06/10]
19. Castle WE. MENDEL’S LAW OF HEREDITY. Science (New York, NY) 1903;18(456):396-406. doi: 10.1126/science.18.456.396 [published Online First: 1903/09/25]
20. McCartney DL, Min JL, Richmond RC, et al. Genome-wide association studies identify 137 genetic loci for DNA methylation biomarkers of aging. Genome biology 2021;22(1):194. doi: 10.1186/s13059-021-02398-9 [published Online First: 2021/07/01]
21. Codd V, Wang Q, Allara E. Polygenic basis and biomedical consequences of telomere length variation. Nat Genet 2021;53(10):1425-33. doi: 10.1038/s41588-021-00944-6
22. Pagoni P, Korologou-Linden RS, Howe LD, et al. Causal effects of circulating cytokine concentrations on risk of Alzheimer’s disease and cognitive function. Brain, behavior, and immunity 2022;104:54-64. doi: 10.1016/j.bbi.2022.05.006 [published Online First: 2022/05/18]
23. Abecasis GR, Altshuler D, Auton A, et al. A map of human genome variation from population-scale sequencing. Nature 2010;467(7319):1061-73. doi: 10.1038/nature09534 [published Online First: 2010/10/29]
24. Chen L, Yang H, Li H, et al. Insights into modifiable risk factors of cholelithiasis: A Mendelian randomization study. 2022;75(4):785-96. doi: 10.1002/hep.32183
25. Pierce BL, Ahsan H, Vanderweele TJ. Power and instrument strength requirements for Mendelian randomization studies using multiple genetic variants. International journal of epidemiology 2011;40(3):740-52. doi: 10.1093/ije/dyq151 [published Online First: 2010/09/04]
26. Burgess S, Scott RA, Timpson NJ, et al. Using published data in Mendelian randomization: a blueprint for efficient identification of causal risk factors. European journal of epidemiology 2015;30(7):543-52. doi: 10.1007/s10654-015-0011-z [published Online First: 2015/03/17]
27. Burgess S, Butterworth A, Thompson SG. Mendelian randomization analysis with multiple genetic variants using summarized data. Genetic epidemiology 2013;37(7):658-65. doi: 10.1002/gepi.21758 [published Online First: 2013/10/12]
28. Burgess S, Davey Smith G. Guidelines for performing Mendelian randomization investigations. 2019;4:186. doi: 10.12688/wellcomeopenres.15555.2
29. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. International journal of epidemiology 2015;44(2):512-25. doi: 10.1093/ije/dyv080 [published Online First: 2015/06/08]
30. Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. International journal of epidemiology 2017;46(6):1985-98. doi: 10.1093/ije/dyx102 [published Online First: 2017/10/19]
31. Bowden J, Davey Smith G, Haycock PC, et al. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genetic epidemiology 2016;40(4):304-14. doi: 10.1002/gepi.21965 [published Online First: 2016/04/12]
32. Bowden J, Del Greco MF, Minelli C, et al. Improving the accuracy of two-sample summary-data Mendelian randomization: moving beyond the NOME assumption. International journal of epidemiology 2019;48(3):728-42. doi: 10.1093/ije/dyy258 [published Online First: 2018/12/19]
33. Verbanck M, Chen CY. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Erratum in: Nat Genet 2018;50(5):693-98. doi: 10.1038/s41588-018-0099-7
34. Burgess S, Davey Smith G. Guidelines for performing Mendelian randomization investigations. Wellcome open research 2019;4:186. doi: 10.12688/wellcomeopenres.15555.2
35. Qin J, Zhang L, Ke B, et al. Causal relationships between circulating inflammatory factors and IgA vasculitis: a bidirectional Mendelian randomization study. Frontiers in immunology 2023;14:1248325. doi: 10.3389/fimmu.2023.1248325 [published Online First: 2023/09/27]
36. Khemka S, Reddy A, Garcia RI, et al. Role of diet and exercise in aging, Alzheimer’s disease, and other chronic diseases. Ageing research reviews 2023;91:102091. doi: 10.1016/j.arr.2023.102091 [published Online First: 2023/10/14]
37. Palmer JM, Huentelman M, Ryan L. More than just risk for Alzheimer’s disease: APOE ε4’s impact on the aging brain. Trends in neurosciences 2023;46(9):750-63. doi: 10.1016/j.tins.2023.06.003 [published Online First: 2023/07/18]
38. Santamaria BA, Corroza J, Sánchez-Ruiz de Gordoa J, et al. Association of Blood-Based DNA Methylation Markers With Late-Onset Alzheimer Disease: A Potential Diagnostic Approach. Neurology 2023 doi: 10.1212/wnl.0000000000207865 [published Online First: 2023/10/13]
39. Liao Q, He J, Tian FF, et al. A causal relationship between leukocyte telomere length and multiple sclerosis: A Mendelian randomization study. Frontiers in immunology 2022;13:922922. doi: 10.3389/fimmu.2022.922922 [published Online First: 2022/08/02]
40. Gao K, Wei C, Zhu J, et al. Exploring the Causal Pathway From Telomere Length to Alzheimer’s Disease: An Update Mendelian Randomization Study. Frontiers in psychiatry 2019;10:843. doi: 10.3389/fpsyt.2019.00843 [published Online First: 2019/12/06]
41. Cao Z, Hou Y, Xu C. Leucocyte telomere length, brain volume and risk of dementia: a prospective cohort study. General psychiatry 2023;36(4):e101120. doi: 10.1136/gpsych-2023-101120 [published Online First: 2023/09/14]
42. Liu R, Xiang M, Pilling LC, et al. Mid-life leukocyte telomere length and dementia risk: An observational and mendelian randomization study of 435,046 UK Biobank participants. Aging cell 2023;22(7):e13808. doi: 10.1111/acel.13808 [published Online First: 2023/05/31]
43. Levstek T, Kozjek E, Dolžan V, et al. Telomere Attrition in Neurodegenerative Disorders. Frontiers in cellular neuroscience 2020;14:219. doi: 10.3389/fncel.2020.00219 [published Online First: 2020/08/08]
44. Bernardes de Jesus B, Blasco MA. Telomerase at the intersection of cancer and aging. Trends in genetics : TIG 2013;29(9):513-20. doi: 10.1016/j.tig.2013.06.007 [published Online First: 2013/07/24]

© Serdi 2024