Abstract
Background:
Several observational studies have found leukocyte telomere length (TL) to be associated with Alzheimer’s diseases (AD) or dementia. However, these findings were based on small sample sizes and cannot clarify whether this relationship was causal. Genome-wide association studies (GWAS) have identified common variants associated with TL, providing a valuable resource for examining the causal effect of TL on AD using Mendelian Randomization (MR) methods.
Objective:
To examine if TL was causally associated with AD using GWAS summary statistics.
Methods:
Using a genetic risk score comprised of seven variants associated with leukocyte TL as an instrumental variable, we tested whether shorter TL was associated with a higher risk of AD by applying an MR approach to the summarized genome-wide association study data.
Results:
The genetic risk score for TL was associated with higher risk of AD [log-odds ratio (OR) = 0.003 for per TL-decreasing allele; 95% confidence interval (CI): 0.001, 0.005, p = 0.005]. Moreover, the MR analysis provided support for shorter TL to be causally associated with a higher risk of AD (log-OR = 0.04 per SD-decrease of TL; 95% CI: 0.01, 0.08, p = 0.01).
Conclusion:
We suggest that TL has a causal effect on the risk of AD.
INTRODUCTION
Telomeres are sequences of repetitive nucleotides at the end of the chromosomes, which protect the chromosomes from deterioration or from fusion with neighboring chromosomes. Observational studies have found associations between shorter telomeres and cognitive decline, Alzheimer’s disease (AD), cardiovascular diseases, and mortality [1 –11], and telomeres have been considered as a marker of biological aging and aging-related outcomes [12, 13]. However, observational studies suffer from residual confounding and reverse causation, making it difficult to draw conclusions on whether telomere length is causally associated with AD, or simply a marker of an underlying pathological process.
For the past decades, the instrumental variable (IV) analysis has been developed for assessing causality using genetic variants in epidemiological research under the name of Mendelian randomization (MR) [14]. MR analysis has the advantage of being independent of common measured and unmeasured confounders when using genetic variants as instrumental variables. However, an MR study also requires a large sample size in order to gain high statistical power because the effect size of genetic variants on the outcome is usually small. A recent genome-wide association study (GWAS) has identified seven variants associated with leukocyte telomere length (LTL) [15]. These genetic variants make up an ideal basis for an IV to be used in a casual examination of the association between telomere length and AD.
In the present study, we aim to examine the causal effect of LTL on AD, we will take advantage of the publicly available GWAS summary data from the latest efforts. In addition to that, we will test the MR assumptions regarding the pleiotropic effects of the selected single nucleotide polymorphisms (SNPs). We will investigate the causal effect of LTL on AD using GWAS summary data from Codd et al. [3] and from the latest AD GWAS [16] by applying MR methods. The present study will update a previous study using the new AD GWAS data and a similar approach [4].
METHODS
Leukocyte telomere length samples
Through the European Network for Genetic and Genomic Epidemiology (ENGAGE) Consortium, a GWAS was conducted to investigate common variants associated with mean LTL [15]. This study consisted of 37,684 individuals with replication of selected variants in an additional 10,739 individuals. All individuals were of European descent. Mean LTL was measured using a quantitative PCR-based technique in all samples. This method expresses telomere length as a ratio (T/S) of telomere repeat length (T) to copy number of a single copy gene (S), in each sample. Seven lead SNPs, including five new loci, were identified to be of genome-wide significance. Details of genotyping methods and quality control criteria have been described previously [15]. All of these seven SNPs were used both as single instrumental variables, and combined in a genetic risk score (GRS), in the following MR analysis.
AD samples
The latest AD GWAS included 455,258 individuals of European ancestry [16]. Meta-analysis was conducted in three phases. Phase 1 recruited 24,087 clinically diagnosed late-onset AD cases, and 55,058 controls from the International Genomics of Alzheimer’s Project (IGAP), the Alzheimer’s disease working group of the Psychiatric Genomics Consortium (PGC-ALZ), and the Alzheimer’s Disease Sequencing Project (ADSP). Phase 2 analyzed an AD-by-proxy phenotype, based on individuals in the UK Biobank (UKB) for whom parental AD status was available (N proxy cases = 47,793; N proxy controls = 328,320). Finally, in phase 3, all individuals of phase 1 and phase 2 were analyzed together and tested for replication in an independent sample. The current publicly available dataset included 13,367,301 genetic variants.
Statistical analysis
To assess the impact of the seven LTL variants on the risk of AD, we performed a GRS analysis by combining summary statistics from each SNP into a score [17]:
The standard error (SE) of this estimate was calculated by:
For the causal effect of LTL on AD (β
IV) using single SNPs as instrumental variables, we used MR methods to calculate the causal estimates by the division of each SNP’s effect on AD (β
SNP_AD) by the corresponding SNP’s effect on LTL (β
SNP_LTL):
The SEs of the IV estimators were calculated by the following equation that has been evaluated for this purpose:
To estimate the causal effect of LTL on AD using GRS as an instrumental variable (β IV), we first calculated the combined GRS for LTL (β GRS_LTL ) and for the effect of LTL on AD (β GRS_AD ) from summary statistics as described elsewhere, before the MR calculation was done. This analysis also has the advantage of being able to test the pleiotropic effects of multiple SNPs on a specific trait. Statistical analyses were performed in R 3.4 using the MendelianRandomization package [18].
RESULTS
All of the seven SNPs were strongly associated with leukocyte LTL (Table 1). The respective effects of seven SNPs on AD were in the same direction. MR analysis showed that longer LTL was causally associated with lower risk of AD from the inverse variance weighted method [log-OR: –0.04 per SD-decrease of LTL; 95% CI: –0.07, –0.01; p = 0.01]. The intercept of MR-Egger regression was close to zero which was not against the violation of MR assumptions. Both simple median regression and weighted median regression showed similar results (Table 2 and Fig. 1).
Genetic variants and their associations with telomere length and AD
Mendelian randomization estimates of the association between telomere length and AD
OR, odds ratio.
The plot of the MR estimates of the association between telomere length and AD.
DISCUSSION
In the present Mendelian randomization study, we examined the causal effect of LTL on AD using summary GWAS data. Based on seven LTL associated SNPs serving as instrumental variables, we found that a decrease in LTL was associated with higher odds of AD. This study, to our knowledge, is the second study investigating the causal effect of LTL on AD using the MR approach. The results suggest that LTL might play an important role in AD etiology.
To date, several studies have been conducted to investigate the association between telomeres and AD. A few case-control studies reported telomere shortening in peripheral blood cells of AD patients compared with controls [1 , 20]. A community based prospective study with more than 9 years follow up duration found that every 1000 base pairs (bps) decreased telomere length was associated with a 21% higher risk of AD incidence [2]. Likewise, another study with two years follow-up time showed telomere length could predict dementia in patients with stroke [21]. However, most of these studies were based on small sample sizes, with a limited number of adjustments made for possible confounders, such as age and sex.
Five of the seven loci used in the GRS contain genes (TERC, TERT, NAF1, OBFC1, and RTEL1) are known to be involved in telomere biology [15]. The telomerase RNA component (TERC, encoded by TERC) and telomerase reverse transcriptase protein (TERT, encoded by TERT) may also have a protective role against tau pathology in AD patients [22]. Zinc finger proteins (ZNFs), such as ZNF208, regulate gene transcription through this binding. Some ZNF proteins have been found to contribute to an anti-apoptotic effect in astrocytes of AD [23]. Acylphosphatase-2, an enzyme encoded by the ACYP2, functions as hydrolyzing phosphoenzyme intermediates of membrane pumps, such as Ca2 +-ATPase. Elevated cytosolic calcium may result in calcium overload of mitochondria and further inhibits mitochondrial ATP production and other energy-dependent functions [24]. Finally, an energy deficit results in neuronal damage.
The main limitations of our study are due to MR assumptions, some of which could be outweighed by using summary GWAS data. The SNPs we selected were strongly associated with LTL such that they meet the strong IV assumption for MR study. However, the second assumption, that IVs should be independent of confounders, could not be statistically tested when using summarized data albeit it could be partially examined by testing the associations between IV and other confounders when using individual-level data. The third assumption, no pleiotropic effects, can be partly assessed when using multiple SNPs in a GRS. We tested this and found no evidence for pleiotropy in any of the seven loci used. The effect estimates of variants associated with LTL and AD were obtained from two different data collections, both of which used cohorts of European ancestry with large sample sizes. Hence, we assumed they were representative of the general European populations. However, provided a non-linear relationship between LTL and AD existed, this association cannot be assessed in our study using summarized GWAS data [17, 25]. The publicly available GWAS summary statistics data on AD, unfortunately, do not have sex-specific analyses. Therefore, we cannot perform sex-specific analyses or examine if ApoE could have an influence on the observed associations due to the limitation in the publicly available AD GWAS summary statistics. The MR analyses cannot distinguish if the exposure (e.g., LTL) occurred in the prodromal or early phases of the disease (e.g., AD). In this study, telomeres were measured in blood leukocyte. A previous study found that telomeres in blood had a moderate correlation with that in other tissues [26]. However, due to the sample availability, blood leukocyte has been measured and considered in most epidemiological studies.
In summary, our present study applied an MR approach to providing evidence for a causal relationship between LTL and AD. Further elucidation of this association could provide pivotal insights into the physiological roles of telomeres in AD pathogenesis.
