Introduction
Late-onset Alzheimer’s disease (LOAD) is the most common cause of dementia in the elderly. It is a complex disorder that results from a combination of genetic and nongenetic risk factors, where the environment plays an important role in its development.Reference Scheltens, Blennow and Breteler 1 Age is the main risk factor for developing LOAD: the disease is present in 1% to 3% of those between ages 60 and 70, 3% to 12% of those between ages 70 and 80, and 25% to 35% of those older than age 85.Reference Farrer, Cupples and Haines 2 , Reference Reitz, Brayne and Mayeux 3 Some risk factors for LOAD are modifiable, that is, they are relevant for prevention. Factors such as smoking, years of education, cognitive stimulation, exercise, and diet have been found to play a role in the disease.Reference Silva, Loures C de and Alves 4 , Reference Robinson, Lee and Hane 5
Genetic variants in multiple genes have been identified that increase the risk for LOAD.Reference Cuyvers and Sleegers 6 , Reference Andrews, Fulton-Howard and Goate 7 Furthermore, there are also genetic variants that may be protective against the development of dementia at advanced ages.Reference Suri, Heise and Trachtenberg 8 The level of protection in each individual depends on the effect of multiple factors such as aging, genetics (protective and risk variants), lifestyle, cardiovascular health, and others. It has been proposed that individuals with a positive balance of protective factors can reach advanced ages without developing dementia,Reference Vemuri, Knopman and Lesnick 9 which we will refer to as successful cognitive aging (SCA).
It has been observed that one gene can carry both risk and protective variants for late-onset dementia. This is the case for apolipoprotein E (APOE)Reference Scacchi, De Bernardini and Mantuano 10 ; the APOE-ε4 allele increases the risk of LOAD, and the APOE-ε2 allele protects against it.Reference Corder, Saunders and Strittmatter 11 , Reference Corder, Saunders and Risch 12 However, some studies have found that the association between the APOE-ε4 allele and dementia is not present in individuals over 80 years of age.Reference Scacchi, De Bernardini and Mantuano 10 , Reference Valerio, Raventos and Schmeidler 13 These findings could support a survivor effect model for SCA in old age. The protected survivor model proposes that a minority of the general population has a protective factor that mitigates the negative effect of a risk factor on successful cognitive aging for the unprotected majority. As age increases, the proportion of survivors with protection increases. Therefore, although the association of the risk factor with survival does not change within an individual, the association in the surviving population changes as its age increases.Reference Silverman and Schmeidler 14
In the last years, epigenetic modifications of the genome have gained attention in complex diseases such as LOAD, and the identification of epigenetically dysregulated genes has been increasing.Reference Sanchez-Mut and Gräff 15 , Reference Zhang, Silva and Young 16 Some genes, such as ankyrin 1 (ANK1), sorbin and SH3-domain-containing 3 (SORBS3), and histone deacetylase 2 (HDAC2) genes have been reported as dysregulated by independent studies in humans.Reference Sanchez-Mut and Gräff 15 Epigenetic dysregulation of enhancers in neurons in AD has also been described, and the authors propose a hypothesis where hypomethylation of enhancers could induce the formation and progression of amyloid-beta plaques and neurofibrillary tangle pathology through BACE1 activation.Reference Liu, Lauro and Ding 17 Another recent whole-genome methylation analysis proposed a role for the immune system and polycomb complex involvement in AD.Reference Zhang, Silva and Young 16
The validity of using peripheral blood for epigenetic studies of disorders of the brain (or other tissues) has often been questioned. This is unavoidable in most studies, because of the challenge or impossibility of obtaining brain tissue for analyses. Recent research has shown that, in general, there is a robust correlation in methylation levels between blood and the brain, although this can vary greatly for different genomic regions.Reference Walton, Hass and Liu 18 - Reference Chen, Zang and Braun 21 It has been suggested that interpretation of blood methylation results for brain disorders should focus on CpG sites with a high correlation in DNA methylation across both tissue types, and some tools have been developed for this purpose.Reference Edgar, Jones and Meaney 19 , Reference Braun, Han and Hing 20
While previous work has shown that some risk factors for LOAD could have an age-dependent effect,Reference Scacchi, De Bernardini and Mantuano 10 , Reference Valerio, Raventos and Schmeidler 13 most of the analyses have been performed on individuals under 80 years. Above those ages, the effect of risk factors for LOAD remained underexplored. Similarly, most epigenetic and nonepigenetics studies of SCA have focused on individuals between 65 and 85 years old, and representation of nonagenarians or centenarians has been scarce.Reference Depp and Jeste 22 It has been proposed that cognitively healthy individuals aged 90 years and above are an optimal population to study genetic protective factors for LOAD.Reference Silverman, Schnaider-Beeri and Grossman 23 Their first-degree relatives have been observed to maintain intact cognitive function more frequently than relatives of younger nondemented elderly. Also, a high heritability of cognitive functions such as memory has been identified in nonagenarians.Reference Silverman, Schnaider-Beeri and Grossman 23 , Reference Greenwood, Beeri and Schmeidler 24
Costa Rica is the second country in the American continent with the greatest longevity after Canada.Reference Rosero-Bixby and Dow 25 At the age of 90, Costa Rican nonagenarians have a life expectancy of 4.4 years, which is half a year more than any other country in the world.Reference Rosero-Bixby 26 Access to an almost universal health system, lifestyle factors, and family support have been proposed to influence the observed longevity in the country.Reference Rosero-Bixby and Dow 27 This study aims to identify differentially methylated regions (DMRs) in the genomes of a sample of cognitively healthy individuals and a sample of individuals with LOAD, all of them over 90 years of age, from the Central Valley of Costa Rica (CVCR).
Methods
Study population
The subjects were recruited in the study “Successful Cognitive Aging and Cardiovascular Risk Factors in the Central Valley of Costa Rica,” funded by an NIH Fogarty International Center & National Institute on Aging grant (R21TW009258), and a P01-AG02219 grant funded by the Alzheimer’s Association. The project and consent forms were reviewed and approved by the Institutional Review Board of the University of Costa Rica and the Mount Sinai Medical School. The study was explained to each subject and written informed consent was obtained. If the subject was unable to give consent because of cognitive impairment, consent was obtained from the spouse or the primary caretaker relative. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
The sample consisted of 32 nonagenarians, of which 11 had a diagnosis of probable LOAD. For this study, we only included female probands over 90 years old, with between 0 and 9 years of schooling, at least one pregnancy, married or widowed, menopause between 50 and 55 years of age, and without hormone replacement therapy with estrogen.
All subjects were clinically assessed by a medical geriatrician (DV) and a psychologist (LM-V) with a general medical examination, the Clinical Dementia Rating scale (CDR), and the Mini-Mental State Examination (MMSE).Reference Hughes, Berg and Danziger 28 , Reference Folstein, Folstein and McHugh 29 In the CDR, clinical information is collected from both an informant and the subject. A CDR score of zero indicates no dementia or a recent decline in cognition or functioning, and a score of 3 indicates severe dementia. In the MMSE, a score between 27 and 30 indicates the absence of cognitive impairment, and values below 6 indicate severe dementia. Diagnosis of probable LOAD was defined by the geriatrician based on the clinical history and the CDR and MMSE scores. Patients with a history of stroke or the presence of a disorder other than AD that potentially causes dementia were excluded. In the sample of individuals with probable LOAD we decided to include subjects with a CDR score of 2 (moderate; N = 3) or 3 (severe; N = 8). Individuals with mild cognitive impairment (with CDR scores above zero, but less than 2) were not included in our sample.
Genome-wide methylation analysis
Microarray analysis
Whole blood was collected by peripheral venipuncture from participants and DNA was extracted using the sucrose method.Reference Cheung, Hubert and Landry 30 DNA was bisulfite converted with the Zymo Research EZ DNA Methylation Kit (Irvine, CA). We obtained genome-wide methylation profiles using the Infinium HumanMethylationEPIC BeadChip (Illumina, Inc., San Diego, CA). The methylation data were converted to idat files in the GenomeStudio software (Illumina, Inc., San Diego, CA).
Idat files were processed with the minfi R package (https://www.bioconductor.org/packages/devel/bioc/vignettes/minfi/inst/doc/minfi.html).Reference Aryee, Jaffe and Corrada-Bravo 31 Quality control was first performed to detect samples that failed to adequately detect DNAm (in our case, all samples passed quality control). Then, we removed probes that could cross-hybridize or overlap with SNPs, which could confound results. Probes with a detection p > .05 and probes that failed in more than 50% of the samples were removed. The total number of probes post-filtering was 794,770. We used FunNorm normalization to remove unwanted variation by regressing out variability explained by the control probes present on the array.Reference Fortin, Labbe and Lemire 32 The proportion of DNAm at a particular CpG site (β values) was ascertained by taking the ratio of the methylated (C) to unmethylated (T) signal, using the formula: β = intensity of the methylated signal/(intensity of the unmethylated signal + intensity of the methylated signal + 100). β values range from 0 (completely unmethylated) to 1 (completely methylated).Reference Bibikova, Barnes and Tsan 33
Data availability
The data presented in this study are available upon request from the corresponding author; upload to a public repository was not done due to privacy and ethical issues.
Prediction of epigenetic age
The Horvath method in the R package was used to determine DNAm-based age prediction. This method uses a weighted average of DNA methylation at 353 clock CpG sites, which is then transformed to DNAm age using a calibration function (http://dnamage.genetics.ucla.edu).Reference Horvath 34 Mean differences between groups were assessed using the Kruskal–Wallis test.
Estimation of cell type proportions
We estimated the cell type proportions in our two groups because whole blood is made up of many different cell types, each with different methylation profiles that can vary in proportion with disease status. These cell types include CD8+ T lymphocytes, CD4+ T lymphocytes, natural killer cells, B cells, monocytes, and granulocytes. For this purpose, we applied Houseman’s algorithm using the minfi R package.Reference Houseman, Accomando and Koestler 35 A Mann–Whitney U test was used to evaluate the differences between groups for each cell type category.
Identification of differentially methylated probes and DMRs
We applied two different methods to identify differentially methylated probes (DMPs) and DMRs between nonagenarians with and without LOAD: DMRcate and ChAMP-Bumphunter.Reference Morris, Butcher and Feber 36 We assigned a p-value cutoff of .05, after false discovery rate (FDR) correction, to determine DMPs and DMRs in both DMRcate and Bumphunter. We also used the ChAMP-Bumphunter R package to do the gene set enrichment analysis based on our DMRs.
Once the DMRs were identified, we used the web application BECon to verify the correlation between the methylation in blood and the methylation in the brain for each one of the CpG sites of the DMRs.Reference Edgar, Jones and Meaney 19 This tool uses a DNA methylation database from paired samples of blood and three postmortem brain regions from individuals to show how informative DNA methylation from the blood is for brain DNA methylation.
Gene-level analysis
After reviewing existing literature on the genes included in the DMRs, we decided to focus on the PM20D1 gene, for which altered expression in AD has previously been reported.Reference Sanchez-Mut, Heyn and Silva 37 Specifically, two SNPs, rs708727 and rs960603 have been reported as expression and methylation QTLs for the gene.Reference Sanchez-Mut, Heyn and Silva 37 , Reference Sanchez-Mut, Glauser and Monk 38 Sanger sequencing was performed to genotype both SNPs. Average methylation (β values) in the DMR between the genotypes for rs708727 was compared with a Mann–Whitney U test (because only two out of the three possible genotypes were observed). For rs960603 all three possible genotypes were observed, and a comparison of average methylation between them was done with a Kruskal-Wallis test. The same comparisons of methylation levels between genotypes for both SNPs were done for each CpG site in the DMR. A stratified comparison by genotype of the methylation level between SCA and LOAD was also performed, using Mann–Whitney U tests for each pairwise comparison. The AA genotype for rs960603 was not included in the analysis for statistical reasons (only one individual each for SCA and LOAD was available).
Results
Epigenetic age
We assayed DNAm profiles of 32 females, of which 21 were cognitively healthy and 11 had a probable LOAD diagnosis. The age range is between 90 and 103 years old. The average chronological age of the LOAD group was 95 (SD = 3.36), and the average epigenetic age was 73 (SD = 5.99). In the SCA group, the average chronological age was 93 (SD = 2.77), while the average epigenetic age was 73 (SD = 5.85). Neither the epigenetic age (p-value = .53) nor the difference between epigenetic age and biological age shows differences between groups (p-value = .4200). Nevertheless, when analyzing the sample as a whole, a significant difference was obtained between chronological and epigenetic age (p = .0045). The age acceleration was lower on individuals diagnosed with LOAD compared to those cognitively healthy, but not statistically significant (p = .1972).
Differentially methylated regions
No statistically significant differences were found in the cellular composition between the LOAD and SCA groups, in any of the cell types, so this variable was not included in the analysis (Table 1). We identified several differentially methylated regions in our analysis, but no statistically significant differences at the probe level (DMPs) were detected. After detection of DMRs with the DMRcate and Bumphunter methods, we chose the 11 DMRs that both methods found in common for further analysis (Table 2). Four of the 11 DMRs are hypomethylated in the SCA group compared with the LOAD group, while seven of them are hypermethylated. The mean length of the DMRs was 603 bp, with 197 bp in the shortest region and 1736 bp in the longest. On average, in the 11 DMRs observed, there were 11 CpG sites per region, with a range of 6 to 18 CpG sites. Six out of the 11 DMRs include known genes, and two include pseudogenes. From the eight DMRs associated with genes and pseudogenes, six were located in promoters. The results of the correlation in blood and brain methylation for each of the regions are also shown in Table 2. The DMRs, were enriched on 94 diverse pathways, but the top 3 pathways were: Pilon KLF1 targets downregulated (adj. p = 4.44e-09), Martens bound by PML-RARA fusion (adj. p = 1.26e-06), and Blalock Alzheimer’s Disease up (adj p = 1.26e-06).
Cell Type Composition Comparison Between the SCA and LOAD Groups.
Differentially Methylated Genomic Regions Between Individuals with SCA and LOAD.
Note: Δβ: (Delta beta; absolute mean or max difference of β values between groups (SCA-LOAD).
* Genome coordinates from Human Genome GRCh37/hg19 Assembly.
After a review of existing literature, we decided to focus on the peptidase M20-domain-containing protein 1 (PM20D1) gene. An association of this gene with AD has previously been reported, as well as the existence of SNPs that act as expression and methylation QTLs for the gene.Reference Sanchez-Mut, Heyn and Silva 37 , Reference Sanchez-Mut, Glauser and Monk 38 The CpG sites included in this DMR are cg17178900, cg26354017, cg14159672, cg14893161, cg07533224, cg12898220, cg05841700, cg11965913, cg07167872, cg24503407, cg16334093, and cg07157834.
mQTL SNPs in PM20D1
We tested whether there is a relationship between the genotypes at mQTL and eQTL SNPs rs708727 and rs960603 and methylation level in the CpG regions of PM20D1. For rs708727, average methylation (β values) in the PM20D1 DMR differs between genotypes, with higher average methylation in heterozygotes (AG) when compared to individuals homozygous for the G allele (p = 1.4E-05, Figure 1). For these SNPs, there were no homozygotes for the A allele in our sample. In the case of rs960603, no significant differences in average methylation levels were found between genotypes (p = 0.31, Figure 1). A comparison of methylation levels between genotypes for both SNPs at the 12 individual CpG sites in the DMR is presented in Table 3. The pattern is the same for the DMR as a whole: there is a significant difference in methylation between genotypes in all CpG sites for rs708727, and no difference in any of the sites for rs960603. No significant differences in methylation levels between the SCA and LOAD groups were found within each genotype for either rs708727 or rs960603 (p > 0.05 for all pairwise comparisons; Figure 2).
Comparison of average methylation (β values) between genotypes for rs708727 and rs960603.
Stratified comparison per genotype of average methylation (β values) between SCA and LOAD for rs708727 and rs960603. The p-values for the Mann-Whitney U test are shown for each comparison.
Comparison of the Average Methylation Level (β Values) Between Genotypes for rs708727 and rs960603 for all CpG Sites of the PM20D1 DMR.
Note: Values in the table are p-values for the Mann-Whitney U test for rs708727, and for the Kruskall-Wallis test for rs960603.
Discussion
Even though it was not the main goal of our study, the availability of whole-genome methylation information allowed us to calculate epigenetic age for the subjects in our sample. DNA methylation levels have been proposed as biomarkers of aging since chronological age correlates with DNA methylation in most human tissues and cell types.Reference Horvath 34 , Reference Horvath and Raj 39 In addition, this measure has predictive value for all-cause mortality, and estimated from blood-extracted DNA, correlates with measures of cognitive and physical fitness in 70 year-olds.Reference Marioni, Shah and McRae 40 , Reference Marioni, Shah and McRae 41 Although we found a 20-year or greater difference between chronological and epigenetic age in all our samples, no difference was observed between the SCA and AD groups using Horvath’s DNAm age predictors. This is in contrast with the results in brain tissue of Levine et al,Reference Levine, Lu and Bennett 42 who found that postmortem DNAm age in the dorsolateral prefrontal cortex was associated with neuropathological variables and postmortem measures of cognitive decline among individuals with AD.
Nevertheless, our results are valuable in the context of studies dealing with long-lived individuals. As in other studies of such individuals, epigenetic age was lower than chronological age; however, the extent of this difference was markedly larger. McEwen et al studied DNA methylation in a sample from Nicoya,Reference McEwen, Morin and Edgar 43 a Costa Rican high longevity region. No difference was found in epigenetic age between Nicoyans and a sample of people of other regions of Costa Rica. However, they reported a difference of −6.9 years between epigenetic age and chronological age when analyzing the whole sample and −12.7 years in centenarians. These results are consistent with the report of an epigenetically younger age in Hispanic populations.Reference Horvath, Gurven and Levine 44 Similar results have also been found in other populations. Nonagenarians from Sydney, Australia showed a difference of −9.56 years between epigenetic and chronological age.Reference Armstrong, Mather and Thalamuthu 45 Furthermore, centenarians from an Italian cohort were 8.6 years younger than their chronological age, while their offspring have a lower epigenetic age than age‐matched controls.Reference Horvath, Pirazzini and Bacalini 46 More research is required to understand the reasons for these differences.
When comparing whole-genome methylation between the SCA and AD groups, we found 11 differentially methylated regions. Some of them have been previously reported to be associated with AD or related to pathways associated with AD. The gene known as peptidase M20-domain-containing protein 1 (PM20D1) was found to be hypomethylated in the SCA sample. This gene has been associated with AD, and additional evidence shows a correlation between DNA methylation, RNA expression, and genetic background; it is both methylation and an expression QTL.Reference Sanchez-Mut, Heyn and Silva 37 , Reference Sanchez-Mut, Glauser and Monk 38 , Reference Heyn, Moran and Hernando-Herraez 47 , 48 Recent evidence indicates that PM20D1 expression might provide a potential cellular defense mechanism against AD.Reference Sanchez-Mut, Heyn and Silva 37 In vitro assays showed an increased PM20D1 expression in neuroblastoma cells treated with reactive oxygen species (ROS) and amyloid-β (Aβ), emulating an AD model. Meanwhile, an analysis of brain tissue in a mouse model of AD, which develops AD-related pathologies with age, such as amyloid plaques, astrogliosis, and learning deficits, showed increased PM20D1 expression in the frontal cortex at symptomatic stages of the disease in comparison with presymptomatic stages and controls. In addition, manipulation of PM20D1 levels showed that overexpression of PM20D1 reduced cell death and decreased Aß levels in vivo and in vitro assays. Additionally, a recent study detected association of a DMR (in peripheral blood) in the PM20D1 gene with both the transition between cognitive-normal to mild cognitive impairment, and the rate of cognitive decline in Alzheimer’s disease.Reference Li, Vasanthakumar and Davis 49
Sánchez-Mut et al. found a significant increase of PM20D1 repression in AD when compared with nondemented individuals by analyzing DNA methylation and RNA expression. The repression occurs by a CCCTC-binding-factor-mediated chromatin loop that depends on an AD-associated haplotype. Genetic analysis of human brain cortex samples has shown an allele-dependent correlation between the haplotype of two mQTL associated SNPs, rs708727 and rs960603, and PM20D1 promoter methylation. As expected, PM20D1 expression was inversely correlated with the methylation level of its promoter.Reference Sanchez-Mut, Heyn and Silva 37 In the present study, we also found a significant change in average methylation at the PM20D1 DMR according to the genotype for rs708727, while for rs960603 there is no significant effect. Similarly, a strong effect of rs708727 allele dosage on methylation, but not for rs960603, was recently reported by Wang et al.Reference Wang, Chen and Readhead 50 Another recent study by Sanchez-Mut et alReference Sanchez-Mut, Glauser and Monk 38 found a much stronger correlation with methylation level at PM20D1 CpG sites for rs708727 than for rs960603. The minor allele frequencies for these SNPs in the central valley of Costa Rica are 0.41(A) for rs708727, and 0.38(G) for rs960603 (unpublished data from G. Chavarria-Soley and H. Raventós, based on whole genome sequences of 51 unrelated individuals). In our study, when we performed a stratified comparison per genotype of methylation levels between SCA and LOAD for both SNPs no significant differences were found. However, a tendency toward higher methylation levels in the LOAD sample can be seen for both SNPs, which could be explored with a larger sample size.
A recent longitudinal study of individuals affected with AD has provided evidence in favor of a hypothesis that proposes a change in the methylation status of PM20D1 (possibly coupled with a change in its expression) throughout the AD pathology.Reference Wang, Chen and Readhead 50 At the initial stages of the disorder of mild cognitive impairment, the gene is hypomethylated in comparison to controls. This hypomethylation could be coupled with a higher expression of the gene, which could play a protective role. In later stages of the disease, the gene is then hypermethylated, as has been reported in several studies.Reference Zhang, Silva and Young 16 , Reference Sanchez-Mut, Heyn and Silva 37 Wang et al propose that the turning point for the change in methylation of the gene is at approximately 78–79 years old. Our results fit with this hypothesis, since we observe hypermethylation of the gene in affected individuals, and they are all over 90 years of age and present with moderate or severe dementia. From the point of view of SCA group, the hypomethylation of PM20D1 we detected could play a protective role against the development of LOAD.
Interestingly, PM20D1 has been associated with obesity and diabetes, which are risk factors for AD.Reference Long, Svensson and Bateman 51 - Reference Profenno, Porsteinsson and Faraone 53 PM20D1 lies within the Parkinson’s disease 16 (susceptibility) locus on chromosome 1, which has previously been associated with Parkinson’s disease.Reference Simon-Sanchez, Schulte and Bras 54 In addition, PM20D1 has been reported to be differentially methylated in individuals with obesity and multiple sclerosis patients.Reference Feinberg, Irizarry and Fradin 55 , Reference Maltby, Lea and Sanders 56 The protective role of PM20D1 may be explained by the fact that it has previously been shown to activate mitochondrial uncoupling, which promotes neuronal survival because it contributes to the adaptive responses to bioenergetic and oxidative stressors.
The gene LHX6 encodes a member of a protein family that contains the LIM domain, a unique cysteine-rich zinc-binding domain. The protein is a transcription factor involved in embryogenesis and in the expression of a subset of genes involved in interneuron migration and development.Reference Lavdas, Grigoriou and Pachnis 57 , Reference Zhang, Gutierrez and Li 58 The gene is highly expressed in neural crest-derived mesenchyme cells.Reference Grigoriou, Tucker and Sharpe 59 In our SCA sample, this region was found to be hypermethylated.
Another one of the differentially methylated genes, CERS3 is a member of the ceramide synthase family of genes, this region was found to be hypermethylated in our SCA sample. This type of enzyme regulates sphingolipid synthesis by catalyzing the formation of ceramides from the sphingoid base and acyl-CoA substrates. Several lines of evidence suggest that there is a causal link between ceramide or sphingolipids levels and neurodegenerative diseases such as AD.Reference Ben-David and Futerman 60 However, CERS3 is the only one out of the six-ceramide synthases that are not expressed in brain tissue.Reference Becker, Wang-Eckhardt and Yaghootfam 61 Its expression has been reported especially in testis and skin.Reference Riebeling, Allegood and Wang 62 , Reference Mizutani, Kihara and Igarashi 63 In addition, the gene is associated with ichthyosis.Reference Radner, Marrakchi and Kirchmeier 64
The gene vtRNA2-1, also known as nc886, encodes a non‐coding RNA that represses PKR, a double‐stranded RNA-dependent kinase, involved in tumor suppression.Reference Jeon, Lee and Lee 65 , Reference Jeon, Johnson and Lee 66 This gene is often hypermethylated and repressed in cancers.Reference Fort, Mathó and Geraldo 67 - Reference Lee, Park and Lee 69 Romanelli et alReference Romanelli, Nakabayashi and Vizoso 70 showed that the vtRNA2-1 region is variably maternally imprinted, namely, it has allele-specific methylation and shows variable levels of methylation levels among tissues. In our sample, this region was found to be hypermethylated in the SCA group.
For the remaining four genes and pseudogenes there is little information regarding their function. The protein phosphatase 1 regulatory inhibitor subunit 2 Pseudogene, PPP1R2P1, is a pseudogene, which we found to be hypomethylated in our SCA sample. Evidence shows that PPP1R2P1 is expressed, but its function remains unknown.Reference Korrodi-Gregório, Abrantes and Muller 71 The gene PRSS22 encodes a member of the trypsin family of serine proteases. The enzyme is expressed in the airway epithelial cells in a developmentally regulated manner.Reference Wong, Yasuda and Madhusudhan 72 DOC2GP is a pseudogene expressed mainly in the heart, spleen, and thyroid, while the functionally uncharacterized C17orf98 is expressed in testis.Reference Fagerberg, Hallström and Oksvold 73
Regarding the correlation between blood and brain methylation in our analysis, it was very variable among the DMRs. The lowest correlation was −0.004, and the highest was 0.701. Three out of the six DMRs that contain genes present a correlation above 0.55. The DMR with the highest correlation contains neither a gene nor a pseudogene. A limitation in the determination of the correlations was that the BECon tool does not have information for all the CpG sites of our DMRs, so the average correlation between blood and brain was calculated excluding some sites from the DMR regions. Recently a comparison of methylation levels between blood and four regions of the brain in an independent cohort was performed for CpG sites specifically in PM20D1. Reference Wang, Chen and Readhead 50 Correlations were high and significant for all comparisons, with a range of correlation coefficients from 0.857 to 0.976. Therefore, it is safe to assume that the methylation changes we observed in the gene are also occurring in the brain. Current evidence suggests that in many cases blood is an appropriate sample for such studies, which has many implications including the possibility of biomarker detection for early diagnosis.
An important limitation of our study is the small sample size. Nevertheless, our confirmation of the previously reported role of PM20D1 in AD proves that valid results can be obtained with small samples. Besides, there is increasing recognition of the value of studying ancestrally diverse populations, and Latin American populations have been historically underrepresented in large-scale genomic and epigenomic studies.Reference Hindorff, Bonham and Brody 74 , Reference Peterson, Kuchenbaecker and Walters 75
Conclusions
In conclusion, our work adds evidence to suggest that long-lived individuals have a lower epigenetic age than their chronological age. This work also supports the possible role of PM20D1 in protection against AD, by showing differential methylation in blood of affected and unaffected individuals in a longer-lived population. We also confirmed the association between rs708727 genotypes and methylation levels in the gene’s promoter, which provides further evidence in favor of the influence of genetic factors on PM20D1 expression (which in turn may influence susceptibility to develop AD). Finally, we found other differentially methylated regions including genes involved in cell cycle regulation, embryogenesis, synthesis of ceramides, and migration of interneurons to the cerebral cortex. These genomic regions might play a role in AD and SCA, and merit further studies.
Acknowledgment
We thank the participants for their collaboration.
Funding Statement
This work received funding support from NIH Fogarty International Center & National Institute on Aging (grant: R21TW009258), Alzheimer’s Association, Universidad de Costa Rica (projects A4323 and B8377), and Sistema de Estudios de Posgrado from Universidad de Costa Rica. This study received partial funding from the “Instituto Nacional de Medicina Genomica.”
Disclosures
The authors have no conflicts of interest to disclose.
Author Contributions
Conceptualization: C.C.-V., L.M.-V., D.V., J.M.S., H.N., H.R., G.C.-S.; Data curation: C.C.-V., L.M.-V., D.V., A.D.G.-M.; Formal analysis: C.C.-V., J.J.M.-M., D.V., A.D.G.-M., G.C.-S.; Funding acquisition: J.M.S., H.N., H.R.; Investigation: C.C.-V., J.J.M.-M., L.M.-V., D.V., A.D.G.-M., J.M.S., H.N., H.R., G.C.-S.; Methodology: C.C.-V., J.J.M.-M., L.M.-V., D.V., A.D.G.-M., J.M.S., H.N., H.R., G.C.-S.; Project administration: L.M.-V., H.N., H.R., G.C.-S.; Supervision: C.C.-V., J.J.M.-M., G.C.-S.; Validation: J.J.M.-M.; Writing—original draft: C.C.-V., J.J.M.-M., L.M.-V., D.V., J.M.S., H.N., H.R., G.C.-S.; Writing—review and editing: C.C.-V., J.J.M.-M., L.M.-V., D.V., A.D.G.-M., J.M.S., H.N., H.R., G.C.-S.
