Epigenetic dysregulation of transposable elements in cognitive impairment and Alzheimer’s disease

Abstract

Aging and cognitive impairment increase the risk for Alzheimer’s disease (AD), and growing evidence suggests that transposable elements (TEs) in the genome play a role in aging and AD. The mechanisms of TE dysregulation in this context are unclear, but one possibility is that epigenetic changes, including DNA hypomethylation and/or reduced chromatin structure, underlie age- and AD-related TE activity. Therefore, the purpose of the present study was to generate a resource for studying TE epigenetics in aging and AD, and to use it to determine if epigenetically dysregulated TEs are related to age/AD-relevant clinical outcomes. We performed RNA-seq on peripheral blood samples from 45 healthy older adults, mild cognitive impairment (MCI) and AD dementia patients, and we observed a pattern of undulating TE transcript expression with MCI and AD, similar to previous reports. We then used whole-genome bisulfite sequencing (WGBS) and transposase-accessible chromatin sequencing (ATAC-seq) to characterize global DNA methylation and chromatin accessibility in the same subjects. We found that most TEs that were enriched/dysregulated in our RNA-seq data with MCI and AD could be found within hypomethylated and chromatin-accessible regions of the genome. These TEs included several that have been directly linked to inflammation and disease in humans, and they were related to cognitive/functional diagnosis, age, and biomarkers of inflammation and neurodegeneration in the subjects we studied. Collectively, these findings are consistent with the idea that epigenetic alterations may contribute to TE transcript dysregulation that plays an important role in aging, cognitive decline, and AD.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11357-025-01765-9.

Introduction

Advancing age is the primary risk factor for late-onset Alzheimer’s disease (AD) dementia. How age increases AD risk is poorly understood. However, brain aging involves declines in cognitive function, which may develop into mild cognitive impairment (MCI) that increases the risk of developing AD dementia [1, 2]. Brain aging and MCI are also characterized by similar, adverse biological events that are exacerbated in AD (e.g., the accumulation of tau and amyloid beta [Aβ]), and most AD cases are sporadic and age-related (i.e., not genetic) [3]. As such, identifying and understanding mechanisms of aging and MCI that contribute to age-related AD per se is an important research goal.

One key driver of brain aging, MCI, and AD dementia is inflammation, which is a central hallmark of aging. Systemic/peripheral inflammation is also closely related to neuroinflammation in the brain [4, 5], which is characterized by innate immune activation and neurotoxic, inflammatory cytokine production [2, 6]. The precise causes of age-related peripheral inflammation and neuroinflammation are unknown. However, recent studies suggest that transposable elements (TEs) may play an important role in this context [7–11], and as a result, clinical trials targeting TEs are underway [12]. TEs make up ~ half of the human genome, and they consist of non-coding, repetitive sequences that have accumulated throughout evolution, including primarily: (1) retrotransposons, such as long terminal repeats (LTRs), long interspersed nuclear elements (LINEs) and short interspersed nuclear elements (SINEs); and (2) DNA transposons [13]. Retrotransposons are of particular interest in aging/AD, because several (e.g., from the LINE1 family in humans) remain active, with the ability to “copy and paste” into different regions of the genome via reverse transcription and re-insertion into the genome [13, 14].

We and others have shown that transcripts from TEs, including LINEs, increase with aging and AD in multiple tissues such as brain and blood [9, 11], and recent data show that TE transcripts (in peripheral blood/cells) are reduced in MCI before a “storm” of TE dysregulation in AD [15]. This TE dysregulation could play a role in age- and AD-related inflammation, as TE transcripts can form immunostimulatory double-stranded RNA (dsRNA) and/or complementary DNA (cDNA) [14, 16–18]. Indeed, model organism studies show that TE transcripts contribute to aging via the activation of anti-viral inflammatory signaling [19–21], and that TE-derived dsRNA and cDNA are elevated in the aging/AD brain [11, 22]and in transgenic tau pathology models [23–25], consistent with the idea that TEs may contribute to age/AD-related inflammation [21, 26].

Importantly, because of their potential to cause these adverse cellular events, TEs are typically epigenetically suppressed via DNA methylation and heterochromatin. However, increasing evidence suggests that epigenetic changes (e.g., reduced DNA methylation and chromatin structure) may drive TE dysregulation with aging and AD [27–30]. For example, studies have shown that DNA methylation patterns are altered with aging, MCI, and AD dementia [31–33], and that TEs may be particularly impacted [29, 34]. Data also show that age- and disease-associated heterochromatin decondensation may be linked with TE dysregulation [30, 35, 36], and that epigenetically modulating TEs can extend lifespan [37, 38].

Despite the existing evidence, the epigenetics of age- and AD-related TE dysregulation have not been comprehensively characterized in humans, especially at the whole-genome level. Therefore, the purpose of the present study was to: (1) generate a resource for those interested in TE dynamics/epigenetics in aging and AD; and (2) use it to determine if epigenetically dysregulated TEs are related to age/AD-relevant clinical metrics. To do this, we performed a multi-omics analysis on peripheral blood/cells, an accessible sample type that may reflect systemic changes with aging and AD [39, 40]. Specifically, we used transcriptomics to profile TE transcript differences in samples from asymptomatic, healthy older adults and adults with either MCI or dementia due to possible AD, and we characterized genome-wide DNA methylation and chromatin accessibility alterations that may explain these differences. We found that most TE transcripts with increased expression could also be traced to genome regions with reduced methylation and chromatin, and that transcripts from the most epigenetically dysregulated TEs were related to cognitive function and markers of systemic inflammation and neurodegeneration.

Materials and methods

Participants

Samples and data for the present analyses were obtained from a study (“Bio-AD”) conducted at the University of Colorado Anschutz Medical Campus Alzheimer’s and Cognition Center. All analyses were performed on samples from a carefully selected group of 45 older adults (aged 53–83) including healthy, asymptomatic middle-aged/older adults (ASM, n = 11) and asymptomatic older adults (ASO, n = 14; to evaluate potential aging effects), as well as symptomatic older adults with MCI (n = 10) and dementia (n = 10) due to possible AD. Details on participants are presented under Results. All participants underwent a battery of cognitive function testing, health history assessments, neurological and physical examinations, and informant interviews. To be included as asymptomatic, participants had to be community dwelling older adults with no diagnosis of MCI or dementia, and no evidence of a neurodegenerative disease based on a neurological exam and an informant interview. Symptomatic participants were referred from the University of Colorado Memory Disorders Clinic, and severity (MCI or dementia due to possible AD) was adjudicated using established protocols [41–43]. Participants with major psychiatric disorders, non-AD neurological conditions (e.g., Parkinson’s disease), recent history of a focal brain lesion, substance abuse, significant medical illness or conditions that would interfere with cognitive testing were excluded from the Bio-AD study. All participants/cases were evaluated and agreed on by a board-certified neuropsychologist, board-certified behavioral neurologist, and clinical research coordinator. All participants provided informed consent under a protocol approved by the Colorado Multiple Institutional Review Board [44].

Cognitive assessments

The protocols used for all cognitive function tests have been described in detail elsewhere [41, 45, 46]. Briefly, testing included the Montreal Cognitive Assessment (MoCA, a key clinical tool for evaluating cognitive health and dementia status) and tests in the following subdomains of cognitive function as part of the Spanish English Neuropsychological Assessment Scales (SENAS): verbal episodic memory, spatial location, executive function, and language/semantic knowledge [41, 46]. Item response theory composite scores were used for each of the SENAS domains. Verbal episodic memory was measured because of its importance for memory formation that is impacted in early AD [47], and verbal memory composite scores were determined by a multi-trial list-learning measure as previously described [41]. Spatial composite scores were calculated based on the spatial localization scale, which assesses the ability to observe and replicate two-dimensional spatial relationships [45]. Executive function scores were determined with working memory (visual and digit backward recall, list sorting), and fluency testing. Semantic composite scores were measured using a nonverbal picture association and verbal object naming tasks, as previously described [46].

Biomarker assessment

Blood samples were collected via venipuncture, centrifuged at 22 °C for 15 min at 1500 × g, and plasma was removed. Isolated plasma was centrifuged at 4 °C for 10 min at 2200 × g and stored at − 80 °C for subsequent analyses. Biomarker analyses of glial fibrillary acidic protein (GFAP) and neurofilament light chain (NfL, a marker of neuronal damage and neurodegeneration) were quantified using the Quanterix ultra-sensitive single molecule array (SIMOA) immunoassay and SR-X Analyzer with manufacturer-supplied antibody kits. Peripheral markers of inflammation, including vascular cell adhesion molecule 1 (VCAM1), interferon gamma-inducible protein 10 (IP10), macrophage inflammatory protein 1-alpha (MIP1ɑ), C-reactive protein (CRP), macrophage-derived chemokine (MDC), and intercellular adhesion molecule 1 (ICAM1) were measured on Meso Scale Diagnostics panels on a MESO QuickPlex SQ 120 and quantified with Discover Workbench 4.0 software.

Nucleic acid isolation and sequencing

RNA-seq

RNA was recovered from separate whole blood samples collected in PaxGene blood collection tubes (Qiagen) using PaxGene-specific RNA isolation kits (Qiagen), and total RNA-seq was performed as previously described. Briefly, RNA libraries were generated using Tecan Globin/Ribo depletion kits and sequenced on an Illumina NovaSEQ 6000 platform to generate > 40 million 151-bp, paired-end reads per sample. Differential gene expression was analyzed using standard techniques as previously described [9–11, 48, 49]. Reads were trimmed and quality filtered with fastp [50], aligned to the human genome (UCSC, hg38) using the STAR aligner [51], and gene and TE counts were generated with the TEtranscripts program [52]TEcount feature, which algorithmically estimates TE abundance and generates individual TE transcript type counts accounting for multi-mapping. Differentially expressed genes and TEs were then identified using DESeq2 for R [53].

Whole-genome bisulfite sequencing (WGBS)

Because this was a retrospective study, sequencing was performed on available samples (i.e., whole blood in PaxGene tubes for RNA-seq, above). Identical DNA samples were not available. Therefore, to profile methylation differences, WGBS was performed on genomic DNA extracted from CryoStor-frozen white blood cells (collected at the same time as whole blood) using the Zymo DNA Miniprep Plus kit. Samples with low DNA concentrations or residual contamination were re-processed using the Zymo Clean & Concentrate Kit. For WGBS, genomic DNA was subjected to bisulfite conversion, and libraries were constructed using the Swift Accel-NGS Methyl-Seq DNA library kit, then sequenced at 60 Gb (~ 200 million reads) per sample on an Illumina NovaSEQ 6000 (151-bp, paired-end reads). WGBS data were processed and analyzed using the MethPipe program as previously described [54]. Briefly, reads were mapped using the abismal program, and MethPipe was used to remove duplicate reads, calculate methylation levels at individual genomic loci, estimate bisulfite conversion rates, and identify hypo- and hyper-methylated regions of the genome. Within the MethPipe workflow, differentially methylated regions were calculated using hyper/hypomethylated regions called by a hidden Markov model to determine average site-specific methylation level and variance, and differential methylation scores for individual sites were generated using a modified Fisher’s exact test as previously described [55]. The resulting list of differentially methylated regions was then intersected with a gtf/bed file of annotated repetitive DNA sequences in the hg38 human genome (obtained from www.repeatmasker.org) to identify the TEs located in these regions using Bedtools [56]. Only differentially methylated regions with at least one significant CpG site were retained for downstream analyses.

Assay for transposase-accessible chromatin using sequencing (ATAC-seq)

To characterize global chromatin accessibility, ATAC-seq libraries were prepared from the same samples used for WGBS using the Active Motif ATAC-seq library kit as previously described [57, 58]. ATAC-seq library preparation requires live cells for nuclei isolation; therefore, cryopreserved samples were rapidly thawed at 37 °C for 2 min, gently mixed to resuspend, and intact nuclei were isolated (~ 75,000/sample) and subjected to tagmentation to select transposase-accessible DNA fragments (i.e., in open chromatin regions). Tagmented DNA fragments were purified and then amplified via PCR, then sequenced on an Illumina NovaSEQ 6000 to generate > 50 million 151-bp, paired-end reads per sample. ATAC-seq reads were trimmed and filtered using fastp, and aligned to the human genome with Bowtie2 [59]. ATAC peak identification and differential enrichment analyses were performed with MACS2, and differentially accessible peaks that met an adjusted p-value threshold of 0.01 were retained for downstream analyses. Bedtools [56]was used as above for WGBS data to intersect ATAC peaks and TE loci (by intersecting peaks with a RepeatMasker hg38 annotation) to identify TEs in open chromatin regions.

Statistical analyses

Clinical data were tested for normality using GraphPad Prism. Normally distributed data were analyzed using a one-way ANOVA followed by Tukey’s multiple comparisons test. Differences among groups that were not normally distributed were assessed using the Kruskal–Wallis test and two-step Benjamini, Krieger and Yekutieli correction for multiple comparisons. Differential expression of genes and TE transcripts was quantified with DESeq2 [53] using all genes and TEs together (to estimate TE differences more conservatively). Hierarchical clustering analyses were performed using Morpheus (Broad Institute). Regression analyses for TE transcript predictors of cognitive functional/diagnoses, age, neurodegeneration and inflammatory biomarkers were performed in R (MASS and tidyverse packages) using forward stepwise regression models with AIC criteria for TE predictor inclusion. To assess whether selected TE predictors outperformed random predictors of the same number, a permutation-based benchmarking approach was used in which 1000 replicate models with an equal number of randomly selected TEs were generated. R² values from the selected and random models were compared as the proportion of random models with R² greater than or equal to that of the selected model.

Results

Participant characteristics and TE transcript levels

To evaluate the roles of aging and disease in TE dysregulation, subjects were grouped initially as ASM, ASO, MCI, and AD. All major anthropometric characteristics (e.g., body mass index, blood pressure) were similar among these groups (Table 1). However, as expected, global cognitive function, as reflected by total MoCA scores, was lower in MCI compared to both ASM and ASO subjects, and even lower in AD dementia vs. asymptomatic subjects (Fig. 1A). In addition, MCI and AD dementia subjects had greater plasma levels of NfL (Fig. 1B) and GFAP (Fig. 1C), important clinical biomarkers of neurodegeneration and neuroinflammation, respectively [41, 60–62]. NfL was also increased with aging alone (i.e., in ASO vs. ASM), in line with the idea that NfL may be a marker of cognitive aging as well as neurodegeneration [63]. Other circulating markers of inflammation were not different among groups, although most were somewhat greater in older subjects (Table 2). As such, subject groups were largely similar with the exception of lower cognitive function and increased neurodegeneration biomarkers in MCI/AD, consistent with what others have reported.

Table 1.

Subject characteristics by group

	ASM	ASO	MCI	AD dementia
Age (years ± SD)	61 ± 4.8	76 ± 3.6*	73 ± 6.0*	70 ± 8.5*#
Sex	5 M/6 F	6 M/8 F	5 M/5 F	5 M/5 F
Education (years)	16.6 ± 2.4	16.5 ± 2.7	16.7 ± 2.6	15.8 ± 2.2
Body mass index (kg/m2)	25.4 ± 3.1	27.3 ± 3.7	28.5 ± 5.4	25.4 ± 3.1
Systolic BP (mmHg)	119 ± 9.3	127 ± 8.9	129 ± 12.7	133 ± 12.5
Diastolic BP (mmHg)	71 ± 8.2	69 ± 5.1	70 ± 9.1	73 ± 7.6
APOE genotype (% with ≥ 1 ɛ4 allele)	3 (30%)†	2 (16.7%)†	6 (60%)	6 (60%)

Data shown as mean ± SD. ASM: middle-aged asymptomatic, ASO: older asymptomatic, MCI: mild cognitive impairment, AD: Alzheimer’s disease. *p < 0.05 compared to ASM, ^#p < 0.05 compared to ASO, ^†missing values excluded.

Fig. 1.

Lower cognitive function and elevated biomarkers of neurodegeneration and neuroinflammation are associated with TE transcript dysregulation in MCI and AD. A Total Montreal Cognitive Assessment (MoCA) scores, B plasma concentrations of neurofilament light chain (NfL), and C plasma glial fibrillary acidic protein (GFAP) in all subjects by group; D Log2-fold differences in expression of the 4 main TE transcript types among groups. ASM, middle-aged asymptomatic (n = 11); ASO, older asymptomatic (n = 14); MCI, mild cognitive impairment (n = 10); AD, Alzheimer’s disease/dementia (n = 10). *p < 0.05 vs. ASM, ^#p < 0.05 vs. ASO

Table 2.

Circulating markers of inflammation

	ASM	ASO	MCI	AD dementia
VCAM (ng/mL)	319.9 ± 67.5	422.9 ± 112.3	404.3 ± 102.3	345.5 ± 104
IP10 (pg/mL)	464.7 ± 264.7	542.9 ± 235.9	463.8 ± 260.9	371.3 ± 125
MIP1ɑ (pg/mL)	13.77 ± 1.37	19.01 ± 8.11	16.3 ± 3.9	13.4 ± 3.16
CRP (μg/mL)	3.92 ± 7.71	2.33 ± 1.74	3.52 ± 5.93	0.739 ± 0.386
MDC (pg/mL)	838.1 ± 160	848.5 ± 270.3	820.6 ± 481	831.7 ± 156.8
ICAM1 (ng/mL)	278.8 ± 41.7	311.6 ± 61.3	301 ± 69.8	284.7 ± 69.6

Data shown as mean ± SD. ASM, middle-aged asymptomatic; ASO, older asymptomatic; MCI, mild cognitive impairment; AD, Alzheimer’s disease; VCAM1, vascular cell adhesion molecule 1; IP10, interferon gamma-inducible protein 10; MIP1⍺, macrophage inflammatory protein 1-alpha; CRP, C-reactive protein; MDC, macrophage-derived chemokine; and ICAM1, intercellular adhesion molecule 1.

Aging and AD have been linked with transcriptomic changes, including to TEs, in both the brain and peripheral tissues like blood—an accessible sample that may provide insight into systemic changes [9, 11, 21, 39, 40, 64, 65]. Therefore, to characterize differences in TE dynamics among our subject groups (for which brain samples were not available), we performed total RNA-seq and differential expression analyses on peripheral blood samples. Gene and TE transcript differences by group were modest overall, perhaps because of our choice to use total RNA-seq (which spreads reads across genes and non-coding sequences). However, overall gene expression patterns pointed to age- and disease-associated differences in biological processes related to innate immune activity and cellular stress responses (Supplementary Fig. S1). We also observed a pattern of TE dysregulation by group/diagnosis (Fig. 1D), including undulating TE transcript expression with MCI/AD, particularly in DNA transposons, LINEs and SINEs, similar to the pre-AD “retrotransposon storm” others have reported [15]. We did not observe robust subject/TE transcript clustering within groups, or an overrepresentation of group/diagnosis type in subjects with high TE transcript expression (Supplementary Fig. S2). TE transcripts were also not strongly increased with aging (i.e., in ASO vs. ASM), but this could be because the age difference between groups was relatively small. Indeed, we and others have reported significant TE transcript increases with aging (across decades), and in AD [9, 11, 21, 64], both in the brain and peripheral tissues.

Given the above observations, we retained our subject groupings for downstream analyses with a focus on identifying the effects of aging vs. disease on TE epigenetics. Because TE transcript differences by group were modest and recent studies show that DNA hypomethylation and chromatin dysregulation may underlie TE dysregulation with aging/AD [26, 29, 66–68], in these analyses, as described in the following sections, we proceeded to: (1) evaluate the extent to which TE transcript patterns in our subjects aligned with epigenetic differences; and (2) determine if these differences could be used to identify “key” TEs in the context of aging and AD dementia (i.e., beyond the patterns observed in RNA-seq data).

TE transcripts that increase with aging and MCI/AD may originate from hypomethylated DNA

To evaluate DNA methylation as comprehensively as possible (including in genomic regions that are not captured by methylation arrays), we performed WGBS. With aging alone (i.e., in ASO vs. ASM), we found that ~ 42% of TEs that were increased in our RNA-seq data (i.e., positive Log2 Fold Difference) also had loci in hypomethylated genome regions (Fig. 2A, next page). On a percent basis, these TEs were mostly LTRs, and the absolute number of hypomethylated regions coinciding with transcriptionally enriched TEs was also highest for LTRs (Fig. 2B, C). The number of TEs increased in expression and associated with hypomethylated DNA was similar in MCI (~ 45% of elevated TEs), and most were LTRs and DNA transposons (Fig. 2D–F). However, in AD vs. ASM, transcriptionally enriched TEs that were also associated with hypomethylated DNA were increased, with nearly ~ 65% having loci in hypomethylated genome regions (Fig. 2G). Of these, the most abundant TEs on a percent basis were LTRs, DNA transposons and LINEs (Fig. 2H, I), but by absolute number, LINEs and SINEs occurred most often in hypomethylated regions. These broad patterns of TE hypomethylation were similar but somewhat more exaggerated in AD vs. MCI (Fig. 2J–L). Interestingly, among all groups, the overall percentage of genome mobilization (i.e., base pairs within hypomethylated regions) was similar, suggesting that age/AD-related TE dysregulation could be due to changes in methylation patterns rather than global hypomethylation (Supplementary Fig. S3). Overall, these data suggest that similar degrees of TE hypomethylation (particularly for LTRs, the most common type of TE) occur with aging and MCI, but that AD dementia is associated with an exaggeration of these events and further dysregulation of other key TE types (i.e., LINEs and SINEs).

Fig. 2.

Transcriptionally dysregulated TEs are hypomethylated with aging, MCI, and AD. (Left panels) Venn diagrams showing TE transcripts that are increased in RNA-seq data and associated with differentially methylated DNA regions; (middle panels) pie charts showing the relative composition of enriched TEs with loci in hypomethylated DNA regions by type; (right panels) bar graphs showing abundance of TEs by number of occurrences in hypomethylated DNA regions. A–C ASO vs. ASM; D–F MCI vs. ASM; G–I AD vs. ASM; and J–L AD vs. MCI

TE transcripts that increase with aging and MCI/AD may also arise from chromatin-accessible DNA

Existing data indicate that chromatin dysregulation, including increased histone acetylation (which generally increases transcriptional accessibility), occurs in aging, AD, and preclinical AD models [27, 69]. To determine the extent to which chromatin differences might be associated with TE dysregulation in our subjects/samples (i.e., in addition to hypomethylation), we next performed ATAC-seq. In ASO vs. ASM, we found that > 75% of TEs with increased expression in our RNA-seq data also had loci in chromatin-accessible genome regions (i.e., within ATAC-seq peaks, Fig. 3A, next page). By major type, most of these enriched, ATAC-accessible TEs were LTRs (Fig. 3B), although DNA transposons were also frequently in chromatin-accessible regions (Fig. 3C). These numbers and patterns were similar in MCI vs. ASM, with ~ 73% of the TEs with increased expression having loci in open chromatin regions, and most being LTRs (Fig. 3D–F). Similar to our observations regarding DNA methylation, AD dementia was associated with an even greater number of increased, chromatin-accessible TEs (~ 85–90% in AD vs. ASM and AD vs. MCI, respectively, Fig. 3G–J, Supplementary Fig. S4). Among these, the most common type of TE on a percent basis was LTRs, but by absolute number, LINEs and SINEs occurred most frequently in chromatin-accessible regions. Similar to our observations of hypomethylation, the proportion of the genome in these chromatin-accessible sites was similar across groups, but slightly greater in ASO and AD subjects, consistent with the reported loss of heterochromatin in aging and AD [28, 29](Supplementary Fig. S3). Collectively, these results suggest that, similar to DNA hypomethylation, increased chromatin accessibility occurs across most TEs with aging and MCI (as reflected by numerous chromatin-accessible LTRs), but that AD dementia is associated an increase in accessibility of additional TE classes (e.g., LINEs and SINEs), which is consistent with current evidence of a role for LINEs and Alu/SINE elements in neurodegeneration [68].

Fig. 3.

Transcriptionally dysregulated TEs are chromatin-accessible with aging, MCI, and AD. (Left panels) Venn diagrams showing TEs that both increased in RNA-seq data and were associated with differentially chromatin-accessible DNA regions; (Middle panels) pie charts showing the relative composition by type of enriched TE transcripts that may be located within open chromatin regions; (right panels) bar graphs showing abundance of TEs by number of occurrences in chromatin-accessible DNA regions. A–C ASO vs. ASM; D–F MCI vs. ASM; G–I AD vs. ASM; and J–L AD vs. MCI

Epigenetically dysregulated TE transcripts are related to age- and AD-relevant clinical metrics

Because both aging and disease contribute to TE dysregulation, we next examined the relative contributions of aging, MCI, and AD dementia to TE hypomethylation and chromatin accessibility, and we used this as a framework to determine if epigenetically dysregulated TEs might be particularly related to clinical outcomes. First, we examined our WGBS data, and we found 14 TEs that were dysregulated (increased in expression and associated with hypomethylated DNA) with older age, MCI, and AD dementia (Fig. 4A, next page). LTRs made up the largest proportion of these TEs (Supplementary Fig. S4), in line with recent evidence of a key role for LTR transcripts in aging and disease [70–73]. Aside from these central TEs, however, most methylation-dysregulated TEs (317) were associated with AD dementia (Fig. 4A). Of these, LINEs, LTRs, and DNA transposons made up ~ 30% each (Supplementary Fig. S4). Interestingly, many fewer hypomethylated TEs were associated with aging and cognitive decline/MCI but not AD dementia (Supplementary Table 1) and aging alone (Supplementary Table 2). Next, we examined our chromatin accessibility data, and we found similar relative contributions of age and cognitive impairment to chromatin-associated TE dysregulation (Fig. 4B), with 44 elevated TE transcripts that had ATAC-accessible loci common to older age and disease. Of these TEs, > 66% were LTRs (Supplementary Fig. S5). As in our WGBS data, most TEs that were increased in expression and chromatin dysregulated (472) were AD-related but not associated with aging or MCI alone. These comprised ~ 25% LINEs, ~ 38% LTRs, and ~ 31% DNA transposons (Supplementary Fig. S5), whereas few individual TEs were associated with age and/or MCI alone (Supplementary Tables 3 and 4). Taken together, these data suggest that hypomethylation and chromatin-associated TE dysregulation may occur in parallel, and that these events, although present in aging, are exacerbated in AD dementia.

Fig. 4.

Epigenetic TE dysregulation with aging, MCI, and AD. A Venn diagram showing TEs increased in RNA-seq data with loci in hypomethylated regions of the genome in WGBS data; B Venn diagram showing TEs increased in RNA-seq data with loci in chromatin-accessible genome regions

Finally, to determine if epigenetically dysregulated TE transcripts might be clinically relevant, we identified the most increased/accessible TEs in both our WGBS and ATAC-seq analyses, and we evaluated their potential to predict clinical phenotypes. Our rationale was that these particularly epigenetically dysregulated TEs might provide insight beyond their modest changes in RNA-seq data. We found 13 TEs that were enriched (increased) with aging/AD in our RNA-seq data and had loci in both hypomethylated and open chromatin regions of the genome (Fig. 5A, next page, and Supplementary Tables 5–6, Supplementary Fig. S6). Many of these TEs have been implicated in gene regulation, cancer, and other inflammatory conditions (e.g., LTR51, MER4D, MER81) [74–77], but most notable among them were L1HS and LTR5Hs. L1HS is a LINE that is among the only retrotransposition-competent TEs in humans, and LTR5HS, a prominent HML-2 human endogenous retrovirus (HERV), has been directly linked to inflammation and disease in humans [19, 21]. Interestingly, only one elevated TE transcript was unique to the WGBS data: MER92C (Fig. 5A), an ERV that can form immunogenic dsRNA and is activated in in vitro cancer models [78]. To determine if the 13 particularly epigenetically dysregulated TEs might be clinically relevant, we used multiple regression to relate them to subject phenotypes (Fig. 5B). For each clinical phenotype/metric, forward stepwise regression models based on these 13 TEs yielded R² values greater than those for the same number of randomly sampled TEs used to predict the same outcomes via the same regression approach. This was true for clinical diagnosis/group, age, plasma markers of neurodegeneration and inflammation, including plasma NfL and GFAP concentrations, as well as global and two subdomains of cognitive function (MoCA, verbal episodic memory, spatial location). Additionally, although peripheral markers of inflammation were not different among groups (i.e., in Table 2), levels of these 13 particularly epigenetically dysregulated TE transcripts were predictive of circulating markers of inflammation and vascular injury associated with cognitive decline, such as VCAM1 [79], IP10 [80], and MIP1⍺ [81], and of systemic markers of inflammation like CRP. Collectively, these findings suggest that the epigenetic dysregulation of TEs may play a particularly important role in age- and AD-related cognitive dysfunction.

Fig. 5.

TE transcripts associated with age- and AD-related hypomethylation and chromatin dysregulation are predictive of age, neurodegeneration and inflammatory biomarkers, and cognitive function. A Venn diagram showing overlap and 13 most dysregulated TEs (i.e., those in both ATAC-seq and WGBS data). B Comparison of R² values for regression models based on transcript levels of the 13 most dysregulated TEs vs. R.² values for regression models generated by randomly sampling the same number of other TEs 1000 times to predict: group/diagnosis; age, Montreal Cognitive Assessment (MoCA) scores; plasma neurofilament light chain (NfL); plasma glial fibrillary acidic protein (GFAP); circulating inflammatory factors (ICAM1, IP10, MCP1, MDC, MIP1ɑ, VCAM1); and verbal episodic memory (VM) spatial localization (SpLoc), semantic memory, and executive function cognitive testing scores. *p < 0.05

Discussion

Aging and AD are characterized by both chronic, systemic inflammation and neuroinflammation in the brain. The exact upstream mechanisms that lead to these inflammatory processes are not fully understood, but growing evidence shows that TE transcripts are systemically dysregulated with aging and AD, and may stimulate inflammation in various tissues, including the brain. As such, TE dysregulation could play a central role in brain aging and AD, and the loss of epigenetic control of TEs could be an important, upstream contributor to neurodegeneration. Here, we have presented a resource profiling TE epigenetics in older and MCI/AD subjects, and we have tested the idea that epigenetically dysregulated TEs may be particularly important in these subjects. Our primary findings are that most TE transcripts that are increased in expression with aging, MCI, and AD dementia can be traced to regions of the genome that are hypomethylated and have open chromatin, making the DNA at these loci more transcriptionally accessible. We also identify a subset of these TEs that are especially epigenetically dysregulated and show that transcripts from these TEs are related to clinical diagnosis, age, cognitive function, and clinical markers of neurodegeneration. Our conclusions are limited by the retrospective nature of the study, including the fact that we used biobanked peripheral blood/cells for our analyses, but our findings suggest several important, general ideas regarding TE dynamics in aging and AD, as described in the following sections.

Cognitive decline with aging can lead to MCI, which increases the risk for AD dementia [82, 83], but monitoring this progression is challenging. Therefore, to characterize subjects and evaluate the effects of aging and disease on TE dynamics in the present study, we used established metrics of cognitive function and neurodegeneration to compare groups of asymptomatic middle-aged and older adults, those with MCI and those with dementia due to likely AD. In these participants, the neurodegeneration and neuroinflammation biomarkers NfL and GFAP were increased peripherally with MCI and AD. These observations are in line with current data, as NfL is a biomarker of neurodegenerative diseases including AD [84–86], and it can predict progression from MCI to AD⁶². We also observed an increase in NfL in older asymptomatic participants even though they were cognitively normal, consistent with the idea that age-related NfL increases may predict future disease [63, 87]. GFAP is a classic marker of reactive astrogliosis, and recent studies also show that plasma GFAP is increased in AD and other neurodegenerative diseases [60, 61, 88], may increase earlier than neuron-specific markers like NfL [89, 90], and is closely tied to amyloid deposition and AD severity [88]. Thus, the characteristics of the research participants studied here are consistent with what others have reported for cognitive aging, MCI and AD, and with a role for inflammation/neuroinflammation in neurodegeneration. We therefore used these groupings in our downstream multi-omics analyses as a framework for identifying aging effects (ASO vs. ASM), aging + cognitive decline effects (MCI vs. ASM), and aging + AD dementia effects (AD vs. ASM), which we dissected by intersecting RNA and epigenetics patterns. However, we note that this is only one approach, and future studies could leverage the multi-omics data we have generated to evaluate gene and/or TE expression differences across aging and AD using more granular, integrated approaches. For example, individual-level analyses could be performed in which data types are correlated to each other across all subjects to determine the extent to which epigenetic differences track with individual TE (and gene) expression differences. Such analyses could be helpful in efforts to understand the heterogeneity of cognitive aging, at least as it relates to peripheral immune cell epigenetics.

Importantly, age-related cognitive decline can begin in late middle age, and when MCI or AD dementia begin to develop, aging and disease processes may interact, including at the epigenetic/transcriptomic level [91, 92]. To examine the effects of age and symptomatic disease on TE dysregulation in the present study, we first assessed age- and MCI/AD dementia-related differences in TE transcripts via whole blood RNA-seq. Although TE transcripts reportedly increase with aging and AD [9, 11, 21, 64], we did not find a strong age-related increase in in older vs. middle-aged asymptomatic subjects, and gene expression differences among groups were also modest. This could have been due to our choice to perform total RNA-seq rather than poly(A)/mRNA-seq. We performed total RNA-seq to more broadly capture TE transcripts, which are often found in intergenic regions or introns that are not retained during mRNA processing; however, this process also dilutes reads, increases variance, and can make it more difficult to detect individual significant differentially expressed transcripts [93, 94]. It also is possible that the sample type we used (whole blood in PaxGene collection tubes), which is what we had access to for this retrospective study, is inherently noisy at the transcriptome level. Indeed, we did not deconvolve gene/TE expression patterns to account for cell type contributions, which could play a role in group differences (tools for doing so can be found in several recent publications [95, 96]). Still, we did find that overall TE transcript expression, particularly of SINEs, undulates with MCI/AD dementia (lower at first but increasing in later stages of symptomatic disease), which is consistent with previous evidence [15]. Next, using WGBS, we found that most TE transcripts that were increased with aging could be traced to hypomethylated DNA regions, and that these TEs were largely DNA transposons and LTRs. These data are perhaps not surprising, as LTRs are the most common type of TE. Interestingly though, most TE transcripts that were increased in MCI and AD dementia were also hypomethylated, but these were more enriched for other TE types—e.g., LINEs (which have many copies in the human genome) in MCI, and LINEs and SINEs in AD. Thus, our observations do not simply reflect the expected patterns of TE distribution throughout the genome (i.e., LTR and LINE overrepresentation), but rather suggest particular patterns of epigenetic dysregulation of TEs in aging/MCI/AD dementia.

Our findings regarding TE methylation are important for two reasons. First, age- and disease-associated changes in DNA methylation are the foundation of many “biological clocks” that estimate biological age (vs. chronological age in years), and others have reported accelerated methylation age in AD [97–101], as well as age-related hypomethylation in TE-rich regions [29, 36, 102, 103]. Second, most prior studies on methylation in AD vs. control brains [98] have not evaluated TEs and were based on DNA methylation arrays that only capture coding regions, gene enhancers and CpG islands, whereas TEs are located throughout the genome [104, 105]. As such, our whole-genome methylation data (albeit in blood) extend on the existing literature with a new, comprehensive picture of TE epigenetics in aging/AD. Although not a goal of our study, these findings could be useful in future efforts to use methylation as a predictive biomarker (e.g., if TE methylation measures improve biological clock accuracy), but significantly larger sample sizes would be needed for sufficient power in such studies [99, 101]. It should be noted that a limitation of our study is that our RNA-seq and epigenomic libraries were prepared from slightly different cell populations (whole blood and peripheral white blood cells, respectively, again because of sample availability). While much of the RNA in whole blood may be derived from white blood cells, future prospective studies would benefit from using similar sample types.

Like DNA methylation, chromatin maintenance generally declines with age, and chromatin is especially dysregulated in AD [27]. ATAC-seq is a relatively new method for assessing genome-wide chromatin accessibility, and few studies have used ATAC-seq specifically to profile TE accessibility [106], especially in the context of cognitive aging and AD. In our ATAC-seq analyses, we found that many TE transcripts that were increased with aging and AD could be traced to open chromatin regions. Moreover, similar to our WGBS findings, we found fewer transcriptionally enriched TEs in chromatin-accessible regions in MCI, whereas AD dementia was associated with a broad increase in chromatin-accessible TEs. This finding is in line with the concept that certain TE classes may be suppressed in MCI prior to pheno-conversion to AD dementia [15]. Our data are also consistent with recent chromatin immunoprecipitation sequencing (ChIP-seq) data suggesting that altered chromatin structure is a mediator of TE transcript dysregulation [30, 107], and with reports of chromatin dysregulation including increased histone acetylation (i.e., activation) in AD and preclinical AD models [27, 69]. Although these prior studies were based on brain tissue and our study involved peripheral blood/cells, ours are the first analyses to use ATAC-seq to relate TE accessibility in human subjects to systemic and neuroinflammation with aging and AD.

Interestingly, while the proportions of epigenetically and transcriptionally dysregulated TEs were similar in our WGBS and ATAC-seq data, more copies of dysregulated TEs tended to be located within open chromatin than hypomethylated regions. These findings could be due to the different sizes of hypomethylated vs. chromatin-accessible regions identified by current bioinformatics pipelines, and/or they could suggest that chromatin changes may be a key factor (perhaps more so than DNA hypomethylation) in TE dysregulation. In this context, an important limitation of our data is that any one TE transcript could originate from multiple loci in the genome, and we did not link specific hypomethylated and/or chromatin-accessible loci to specific transcripts. We also note that some TE transcripts that were nominally increased were not traceable to hypomethylated and/or chromatin-accessible genome regions. This could be because epigenetic control is complex, and DNA methylation and chromatin changes at specific loci do not necessarily lead to altered transcription. There is also evidence that read-through transcription (i.e., failure to terminate transcription at stop site) and intron retention during mRNA splicing may contribute to TE transcript dysregulation with both aging and AD [108, 109]. Thus, future work in this area would benefit from identifying specific TE loci of epigenetic dysregulation (i.e., within ATAC peaks, differentially methylated regions). Emerging bioinformatics programs may make such analyses possible; in general, they may require longer read RNA/DNA sequencing approaches than those used here [110, 111], but the data we have generated may serve as a starting point for others interested in this topic.

Finally, a key question in our analyses was: Does epigenetic dysregulation characterize TEs that are relevant to clinical phenotypes? Interestingly, among the TEs, we identified as particularly epigenetically dysregulated (i.e., increased in expression, hypomethylated, and chromatin-accessible), we found 13 that were common to aging, MCI, and AD dementia. These TEs included an active LINE (L1HS) and an important HML-2 HERV-K transcript (LTR5_Hs). L1HS has been implicated in multiple aging/disease phenotypes [21, 64], and HERV-K TEs are evolutionarily young, active HERVs that may be capable of producing retrovirus-like particles [73]. In fact, transcription of HERV-K has been shown to induce innate immune, pro-inflammatory signaling associated with neurodegeneration⁷⁰. Whereas many studies on LINEs and HERV-K have used cultured animal or human cell lines [112], ours are the first analyses to profile these TEs using both WGBS and ATAC-seq in humans, and to relate them to clinically relevant outcomes. We found that transcript levels of these and other epigenetically dysregulated TEs in our study were predictive of clinical diagnosis, age (the primary risk factor for AD), as well as NfL and GFAP, cognitive function scores in three different domains, and circulating markers of inflammation including VCAM1 and IP10. VCAM1 is a mediator of leukocyte-endothelial cell signaling that is upregulated by pro-inflammatory cytokines, which can be released downstream of dsRNA sensors like MDA5 and RIG1, and it is associated with aging, impaired cognitive function and MCI/AD dementia [113–116]. IP10, released from monocytes and endothelial cells in response to interferon gamma, is also a key cytokine produced in response to dsRNA, and peripheral IP10 is associated with cognitive decline during brain aging and AD [65, 115]. Because our data show that epigenetically dysregulated TE transcripts are related to these and other biomarkers of aging and inflammation, they could suggest a link between TE transcripts (which are prone to form dsRNA) and age/AD-related inflammation, as others have suggested [25]. Given that our analyses were performed on peripheral blood samples, future studies should determine if these patterns are conserved in the brain, which cell types contribute, and if the epigenetically dysregulated TEs identified here can directly lead to inflammation via the formation of dsRNA or cDNA [22](e.g., by performing RNA immunoprecipitation sequencing on sensors to which they may bind, such as MDA5 and RIG-1).

Conclusions

Growing evidence suggests a role for TEs in aging and AD, and ongoing clinical trials are aimed at targeting TEs in this context. To the best of our knowledge, this study is the first to demonstrate genome-wide epigenetic TE dysregulation relevant to aging and AD in blood using both WGBS and ATAC-seq, and to show that most TE transcripts that are elevated with aging, MCI and AD dementia may originate from hypomethylated and/or chromatin-accessible regions of the genome in such samples. Because these TEs were related to clinical diagnosis, age, cognitive function, and circulating markers of inflammation and neurodegeneration, our findings may provide a basis for additional studies aimed at targeting TE-associated inflammation, perhaps using epigenetic modifiers.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

A.N.C. and T.J.L. conceptualized this study. B.M.B.’s research group conducted the study from which samples were obtained. A.N.C., G.T.M., M.E.S., C.M.M., and T.J.L. performed RNA and DNA isolation and related analyses. C.C. performed clinical biomarker measurements. A.N.C. and T.J.L. performed transcriptome, epigenome, and statistical analyses. A.N.C. and T.J.L. wrote the manuscript with editing assistance from all co-authors.

Funding

This study was supported by the National Institutes of Health grants R21AG060302 and R01AG078859 (T.J.L.), F31AG079594 (A.N.C.), R01AG058772 and R21AG072153 (B.M.B), and by the University of Colorado Alzheimer’s and Cognition Center (Huntington Potter, Director).

Data availability

Raw RNA-seq, WGBS, and ATAC-seq data are available on the NCBI Gene Expression Omnibus (GEO) at accession GSE270454. A code repository can be found at https://github.com/ancavalier/2025-Geroscience.git.

Declarations

Conflict of interest

The authors declare no competing interests.