Brain transcriptome-wide association study reveals selective long-term potentiation enrichment and negative directional skew of senescence-regulation pathways in Alzheimer's disease

Abstract

Background

Alzheimer's disease (AD) has a strong genetic component, but many risk loci are non-coding, limiting biological interpretation. Transcriptome-wide association studies (TWAS) help address this by imputing genetically regulated gene expression from genome-wide association study (GWAS) data.

Objective

To test whether AD-associated genetically regulated brain expression is enriched in pathways related to synaptic plasticity, glial-immune biology, senescence, mitochondrial function, nicotinamide adenine dinucleotide (NAD) metabolism, and metabolic signaling.

Methods

We applied S-PrediXcan to the European-ancestry AD GWAS meta-analysis from Bellenguez et al. (2022) across six brain tissues. Tissue-specific results were combined using Stouffer meta-analysis. Approximately 40 curated gene sets from public pathway databases and adult brain cell-type signatures were tested. Enrichment was assessed using Stouffer z-scores, Wilcoxon tests, and permutation p-values. Sensitivity analyses examined tissue consistency, top-driver removal, gene-level Z thresholds, gene-set overlap, and MHC/immune-gene exclusion.

Results

The strongest positive enrichment was long-term potentiation (LTP), particularly KEGG LTP ALL, with a Stouffer z-score of +5.267. Astrocyte-expressed genes were positively enriched and exceeded microglial signatures. Complement enrichment was largely driven by CR1, while synapse pruning was negatively enriched. Insulin, GLP-1, and PI3K–AKT–mTOR pathways also showed positive enrichment. Senescence regulation and oxidative phosphorylation showed the strongest negative skews. NAD-related sets showed mixed gene-specific effects. Sensitivity analyses identified senescence regulation, LTP, oxidative phosphorylation, and PI3K–AKT–mTOR as the most robust signals.

Conclusions

AD genetic liability captured by brain TWAS converges on glial-immune signaling, selective synaptic potentiation, and metabolic pathways, while senescence regulation and oxidative phosphorylation show directional depletion. These findings support pathway-based stratification to guide precision therapeutic approaches in AD.

Keywords

Alzheimer's disease astrocytes brain cellular senescence complement system genome-wide association studies long-term potentiation oxidative phosphorylation precision medicine transcriptome-wide association study

Introduction

Alzheimer's disease (AD) is the leading cause of dementia, and twin studies have estimated its heritability at approximately 60% to 80%.¹ Polygenic analyses and recent large genome-wide association studies (GWAS) have shown that common variation contributes substantially to risk, but many associated variants fall in non-coding regions, making it difficult to assign them to causal genes or to the biological pathways through which inherited risk is transmitted.^2–6 Single-locus analyses have highlighted immune-related and lipid-related genes such as CR1, CLU, TREM2, and APOE, but these approaches say less about how genetic effects are organized across broader cellular programs.^4,6–10 There is therefore a clear need for methods that move beyond individual loci and examine coordinated effects across gene sets with shared function (Figure 1).

Figure 1.

From heritable but unresolved Alzheimer's-disease risk to a hypothesis-driven brain TWAS. (a) AD is highly heritable (∼60–80% from twin studies), yet most common-variant risk lies in non-coding sequence and is difficult to assign to causal genes or pathways; established single loci (CR1, CLU, TREM2, APOE) map poorly onto broader cellular programs. (b) TWAS integrates GWAS summary statistics with brain eQTL weights (GTEx v8) to impute genetically regulated expression and test its association with AD risk across six brain tissues, combined by S-PrediXcan with a Stouffer meta-analysis. (c) Partly resolved biological hypotheses spanning synaptic (plasticity/LTP), glial-immune (astrocytes, microglia/complement, synapse pruning), senescence, mitochondrial (OXPHOS), and metabolic (GLP-1/insulin, NAD/sirtuins) domains—each tested for coordinated, expression-mediated genetic effects. (d) ∼40 gene sets drawn from six databases plus brain cell-type signatures were evaluated under three aims: identify enriched sets, resolve their direction (positive versus negative skew), and interpret the overall pattern within current models of AD pathogenesis. AD: Alzheimer's disease; eQTL: expression quantitative trait locus; GLP-1: glucagon-like peptide-1; GTEx v8: Genotype-Tissue Expression project version 8; GWAS: genome-wide association studies; LTP: long-term potentiation; NAD: nicotinamide adenine dinucleotide; OXPHOS: oxidative phosphorylation; S-PrediXcan: summary-statistics PrediXcan; TWAS: transcriptome-wide association studies.

Transcriptome-wide association studies (TWAS) offer one way to do this. By integrating GWAS summary statistics with expression quantitative trait locus (eQTL) reference panels, TWAS can impute genetically regulated gene expression and test whether that predicted expression is associated with disease risk.^11,12 Compared with pathway approaches that work directly at the single-nucleotide polymorphism level, TWAS brings the analysis closer to gene expression itself and allows a more direct test of whether AD-associated regulatory variation clusters within specific biological pathways or cell types. This is particularly useful in the brain, where regulatory architecture differs across tissues and cell populations. Summary-statistics PrediXcan (S-PrediXcan), an extension of the PrediXcan framework, improves power further by enabling tissue-specific imputation from GWAS summary statistics followed by cross-tissue meta-analysis.¹³ Even so, relatively few AD TWAS studies have taken a systematic, hypothesis-driven look across curated pathway collections spanning both well-known and emerging disease mechanisms.

Several important biological hypotheses in AD remain only partly resolved at the genetic level. Synaptic dysfunction is a strong correlate of cognitive decline, but it is still unclear whether inherited risk preferentially targets particular forms of synaptic plasticity, such as long-term potentiation (LTP), rather than more general synaptic remodeling.¹⁴ Glial mechanisms are also increasingly central. Reactive astrocytes accumulate around amyloid plaques and influence inflammation and metabolism, whereas microglia participate in complement-dependent synapse elimination.^15–18 Complement biology itself has been implicated in AD for many years, with CR1 consistently among the strongest loci, but its relationship to broader pruning machinery remains unsettled.^6,7,17 Cellular senescence is another prominent hypothesis, supported by evidence that senescent glial populations accumulate in AD models and may contribute inflammatory mediators.^19,20 Nicotinamide adenine dinucleotide (NAD) metabolism and sirtuin signaling have likewise attracted attention because NAD levels decline with age and mitochondrial quality-control pathways show protective effects in experimental systems.^21,22 Metabolic dysfunction, especially brain insulin resistance and impaired glucagon-like peptide-1 (GLP-1) signaling, has become increasingly relevant because of overlaps with type 2 diabetes biology and the therapeutic interest in GLP-1 receptor agonists.²³ Mitochondrial bioenergetics, particularly oxidative phosphorylation, is also a long-standing theme in AD pathogenesis.^21,22 Despite the depth of literature on each of these topics, relatively few studies have asked whether common variants influence them in a coordinated, expression-mediated way.

To address that gap, we performed a brain-tissue TWAS using S-PrediXcan and the large European-ancestry meta-analysis from Bellenguez et al..⁶ We tested enrichment across a curated set of approximately 40 pathways and signatures chosen to reflect our primary hypotheses concerning NAD metabolism, cellular senescence, GLP-1 and insulin signaling, synapse pruning, and synaptic plasticity, while also including established AD-relevant domains such as complement activation, apoptosis, mitochondrial biology, and glial cell identity. Gene sets were drawn from Hallmark, Reactome, Kyoto Encyclopedia of Genes and Genomes (KEGG), Gene Ontology (GO), WikiPathways, and custom adult human brain cell-type signatures so that the analysis would have both breadth and mechanistic specificity. This design allowed us to ask not only which pathways were enriched, but also whether related pathways showed coherent or divergent directional patterns. The study was therefore intended to connect AD genetic association signals to biologically interpretable modules, while avoiding over-interpretation of any single locus as a pathway-wide mechanism.^24,25

The main aims of the study were therefore threefold: first, to identify which curated gene sets show significant TWAS enrichment for AD risk; second, to determine the direction of those signals, whether positive or negative; and third, to interpret the overall pattern in the context of current models of AD pathogenesis. Rather than positioning the work as a search for entirely novel pathways, we focused on the specific configuration of enrichment and depletion signals, particularly the strong positive enrichment for LTP and astrocyte-related genes alongside negative skews for canonical senescence regulation and oxidative phosphorylation, as a way of refining existing hypotheses and identifying priorities for future functional and therapeutic work.

Methods

GWAS summary statistics

We used summary statistics from a large European-ancestry GWAS meta-analysis of AD and related dementias available at the time of analysis⁶ (Figure 2). This dataset included 39,106 clinically diagnosed AD cases, 46,828 proxy cases, and 401,577 controls. Imputation was performed using the TOPMed reference panel. Only variants with minor allele frequency greater than 0.01 and imputation quality above 0.8 were retained. All analyses were limited to the European-ancestry subset to reduce confounding from population stratification.

Figure 2.

Analytical workflow for the hypothesis-driven brain TWAS of Alzheimer's disease. (a) Input GWAS summary statistics from a European-ancestry meta-analysis of AD and related dementias (39,106 clinically diagnosed cases, 46,828 proxy cases, 401,577 controls). Variants were TOPMed-imputed and retained at minor allele frequency > 0.01 and imputation quality > 0.8, with analysis restricted to the European-ancestry subset to limit population stratification. (b) S-PrediXcan integrates GWAS z-scores with GTEx v8 cis-eQTL weights to impute genetically regulated expression and test its association with AD risk in six brain tissues (amygdala, anterior cingulate cortex BA24, caudate, frontal cortex BA9, hippocampus, nucleus accumbens). Positive z-scores denote higher predicted expression with increased risk; tissue-specific z-scores are combined by a fixed-effects Stouffer meta-analysis with permutation-based p-values. (c) Forty-two gene sets drawn from MSigDB, Reactome, KEGG, Gene Ontology, WikiPathways, and custom adult-brain single-nucleus signatures, organized into mechanistic themes: NAD/mitochondrial biology, cellular senescence, GLP-1/insulin and nutrient sensing, synaptic plasticity, complement/immune processes, apoptosis, and brain cell-type signatures. (d) Each set was summarized by gene coverage, Stouffer combined z, mean/median and mean absolute z, a Wilcoxon rank-sum test versus all other genes, and 10,000-permutation p-values, with FDR correction across the panel and a 30% coverage requirement. Robustness was assessed by leave-one-tissue-out and cross-tissue sign consistency, removal of the top 3/5/10 drivers, strict |Z| ≥ 2.0/3.0 thresholds, pairwise Jaccard overlap, and an 82-gene MHC/immune exclusion filter; because sets overlap biologically, interpretation emphasized effect sizes and consistency rather than strict family-wise correction. AD: Alzheimer's disease; eQTL: expression quantitative trait locus; FDR: false discovery rate; GLP-1: glucagon-like peptide-1; GTEx v8: Genotype-Tissue Expression project version 8; GWAS: genome-wide association studies; KEGG: Kyoto Encyclopedia of Genes and Genomes; MHC: major histocompatibility complex; NAD: nicotinamide adenine dinucleotide; S-PrediXcan: summary-statistics PrediXcan; TWAS: transcriptome-wide association studies.

TWAS methodology

Genetically regulated gene expression was imputed with S-PrediXcan,¹³ which combines GWAS z-scores with precomputed cis-eQTL weights from Genotype-Tissue Expression project (GTEx) version 8²⁶ (Figure 1). We restricted the analysis to six brain tissues with adequate sample size and eQTL coverage: amygdala, anterior cingulate cortex (BA24), caudate basal ganglia, frontal cortex (BA9), hippocampus, and nucleus accumbens. For each gene, S-PrediXcan yields a z-score that reflects the association between predicted expression and AD risk. Positive z-scores indicate that higher genetically predicted expression is associated with increased AD risk, whereas negative z-scores indicate the opposite. To improve power and identify consistent cross-region signals, we combined tissue-specific z-scores with a fixed-effects Stouffer meta-analysis. Permutation-based p-values were also calculated to provide an empirical measure of significance that accounts for correlations among tissues.

Gene-set curation and rationale

We assembled a panel of 42 gene sets chosen to test specific mechanistic hypotheses while still covering the major biological themes already implicated in AD. Gene sets came from the Molecular Signatures Database, Reactome, KEGG, GO, WikiPathways, and custom compilations derived from adult human brain single-nucleus RNA-seq atlases.^27,28

NAD metabolism and related mitochondrial biology were represented by several complementary sets, including WikiPathways NAD metabolism, KEGG nicotinate and nicotinamide metabolism, Reactome nicotinate metabolism, GO molecular function NAD + binding, GO biological process NAD transport, and WikiPathways NAD biosynthetic pathways. These were supplemented with the Mootha mitochondria set and the Hallmark oxidative phosphorylation set because mitochondrial dysfunction and NAD decline are tightly connected in aging and AD. Sirtuin-related biology was captured through WikiPathways NAD metabolism sirtuins and aging and the Reactome set SIRT1 negatively regulates rRNA expression.

Cellular senescence was evaluated through GO biological process regulation of cellular senescence, GO biological process cellular senescence, Reactome cellular senescence, Reactome senescence, the Saul senescence or SenMayo signature, and Reactome telomere maintenance. These sets were included because senescence is increasingly discussed as a source of neuroinflammation and because senolytic approaches are being actively explored.

GLP-1, insulin, and nutrient-sensing pathways were represented by GO biological process insulin secretion involved in cellular response to glucose stimulus, Reactome regulation of insulin secretion, WikiPathways GLP-1 from intestine and pancreas and role in glucose homeostasis, Reactome synthesis secretion and inactivation of glucagon-like peptide 1, Reactome GLP-1 regulates insulin secretion, Hallmark PI3K–AKT–mTOR signaling, Reactome energy-dependent regulation of mTOR by LKB1–AMPK, GO biological process CAMKK–AMPK signaling cascade, and Mootha PGC-1α target genes. These sets directly addressed the metabolic hypothesis of AD and the growing interest in GLP-1 receptor agonists.

Synaptic plasticity and related pathways were represented by GO biological process regulation of synaptic plasticity, KEGG LTP, GO biological process glutamatergic synapse, and GO biological process synapse pruning. The contrast between a broad plasticity set and the more specific LTP set was deliberate, as was the inclusion of a dedicated pruning set to compare against complement-related immune signaling.

Complement and immune processes were assessed with the Reactome complement cascade and a custom HLA complex gene set. Cell-type specificity was examined through custom adult brain signatures for astrocytes, oligodendrocytes, and microglia derived from published single-nucleus atlases. Apoptosis-related programs were included via GO cellular component BCL-2 family protein complex, GO biological process intrinsic apoptotic signaling pathway, Reactome intrinsic pathway for apoptosis, KEGG apoptosis, and Hallmark apoptosis to provide additional context for interpreting the mitochondrial and senescence findings.

Statistical analysis

For each gene set, we calculated the number of genes represented in the TWAS output, the percentage coverage of the original set, the Stouffer combined z-score, the mean and median gene-level z-scores, the mean absolute z-score, and the Wilcoxon rank-sum p-value comparing genes in the set with all other genes. Permutation p-values were generated by randomly resampling gene sets of equal size 10,000 times. A nominal threshold of p < 0.05 was used for Stouffer and permutation results, and false discovery rate (FDR) correction was also applied across the full panel of tested sets. Gene-set coverage had to exceed 30% for a set to be interpreted. We further examined tissue-specific patterns and carried out leave-one-tissue-out sensitivity analyses to test robustness. Additional sensitivity analyses were performed by removing the top 3, 5, and 10 gene-level drivers by absolute Z, retaining only genes with absolute Z of at least 2.0 or at least 3.0, assessing sign consistency across the six brain tissues, estimating pairwise gene-set overlap using the Jaccard index, and filtering a prespecified 82-gene major histocompatibility complex (MHC)/immune exclusion list.

Multiple-testing considerations and interpretation framework

Because many of the tested gene sets overlap biologically, we treated the collection as a hypothesis-generating panel rather than imposing a strict family-wise error correction across all sets. We therefore emphasized effect sizes, especially Stouffer z-scores, together with cross-tissue consistency and permutation support. We also distinguished between positive enrichment, in which genes in a set tended to have positive TWAS z-scores, and negative enrichment, in which genes tended to have negative z-scores. A negative Stouffer z-score was interpreted as indicating that the expression pattern represented by that set was inversely associated with AD risk rather than biologically irrelevant. All analyses were performed in Python with custom scripts built around the S-PrediXcan output structure.

Results

The TWAS analysis was run on the Bellenguez et al.⁶ European-ancestry AD GWAS summary statistics using S-PrediXcan across six brain tissues, followed by a Stouffer meta-analysis. Coverage across curated gene sets was generally high, with most sets exceeding 70% representation in the TWAS results, which allowed stable interpretation of enrichment patterns. Meta-analysis performance was robust, and several sets reached nominal or permutation-supported significance. Tissue-specific effects were most often strongest in caudate basal ganglia, nucleus accumbens, hippocampus, and frontal cortex, although the leading signals generally showed directional concordance across all six regions.

Among positively enriched pathways, LTP, represented by KEGG LTP ALL, showed the strongest meta-analytic signal, with a Stouffer z-score of +5.267 and a permutation p-value of 0.026 (Table 1). This enrichment was driven by genes involved in calcium handling, MAPK and ERK signaling, and transcriptional regulation, including MAPK3, PPP3R1, EP300, and HRAS. By contrast, the broader GO biological process set for regulation of synaptic plasticity did not show similarly strong enrichment, suggesting that AD genetic risk may be linked more specifically to LTP-related machinery than to synaptic plasticity in a general sense. Genes in the glutamatergic synapse set were only mildly positive and showed less consistency across tissues.

Table 1.

Positively enriched biological pathways, meta-analysis ranked by Stouffer Z.

Pathway	Source	n	Cov. (%)	Stouffer Z	Mean Z	PWilc.	Pperm	Key driver genes
Long-term potentiation	KEGG	59/70	84.3	+5.267	+0.686	0.012	0.026	MAPK3, PPP3R1, EP300, HRAS
PI3K–AKT–mTOR signaling	Hallmark	90/105	85.7	+4.468	+0.471	0.085	0.057	Caudate and nucleus accumbens driven
Complement cascade	Reactome	41/115	35.7	+4.188	+0.654	0.248	0.075	CR1 (+27.6), CLU (−12.1)
Insulin secretion, glucose response	GO:BP	56/75	74.7	+3.659	+0.489	0.304	0.115	Nucleus accumbens and frontal cortex
GLP-1 synthesis, secretion, and inactivation	Reactome	13	—	+2.320	+0.644	0.147	0.311	Driver-sensitive incretin signal
Regulation of insulin secretion	Reactome	63/78	80.8	+1.865	+0.235	0.411	0.421	Metabolic–AD link
GLP-1 regulates insulin secretion	Reactome	34/42	81.0	+1.664	+0.285	0.636	0.469	Incretin signaling
Intrinsic pathway for apoptosis	Reactome	48/55	87.3	+1.510	+0.218	0.532	0.509	Mild positive
Apoptosis	KEGG	72/87	82.8	+1.469	+0.173	0.524	0.527	Mild positive

Bold values indicate amended GLP-1-related sensitivity-supported entry. KEGG: Kyoto Encyclopedia of Genes and Genomes; GLP-1: glucagon-like peptide-1; GO:BP: Gene Ontology Biological Process; AD: Alzheimer's disease

Astrocyte-expressed genes, represented by ADULT ASTRO ALL, also showed clear positive enrichment, with a Stouffer z-score of +3.971 and a permutation p-value of 0.085, at 85% coverage (Table 2). This was one of the most prominent cell-type signals in the study. Important contributing genes included ABO, SDC4, ACSBG1, and SLC25A18, several of which have already been linked to astrocytic function in neurodegenerative settings. In contrast, the oligodendrocyte set showed no enrichment and trended slightly negative, while the microglial set was modestly positive, with a Stouffer z-score of +2.038, but did not approach the strength of the astrocyte signal.

Table 2.

Cell-type-specific gene set enrichment for Alzheimer's disease, meta-analysis across six brain tissues.

Cell type	n	Coverage (%)	Stouffer Z	Mean absolute Z	Mean Z	PWilcoxon	Pperm	Top driver genes
Astrocytes	34/40	85.0	+3.971	1.868	+0.681	0.191	0.085	ABO, SDC4, ACSBG1, SLC25A18
Microglia	22/25	88.0	+2.038	1.598	+0.435	0.715	0.372	HLA-DRA (+9.45), A2 M, ITGAX
Oligodendrocytes	27/35	77.1	−0.862	1.400	−0.166	0.804	0.707	No strong drivers

Positive Stouffer Z indicates enrichment toward AD genetic association. Pperm, permutation p-value with 10,000 permutations. n, genes found divided by total in set.

The complement cascade, represented by REACTOME COMPLEMENT ALL, was positively enriched as well, with a Stouffer z-score of +4.188 and a permutation p-value of 0.075, although set coverage was lower at 35.7%. This result was dominated by CR1, which showed an exceptionally strong positive TWAS association, with a meta-analytic z-score of +27.60 across all tissues. Other complement-related genes were more mixed in direction; for instance, CLU was strongly negative, with a z-score of −12.08. This combination is in line with the established role of CR1 as a major AD risk locus while also underscoring the complexity of complement pathway regulation in the brain.

The strongest negative enrichment signal in the entire analysis came from regulation of cellular senescence, represented by GOBP REGULATION OF CELLULAR SENESCENCE ALL (Table 3). This set had a Stouffer z-score of −5.832 and a permutation p-value of 0.013, making it the most statistically supported negative finding. The direction was consistent across all six brain tissues. Broader cellular senescence sets and curated senescence marker panels, including the Saul SenMayo signature, also trended negative, although less strongly. These results suggest that genes involved in canonical senescence regulation were depleted for positive AD genetic associations in this TWAS framework.

Table 3.

Negatively enriched or depleted biological pathways, ranked by Stouffer Z.

Pathway	Source	Stouffer Z	Mean Z	Pperm	Interpretation
Regulation of cellular senescence	GO:BP	−5.832	−0.972	0.013	Strongest depletion; consistent across all six regions
Oxidative phosphorylation	Hallmark	−4.797	−0.380	0.035	Broad electron transport chain/OXPHOS depletion, including NDUFS2 and HADHB
Energy-dependent regulation of mTOR by LKB1–AMPK	Reactome	−3.847	−0.754	0.095	Energy-sensing axis depleted
Senescence, Mayo panel	Saul et al.	−3.232	−0.325	0.174	Curated senescence markers depleted
CaMKK–AMPK signaling cascade	GO:BP	−2.996	−0.947	0.193	Negative; energy-sensing pathway
Synapse pruning	GO:BP	−2.752	—	—	TREM2-influenced negative signal
Cellular senescence	GO:BP	−2.452	−0.271	0.299	Directionally concordant with regulation set
BCL-2 family protein complex	GO:CC	−1.988	−0.629	0.388	Anti- and pro-apoptotic balance

Negative Stouffer Z indicates gene set depletion for AD genetic signals. GO:BP: Gene Ontology Biological Process; GO:CC: Gene Ontology Cellular Component; OXPHOS: oxidative phosphorylation

Oxidative phosphorylation, represented by the Hallmark set, showed a similarly strong negative pattern, with a Stouffer z-score of −4.797 and a permutation p-value of 0.035. Many core electron transport chain genes contributed to this signal with consistently negative z-scores, including NDUFS2, which had a z-score of −11.70, and HADHB, which was also among the strongest negative drivers. Related energy-sensing pathways, such as LKB1–AMPK regulation of mTOR and the CaMKK–AMPK signaling cascade, also leaned negative, though with somewhat weaker support.

The synapse pruning set, GOBP SYNAPSE PRUNING ALL, was negatively enriched in the meta-analysis, with a z-score of −2.752, despite the positive signal observed for complement biology. This was a notable dissociation and was influenced in part by TREM2, which showed a strong negative TWAS association at −8.61. The result suggests that complement-related immune activity and the broader canonical pruning program are not identical from the perspective of inherited AD risk.

Metabolic and insulin-related pathways tended to be positively enriched, although their effect sizes were smaller than those observed for LTP and astrocyte identity. The Hallmark PI3K–AKT–mTOR signaling set had a Stouffer z-score of +4.468 and a permutation p-value of 0.057, with especially pronounced signals in caudate and nucleus accumbens. GO biological process insulin secretion involved in cellular response to glucose stimulus and Reactome regulation of insulin secretion were also positive, with z-scores of +3.659 and +1.865, respectively. GLP-1-related pathways, including Reactome GLP-1 regulates insulin secretion, showed modest positive enrichment, with a z-score of +1.664. A more specific Reactome set for synthesis, secretion, and inactivation of GLP-1 reached a meta-analytic Stouffer z-score of +2.320, although this signal was modest and became non-significant after top-driver removal. Taken together, these results support a role for metabolic signaling in AD genetic risk, though not at the level of the strongest signals in the dataset.

NAD metabolism and sirtuin-related pathways did not show a single coherent pathway-wide pattern. Broader NAD metabolism and biosynthesis sets were neutral to mildly positive overall, without strong enrichment or depletion. At the gene level, however, notable signals were present in both directions. SIRT3 showed a positive association with a z-score of +4.73, and NT5E and NMNAT2 were also positive to a lesser degree. In contrast, QPRT was one of the stronger negative drivers, with a z-score of −5.01. The GO molecular function set for NAD + binding leaned slightly negative. Overall, the results suggest that NAD-related biology contributes to AD genetic architecture in a gene-specific manner rather than through a strong pathway-wide shift.

Comparisons across glial cell-type signatures reinforced the astrocyte result as the clearest of the three glial populations examined. Microglial genes showed moderate positive enrichment with bidirectional drivers; HLA-DRA, for example, was strongly positive at +9.45, whereas several other microglial genes were negative. Oligodendrocyte genes remained close to neutral and showed no strong enrichment. These general patterns remained stable after accounting for differences in gene-set coverage.

At the level of individual genes, the strongest positive TWAS associations included CR1 at +27.60, followed by several HLA-related and astrocyte-enriched genes (Table 4). The strongest negative associations included multiple mitochondrial genes, such as NDUFS2, MTCH2, and TOMM40, as well as complement regulators such as CLU and CD55. Some genes contributed to more than one pathway signal, including MAPK3, which appeared in both LTP and senescence-related contexts, and SIRT3, which linked NAD and mitochondrial biology. Tissue consistency was high for the most extreme signals; CR1 and NDUFS2, for example, showed very large z-scores in the same direction across all six tissues. Other genes, particularly some mitochondrial and synaptic transcripts, varied more in effect size across regions, with caudate and nucleus accumbens often showing the largest magnitudes.

Table 4.

Top driver genes by absolute AD Z from S-PrediXcan meta-analysis.

Gene	Pathway context	AD Z	Direction	Biological note
CR1	Complement	+27.60	↑	Complement receptor 1; top AD GWAS locus. Extreme signal across all tissues.
HLA-DOB	HLA complex	−12.77	↓	MHC class II; strong immune-related negative.
CLU	Complement	−12.08	↓	Clusterin, apolipoprotein J; established AD GWAS gene.
NDUFS2	Mitochondria/OXPHOS	−11.70	↓	Complex I subunit; consistent depletion across all regions.
MTCH2	Mitochondria	−9.83	↓	Mitochondrial carrier homologue.
CD55	Complement	−9.61	↓	Decay-accelerating factor; complement regulation.
HLA-DQA2	HLA complex	−9.48	↓	MHC class II; immune/antigen presentation.
HLA-DRA	Microglia/HLA	+9.45	↑	MHC class II; dominant microglial positive driver.
TREM2	Synapse pruning	−8.61	↓	Microglial receptor; known AD risk gene, negative Z here.
TOMM40	Mitochondria	−8.23	↓	Outer membrane translocase; proximal to APOE.
BCKDK	Mitochondria	+8.18	↑	Rare positive mitochondrial outlier.
NTRK1	Glutamatergic	+7.40	↑	Neurotrophic tyrosine kinase receptor.
MAPK3	LTP/senescence	+5.75	↑	ERK1; key LTP signaling node.
ABO	Astrocytes	+5.00	↑	Glycosyltransferase; top astrocyte marker.
QPRT	NAD metabolism	−5.01	↓	Quinolinate phosphoribosyltransferase; NAD biosynthesis.
SIRT3	NAD/sirtuins	+4.73	↑	Mitochondrial sirtuin; NAD-dependent deacetylase.

AD Z: combined S-PrediXcan Z-score across tissues. ↑ indicates positive enrichment; ↓ indicates negative enrichment or depletion.

Sensitivity analyses clarified which pathway findings were robust and which were more dependent on a small number of drivers. Four pathways were robust to top-driver removal: GOBP REGULATION OF CELLULAR SENESCENCE ALL, KEGG LTP ALL, HALLMARK OXIDATIVE PHOSPHORYLATION ALL, and HALLMARK PI3 K AKT MTOR SIGNALING ALL. Regulation of cellular senescence remained significant after removal of its top 10 drivers, with the Stouffer Z changing from −5.832 to −2.238, and strengthened under strict gene-level filtering, reaching −7.457 at absolute Z of at least 2.0 and −9.713 at absolute Z of at least 3.0. KEGG LTP ALL remained positive after removal of its top 10 drivers, with the Stouffer Z changing from +5.267 to +2.262, and strengthened to +8.740 and +8.332 under the same strict thresholds. Oxidative phosphorylation remained negative after top-10 removal, changing from −4.797 to −2.589, while PI3K–AKT–mTOR signaling remained positive after top-10 removal, changing from +4.468 to +3.310.

The immune-gene filter separated immune-dependent from immune-independent signals. Regulation of cellular senescence was almost unaffected by filtering, with only 1 of 36 genes removed and the Stouffer Z strengthening from −5.832 to −6.428. In contrast, the complement signal was highly immune-dependent: removal of 24 of 41 immune or MHC genes changed the Stouffer Z from +4.188 to −1.549. Synapse pruning was also immune-sensitive, changing from −2.752 to −0.244 after removal of 6 of 16 immune genes. The astrocyte signal was directionally consistent across tissues but fragile to top-driver removal, with the Stouffer Z falling from +3.971 to +0.033 after the top five drivers and to −0.795 after the top 10 drivers were removed. These analyses support treating LTP, canonical senescence regulation, oxidative phosphorylation, and PI3K–AKT–mTOR as the most robust pathway-level findings, while interpreting complement, astrocyte, microglial, and pruning signals as more gene-specific or immune-dependent.

Taken together, the results point to a distinct pattern in which AD genetic liability, as captured by brain TWAS, is most strongly aligned with LTP machinery, astrocyte-expressed genes, and complement signaling, while showing consistent negative skews for senescence regulation, oxidative phosphorylation, and synapse pruning programs. Metabolic signaling pathways sit in between, with supportive but less dominant enrichment.

Discussion

This brain-tissue TWAS points to a fairly coherent structure in the way common genetic risk for AD is organized. Rather than implicating one dominant biological process, the data suggest that inherited risk converges most strongly on glial-immune signaling, especially astrocytes and complement; selective components of synaptic potentiation, particularly LTP; and metabolic signaling pathways involving insulin, GLP-1, and PI3K–AKT–mTOR. At the same time, the analysis shows a clear negative directional skew for canonical senescence-regulation programs and for oxidative phosphorylation genes. Taken together, the findings fit a model in which common variation influences the brain's capacity to maintain synaptic function and glial homeostasis under inflammatory and metabolic stress, rather than acting mainly through diffuse aging pathways.

Mechanistic implications of the LTP enrichment and senescence-regulation depletion signals

The selective enrichment for LTP is one of the most distinctive results. Synaptic dysfunction is already recognized as central to AD, but the fact that the LTP-specific set was strongly enriched while the broader regulation of synaptic plasticity set was not suggests a narrower target (Figure 3). This interpretation was strengthened by the sensitivity analyses. KEGG LTP ALL had the same positive direction across all six brain tissues, remained significant after the top 10 drivers were removed, and became stronger when the analysis was restricted to genes with larger absolute gene-level Z-scores. The top-driver removal step is important because it showed that the LTP result was not reducible to a single transcript such as MAPK3. Even after removing MAPK3, PPP3R1, EP300, HRAS, RAF1, CREBBP, PPP1CC, BRAF, PLCB2, and MAP2K1, the set retained a positive Stouffer Z of +2.262. LTP depends on tightly coordinated calcium signaling, kinase cascades such as MAPK and ERK, phosphatase regulation, and CREB-dependent transcription. The leading LTP genes in this analysis fit that biology: MAPK3 marks the ERK arm of the pathway, PPP3R1 links calcium-dependent phosphatase activity to the LTP/LTD balance, EP300 and CREBBP connect synaptic activity to transcriptional regulation, and HRAS, RAF1, BRAF, and MAP2K1 sit within the Ras–RAF–MEK–ERK signaling axis. Disruption at these points could make synapses more vulnerable to amyloid-β oligomers or tau pathology, both of which are known to impair LTP through calcium dysregulation, NMDA receptor signaling, and downstream transcriptional effects.^14,29–35 The lack of comparable enrichment in the broader plasticity set implies that genetic liability may not affect all forms of synaptic remodeling equally. Instead, the data highlight activity-dependent strengthening mechanisms, particularly in the hippocampus, frontal cortex, caudate, and nucleus accumbens, as a more specific site of vulnerability. That selectivity may help explain why some cognitive functions begin to fail earlier than others.

Figure 3.

Proposed mechanistic model linking genetically regulated brain expression to Alzheimer's disease biology. (a) Activity-dependent synaptic strengthening proceeds from NMDA-receptor Ca²⁺ entry through Ca²⁺-dependent signaling (calcineurin, CaMKII) and the Ras–RAF–MEK–ERK cascade to CREB–CBP/p300-mediated transcription. The model proposes that inherited liability concentrates in this core long-term-potentiation machinery rather than in broad synaptic plasticity, rendering it a specific point of vulnerability to amyloid-β and tau oligomers that disrupt calcium and NMDA-receptor signaling. (b) Canonical senescence regulators (e.g., YPEL3, ZKSCAN3, PAWR, TERF2) are depicted as relatively spared by common genetic risk; instead, senescence is positioned as a downstream consequence of glial activation and metabolic stress. This framing implies that broad senolytic strategies may benefit from biomarker-guided patient selection rather than uniform application. (c) Core oxidative-phosphorylation and TCA-cycle machinery is interpreted as a substrate of mitochondrial resilience, while the PI3K–AKT–mTOR axis—centered on GSK3B as a hub linking insulin signaling, tau phosphorylation, mTOR activity, and synaptic plasticity—provides a metabolic-signaling counterpart. Together they converge on the concept of synaptic resilience under metabolic stress and argue for targeted bioenergetic and insulin-pathway interventions over simple “mitochondrial-boosting” approaches. (d) A genetic dissociation is proposed between a complement-related immune axis (dominated by CR1) and the canonical synapse-pruning program (influenced by TREM2), with astrocytic contributions concentrated in specific functions such as lipid handling and ion transport rather than broad reactive astrogliosis. The model favors selective modulation of complement components or TREM2-related pathways over blanket suppression of microglial pruning.

The strong negative enrichment for regulation of cellular senescence stands out because much of the recent AD literature has framed senescence as a risk-driving process and a plausible therapeutic target.^19,20 Here, however, GOBP REGULATION OF CELLULAR SENESCENCE ALL was the most robust negative pathway in the study. It showed the same negative sign across all six brain tissues, remained significant after removal of the top 10 drivers, strengthened under strict absolute Z thresholding, and was essentially unchanged by the immune-gene filter. Only 1 of 36 genes in the set was removed by the MHC/immune exclusion list, and the Stouffer Z strengthened from −5.832 to −6.428. In this analysis, genes involved in canonical senescence regulation tended to show TWAS associations in the direction opposite to AD risk. The top drivers were biologically interpretable rather than random: YPEL3 has been described as a p53-regulated gene capable of inducing cellular senescence, ZKSCAN3 is a transcriptional repressor of autophagy and lysosome-related genes, PAWR encodes the pro-apoptotic tumor suppressor Par-4, and TERF2/TRF2 is a telomere-protection factor whose dysfunction can connect telomere damage to senescence.^36–41 One reasonable interpretation is that common-variant genetic liability does not primarily act through these canonical senescence-control programs. Instead, senescence in the AD brain may be more of a downstream response to other genetically influenced processes, including glial activation and metabolic stress, which showed clearer evidence in this dataset.^15,16,37 This does not argue against a role for senescent cells in established disease, but it does suggest that broad senolytic approaches may need more careful biomarker-based selection rather than a one-size-fits-all application.

The mitochondrial and metabolic findings help connect the LTP and senescence results. HALLMARK OXIDATIVE PHOSPHORYLATION ALL was also robustly negative, including after top-driver removal and strict thresholding. The direction suggests that higher genetically predicted expression of many core electron transport chain and TCA-related genes, including NDUFS2, NDUFAB1, NDUFA2, HADHB, PHB2, OGDH, IDH1, and DLD, was associated with lower AD risk in this TWAS framework. This is compatible with a mitochondrial-resilience interpretation rather than a simple disease-activation interpretation. In contrast, Hallmark PI3K–AKT–mTOR signaling showed a robust positive signal, with driver genes including PFN1, PLCG1, PIKFYVE, TNFRSF1A, RIPK1, TBK1, RAC1, GSK3B, and MYD88. GSK3B is particularly relevant because it sits at the intersection of insulin signaling, tau phosphorylation, synaptic plasticity, and mTOR-related biology.^21–23^,42–46 This combination points toward synaptic resilience under metabolic stress rather than isolated dysfunction in one pathway.

The relationship between complement and synapse pruning adds another layer of complexity. Complement genes were positively enriched, driven above all by CR1, which is fully consistent with longstanding genetic and functional evidence linking complement biology to AD.^4,6,7 Yet the synapse pruning set was negatively enriched, with TREM2 making a substantial contribution to that negative skew.^9,10 The sensitivity analyses strengthened this distinction. Complement remained positive under strict gene-level thresholds but collapsed after immune-gene filtering, with removal of 24 of 41 immune or MHC genes producing a sign flip from +4.188 to −1.549. Synapse pruning, by contrast, was negative and also immune-sensitive, falling from −2.752 to −0.244 after immune filtering. This suggests that complement-related immune signaling and the broader pruning program are not interchangeable from the perspective of inherited risk. Complement activation can support clearance but can also amplify damaging inflammation, and experimental work has shown that complement and microglia can mediate early synapse loss in AD models.¹⁷ The current pattern is more consistent with a specific complement-related immune axis being genetically aligned with increased AD risk, while the broader pruning machinery leans in a more protective or context-dependent direction. That distinction matters therapeutically, because it argues for more selective modulation of complement components or TREM2-related pathways rather than blanket suppression of pruning-related microglial functions.

The astrocyte result also requires a qualified interpretation. ADULT ASTRO ALL was positive and tissue-consistent, but the top-driver analysis showed that the signal was concentrated in a relatively small number of genes. Removal of ABO, SDC4, ACSBG1, SLC25A18, and SLC39A12 reduced the Stouffer Z from +3.971 to +0.033, and removal of the top 10 drivers produced a sign flip. This does not negate the astrocyte signal, but it suggests that the result may reflect specific astrocytic functions, such as lipid handling, ion transport, extracellular matrix regulation, or metabolic support, rather than a broad reactive astrogliosis program. This distinction is consistent with the wider AD literature in which astrocytes contribute to inflammation, amyloid handling, and metabolic support, but do so in state- and context-dependent ways.^15,18

The negative skew observed for oxidative phosphorylation genes is also notable. Mitochondrial dysfunction has been tied to AD for decades,²¹ but the TWAS results suggest that higher genetically predicted expression of many core oxidative phosphorylation genes is associated with lower disease risk rather than higher risk. In practical terms, this looks more like a protective expression pattern than a disease-promoting one. It may reflect inherited differences that support mitochondrial resilience or preserve energy production under stress. If so, the result argues against simplistic strategies aimed only at globally increasing mitochondrial output and instead supports more targeted approaches that improve mitochondrial quality control, signaling, or cell-type-specific bioenergetic support, particularly in astrocytes, which showed the strongest cell-type enrichment in this analysis.^22,42

Placed alongside prior AD TWAS and pathway work, the present findings both reinforce and refine what is already known. Previous studies have repeatedly emphasized immune loci, complement biology, astrocytic involvement, and metabolic overlap with insulin signaling.^4,6,12,23 What this analysis adds is sharper resolution. Complement and synapse pruning do not move together. LTP appears more clearly implicated than broad synaptic plasticity. Canonical regulation of senescence trends in the opposite direction from what might be expected if it were a direct mediator of common genetic risk. In that sense, the results do not overturn current models, but they do narrow them and make them more testable.

Several limitations should be kept in view. First, the analysis depended on GTEx-based brain eQTL models, which may not fully capture disease-state, developmental-stage, or cell-type-specific regulatory effects present in AD tissue.^12,26 Second, the signed enrichment strategy is useful for detecting directional patterns, but it can still be influenced by gene-set composition and by correlation structures created by linkage disequilibrium. Although coverage was generally strong and permutation testing was included, the tested panel of overlapping gene sets necessarily raises the possibility of false positives, which is why effect size and cross-tissue consistency were emphasized over very strict multiple-testing correction. Third, the present analysis did not incorporate colocalization or fine-mapping, so causal inference at the level of individual genes remains limited. More generally, TWAS identifies association between genetically predicted expression and disease risk; it does not by itself establish causality, identify the causal variant, or prove that the same regulatory effect is active in diseased cell states. The European-ancestry focus of the GWAS also limits generalizability, and the findings should be tested in more diverse populations as suitable TWAS weights and GWAS data become available.^6,12

Even with those limitations, the translational implications are fairly clear. The strongest immediate opportunity lies in pathway-based stratification. Polygenic or polytranscriptomic scores derived from LTP, astrocyte, complement, and insulin/GLP-1-related gene sets could be used to enrich clinical trials for individuals whose biology is most aligned with a particular treatment mechanism. This is especially relevant for GLP-1 receptor agonists, which are already being evaluated in AD and for which the present metabolic enrichment provides additional genetic support.⁴⁷ Glial-targeted interventions, particularly those aimed at complement signaling or restoration of astrocyte homeostasis, also appear to match the dominant pathway-level signals. By contrast, the results provide a weaker rationale for broad senolytic strategies or nonspecific NAD-precursor supplementation as universal interventions, and they suggest that any such approaches would benefit from prior biomarker selection.

Several next steps would help translate these findings more effectively. Colocalization and fine-mapping focused on top driver genes such as CR1, MAPK3, SIRT3, NDUFS2, and TREM2 would help separate likely causal effects from tagging signals. Cell-type-resolved validation using induced pluripotent stem cell-derived systems, spatial transcriptomics, or single-cell datasets from AD brain would clarify whether the pathway-level patterns seen here are preserved at higher resolution. Pathway-specific scores could also be developed and tested in independent cohorts such as ADNI or ROS/MAP to assess their value for prediction and trial enrichment. Finally, these results could be integrated directly into clinical development strategies for metabolic and glial-directed therapies.

Overall, this TWAS analysis sharpens the current picture of AD genetic architecture. Common-variant liability appears to align most strongly with glial-immune signaling, selective synaptic potentiation, and metabolic pathways, while showing a consistent negative directional skew for canonical senescence regulation and broad oxidative phosphorylation programs. That pattern has practical value. It points toward a more focused set of biological modules for functional follow-up and suggests a clearer path toward therapy development guided by molecular subtype rather than by diagnosis alone.

Conclusion

This brain-tissue TWAS of AD shows a distinctive and internally consistent pattern. Genetic liability was most strongly aligned with astrocyte and complement signaling, LTP machinery, and insulin- and GLP-1-related metabolic pathways. At the same time, canonical senescence-regulation pathways and oxidative phosphorylation programs showed consistent negative directional skews. The specificity of these findings matters. LTP, rather than broad synaptic plasticity, emerged as the clearest synaptic signal, and complement biology, rather than canonical synapse pruning, emerged as the more prominent immune axis. Follow-up sensitivity analyses further showed that the LTP, senescence-regulation, oxidative phosphorylation, and PI3K–AKT–mTOR signals were the most robust pathway-level findings, whereas complement, astrocyte, microglial, and pruning signals require more gene-specific interpretation.

These results refine current models of AD rather than replace them. They support the view that inherited risk acts through specific, biologically coherent modules rather than through general aging-related dysfunction. They also point to practical translational priorities, including pathway-based patient stratification and the prioritization of glial and metabolic interventions, especially complement-modulating and GLP-1-related therapies. By contrast, they suggest more caution around broad use of senolytics or generic NAD-boosting strategies without biomarker guidance.

Taken together, the findings support a precision-medicine approach to AD in which genetic and transcriptomic profiling helps identify the dominant biological vulnerabilities in a given patient and improves the match between mechanism and treatment.

Supplemental Material

sj-docx-1-alr-10.1177_25424823261468711 - Supplemental material for Brain transcriptome-wide association study reveals selective long-term potentiation enrichment and negative directional skew of senescence-regulation pathways in Alzheimer's disease

Supplemental material, sj-docx-1-alr-10.1177_25424823261468711 for Brain transcriptome-wide association study reveals selective long-term potentiation enrichment and negative directional skew of senescence-regulation pathways in Alzheimer's disease by Ngo Cheung in Journal of Alzheimer's Disease Reports

Footnotes

Acknowledgements

The author thanks the investigators and participants of the original genome-wide association, GTEx, and public pathway-resource studies that made this analysis possible.

ORCID iD

Ngo Cheung

Ethical considerations

Ethics approval was not required for this study because all analyses were performed using publicly available, de-identified genome-wide association study summary statistics and reference transcriptomic resources. No individual-level or identifiable participant data were accessed.

Consent to participate

Not applicable

Consent for publication

Not applicable

Author contribution(s)

Ngo Cheung: Writing – original draft; Writing – review & editing.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The genome-wide association study summary statistics and transcriptomic prediction resources used in this study are publicly available from the sources described in the Methods section. Derived summary tables and analysis outputs are available at the Supplemental Material.

Supplemental material

Supplemental material for this article is available online.

References

Gatz

Reynolds

Fratiglioni

, et al. Role of genes and environments for explaining Alzheimer disease. Arch Gen Psychiatry 2006; 63: 168–174.

Escott-Price

Shoai

Pither

, et al. Polygenic score prediction captures nearly all common genetic risk for Alzheimer’s disease. Neurobiol Aging 2017; 49: 214.e7–214.e11.

Neuner

Tcw

Goate

. Genetic architecture of Alzheimer’s disease. Neurobiol Dis 2020; 143: 104976.

Kunkle

Grenier-Boley

Sims

, et al. Genetic meta-analysis of diagnosed Alzheimer’s disease identifies new risk loci and implicates Aβ, tau, immunity and lipid processing. Nat Genet 2019; 51: 414–430.

Wightman

Jansen

Savage

, et al. A genome-wide association study with 1,126,563 individuals identifies new risk loci for Alzheimer’s disease. Nat Genet 2021; 53: 1276–1282.

Bellenguez

Küçükali

Jansen

, et al. New insights into the genetic etiology of Alzheimer’s disease and related dementias. Nat Genet 2022; 54: 412–436.

Lambert

Heath

Even

, et al. Genome-wide association study identifies variants at CLU and CR1 associated with Alzheimer’s disease. Nat Genet 2009; 41: 1094–1099.

Harold

Abraham

Hollingworth

, et al. Genome-wide association study identifies variants at CLU and PICALM associated with Alzheimer’s disease. Nat Genet 2009; 41: 1088–1093.

Guerreiro

Wojtas

Bras

, et al. TREM2 Variants in Alzheimer’s disease. N Engl J Med 2013; 368: 117–127.

10.

Jonsson

Stefansson

Steinberg

, et al. Variant of TREM2 associated with the risk of Alzheimer’s disease. N Engl J Med 2013; 368: 107–116.

11.

Gusev

Shi

, et al. Integrative approaches for large-scale transcriptome-wide association studies. Nat Genet 2016; 48: 245–252.

12.

Wainberg

Sinnott-Armstrong

Mancuso

, et al. Opportunities and challenges for transcriptome-wide association studies. Nat Genet 2019; 51: 592–599.

13.

Barbeira

Dickinson

Bonazzola

, et al. Exploring the phenotypic consequences of tissue-specific gene expression variation inferred from GWAS summary statistics. Nat Commun 2018; 9: 1825.

14.

Forner

Baglietto-Vargas

Martini

, et al. Synaptic impairment in Alzheimer’s disease: a dysregulated symphony. Trends Neurosci 2017; 40: 347–357.

15.

Liddelow

Guttenplan

Clarke

, et al. Neurotoxic reactive astrocytes are induced by activated microglia. Nature 2017; 541: 481–487.

16.

Heneka

Carson

El Khoury

, et al. Neuroinflammation in Alzheimer’s disease. Lancet Neurol 2015; 14: 388–405.

17.

Hong

Beja-Glasser

Nfonoyim

, et al. Complement and microglia mediate early synapse loss in Alzheimer mouse models. Science 2016; 352: 712–716.

18.

Frost

. The role of astrocytes in amyloid production and Alzheimer’s disease. Open Biol 2017; 7: 170228.

19.

Baker

Wijshake

Tchkonia

, et al. Clearance of p16Ink4a-positive senescent cells delays ageing-associated disorders. Nature 2011; 479: 232–236.

20.

Zhang

Kishimoto

Grammatikakis

, et al. Senolytic therapy alleviates Aβ-associated oligodendrocyte progenitor cell senescence and cognitive deficits in an Alzheimer’s disease model. Nat Neurosci 2019; 22: 719–728.

21.

Swerdlow

Burns

Khan

. The Alzheimer’s disease mitochondrial cascade hypothesis: progress and perspectives. Biochim Biophys Acta 2014; 1842: 1219–1231.

22.

Fang

Hou

Palikaras

, et al. Mitophagy inhibits amyloid-β and tau pathology and reverses cognitive deficits in models of Alzheimer’s disease. Nat Neurosci 2019; 22: 401–412.

23.

Calsolaro

Edison

. Alterations in glucose metabolism in Alzheimer’s disease. Recent Pat Endocr Metab Immune Drug Discov 2016; 10: 31–39.

24.

Sierksma

Escott-Price

De Strooper

. Translating genetic risk of Alzheimer’s disease into mechanistic insight and drug targets. Science 2020; 370: 61–66.

25.

Karch

Goate

. Alzheimer’s disease risk genes and mechanisms of disease pathogenesis. Biol Psychiatry 2015; 77: 43–51.

26.

GTEx Consortium. The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science 2020; 369: 1318–1330.

27.

Lake

Chen

Sos

, et al. Integrative single-cell analysis of transcriptional and epigenetic states in the human adult brain. Nat Biotechnol 2018; 36: 70–80.

28.

Mathys

Davila-Velderrain

Peng

, et al. Single-cell transcriptomic analysis of Alzheimer’s disease. Nature 2019; 570: 332–337.

29.

Spires-Jones

Hyman

. The intersection of amyloid beta and tau at synapses in Alzheimer’s disease. Neuron 2014; 82: 756–771.

30.

Lacor

Buniel

Chang

, et al. Synaptic targeting by Alzheimer’s-related amyloid β oligomers. J Neurosci 2004; 24: 10191–10200.

31.

Mucke

Selkoe

. Neurotoxicity of amyloid β-protein: synaptic and network dysfunction. Cold Spring Harb Perspect Med 2012; 2: a006338.

32.

Sheng

Sabatini

Südhof

. Synapses and Alzheimer’s disease. Cold Spring Harb Perspect Biol 2012; 4: a005777.

33.

Jin

Koeglsperger

, et al. Soluble Aβ oligomers inhibit long-term potentiation through a mechanism involving excessive activation of extrasynaptic NR2B-containing NMDA receptors. J Neurosci 2011; 31: 6627–6638.

34.

Shankar

Mehta

, et al. Amyloid-β protein dimers isolated directly from Alzheimer’s brains impair synaptic plasticity and memory. Nat Med 2008; 14: 837–842.

35.

Thomas

Huganir

. MAPK Cascade signalling and synaptic plasticity. Nat Rev Neurosci 2004; 5: 173–183.

36.

Hwang

Yoon

Kang

. A comparative analysis of the cell biology of senescence and aging. Cell Mol Life Sci 2009; 66: 2503–2524.

37.

De Strooper

Karran

. The cellular phase of Alzheimer’s disease. Cell 2016; 164: 603–615.

38.

Kelley

Miller

Todd

, et al. YPEL3, A p53-regulated gene that induces cellular senescence. Cancer Res 2010; 70: 3566–3575.

39.

Chauhan

Goodwin

Chauhan

, et al. ZKSCAN3 Is a master transcriptional repressor of autophagy. Mol Cell 2013; 50: 16–28.

40.

Hebbar

Wang

Rangnekar

. Mechanisms of apoptosis by the tumor suppressor Par-4. J Cell Physiol 2012; 227: 3715–3721.

41.

Van Steensel

Smogorzewska

De Lange

. TRF2 Protects human telomeres from end-to-end fusions. Cell 1998; 92: 401–413.

42.

Butterfield

Halliwell

. Oxidative stress, dysfunctional glucose metabolism and Alzheimer disease. Nat Rev Neurosci 2019; 20: 148–160.

43.

Talbot

Wang

Kazi

, et al. Demonstrated brain insulin resistance in Alzheimer’s disease patients is associated with IGF-1 resistance, IRS-1 dysregulation, and cognitive decline. J Clin Invest 2012; 122: 1316–1338.

44.

Caccamo

Majumder

Richardson

, et al. Molecular interplay between mammalian target of rapamycin, amyloid-β, and tau: effects on cognitive impairments. J Biol Chem 2010; 285: 13107–13120.

45.

Hoeffer

Capetillo-Zarate

, et al. Dysregulation of the mTOR pathway mediates impairment of synaptic plasticity in a mouse model of Alzheimer’s disease. PLoS One 2010; 5: e12845.

46.

Oddo

. The role of mTOR signaling in Alzheimer disease. Front Biosci (Schol Ed) 2012; 4: 941–952.

47.

Liang

Doré

Rowe

, et al. Clinical evidence for GLP-1 receptor agonists in Alzheimer’s disease: a systematic review. J Alzheimers Dis Rep 2024; 8: 777–789.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.05 MB