Brain microRNAs are associated with variation in cognitive trajectory in advanced age

Cognitive performance in older adults follows diverse trajectories from stability to rapid decline, influencing dementia risk and age of onset for MCI or dementia. Traditional neuropathologies (β-amyloid plaques, neurofibrillary tangles, micro- and macroinfarcts, Lewy bodies) explain only about 40% of the variance in cognitive trajectory, suggesting additional mechanisms contribute. MicroRNAs (miRNAs) are small non-coding RNAs that regulate gene expression post-transcriptionally and can act as master regulators affecting hundreds of mRNAs. miRNAs have established roles in synaptic plasticity, neurodegenerative pathology aggregation, neuronal survival, learning, memory, and were previously linked to Alzheimer’s disease pathologies (β-amyloid and tau) in cortical tissue. The study hypothesized that brain miRNAs and their downstream mRNA/protein targets contribute to inter-individual variation in cognitive trajectory in advanced age, beyond known neuropathologies, and aimed to identify such miRNAs and characterize their downstream networks using integrated miRNA, transcriptomic, and proteomic analyses from dorsolateral prefrontal cortex (dIPFC).

Prior work shows that common neuropathologies account for only about 40% of variance in late-life cognitive decline, indicating substantial unexplained variation likely due to other biological processes and co-occurring pathologies. miRNAs are key post-transcriptional regulators; each can modulate numerous targets and influence gene expression networks involved in synaptic function and memory. In cortical tissue, miRNAs have been associated with amyloid and tau pathology, and miR-132 has been linked to tau metabolism and AD pathology. Plasma and brain studies have associated altered miR-132 expression with MCI and AD. Age-related decreases in miR-132 expression and roles in learning, neuronal morphogenesis, and inflammation have been reported. However, a comprehensive, global analysis of miRNA associations with longitudinal cognitive trajectories, integrating transcriptomic and proteomic targets, had not been performed prior to this study.

Study design: A discovery and replication global miRNA association study of cognitive trajectory using ROS and MAP cohorts, followed by meta-analysis. Integrative analyses linked significant miRNAs to transcriptome and proteome profiles from dIPFC. Additional PWAS meta-analysis used independent cohorts (Banner, BLSA) for protein associations. Participants: Discovery N=454 and replication N=134 ROS/MAP participants with annual cognitive evaluations (median ~7 years, up to ~16). Inclusion: ≥1 follow-up, no dementia at baseline. Proteomics participants were from Banner and BLSA cohorts with longitudinal neurological/neuropsychological testing. Cognitive trajectory: Person-specific slopes from linear mixed-effects models. In ROS/MAP, global cognition (from 17 tests: episodic memory, perceptual orientation/speed, semantic, working memory) was the longitudinal outcome with follow-up year predictor, adjusting for age at recruitment, sex, education. In Banner/BLSA, MMSE was the outcome with follow-up year predictor, adjusting for sex, education, age; random intercept and slope per subject. Neuropathologies: Eight cerebral pathologies assessed in ROS/MAP: neurofibrillary tangles, β-amyloid, Lewy bodies, gross infarct, microinfarcts, cerebral atherosclerosis, cerebral amyloid angiopathy, hippocampal sclerosis. miRNA profiling: Total RNA (including miRNA) from dIPFC of ROS/MAP. Nanostring nCounter Human miRNA platform; retained miRNAs with call rate ≥95% and absolute value >15 in ≥50% samples. Batch effects removed using ComBat. 292 miRNAs passed QC for association study. Transcriptomics: RNA-seq from dIPFC (Illumina HiSeq). STAR alignment to GRCh38; gene-level counts generated. Genes with <1 CPM in ≥50% samples or missing length/GC removed; 15,582 genes analyzed. Cell-type proportions (neurons, astrocytes, oligodendrocytes, microglia) estimated with CIBERSORT using Darmanis signatures to adjust for tissue heterogeneity. Proteomics: dIPFC proteomes from Banner and BLSA cohorts. Proteins quantified in ≥90% samples retained (Banner 3,710; BLSA 3,933). Log2 transform, ComBat batch correction; age at death, sex, PMI effects regressed using bootstrap regression. Statistical analyses: Global miRNA association with cognitive trajectory in discovery and replication sets using limma, adjusting for sex, age at death, RIN, PMI, study (ROS vs MAP), and cell-type proportions (cell types not available in replication). Meta-analysis with METAL using effect sizes and SEs. Multiple testing: Benjamini–Hochberg FDR. Pairwise miRNA correlations from normalized residualized miRNA profiles (regressed on covariates), Pearson correlations with multiple-testing adjustment. VariancePartition (fixed-effect) used to estimate percent variance in cognitive trajectory explained by individual miRNAs and each pathology, jointly modeling variables and adjusting for covariates. Module correlations: Spearman correlations between normalized miRNA levels and 47 co-expression gene modules (SpeakEasy network) from dIPFC; module represented by mean expression; FDR correction. Transcriptome-wide differential expression (DE): voom-limma with sex, age at death, study, RIN, PMI, RNA-seq batch, and cell-type proportions; FDR control. Putative miRNA targets identified by intersecting DE genes with TargetScan v7.2 predicted targets for miR-132-3p and miR-29a-3p. Proteome-wide association study (PWAS): Conducted separately in Banner and BLSA via linear regression of cognitive trajectory on normalized protein abundance (sex, age, PMI regressed out of proteome; sex, age, education regressed in trajectory). Meta-analysis with METAL; FDR control. Putative protein-level targets identified by intersecting PWAS-significant proteins (FDR<0.05) with TargetScan predicted targets. Experimental validation: Luciferase reporter assays (HEK293T) for selected 3'UTR constructs of putative targets of miR-132-3p and miR-29a-3p. Co-transfection with pre-miRNA vs control; rescue experiments with mutant 3'UTRs to assess direct targeting.

• Six miRNAs associated with cognitive trajectory (meta-analysis, adjusted p<0.05): miR-132-3p, miR-129-5p, miR-129-3p (higher in stable trajectories), and miR-29a-3p, miR-99b, miR-19b (lower in stable trajectories).
• After adjusting for eight cerebral pathologies, four miRNAs remained significant: miR-132-3p, miR-129-5p, miR-129-3p, miR-29a-3p. In a joint model, only miR-132-3p and miR-29a-3p retained significance, indicating independent contributions beyond pathologies.
• Variance explained: Using normalized miRNA profiles, miR-132-3p explained 18.2% and miR-29a-3p 1.6% of cognitive trajectory variance; after regressing out the eight pathologies, miR-132-3p explained 11.8% and miR-29a-3p 2.0%.
• miR-132-3p and miR-29a-3p were mildly inversely correlated (r = -0.20; adjusted p = 3.22×10^-5), suggesting largely independent effects.
• Module associations: miR-132-3p associated with 24/47 gene co-expression modules (including top modules m109, m13, m7, m127, m131 linked to cognitive decline), with correlation coefficients ranging from about -0.32 to 0.28. miR-29a-3p associated with 3 modules (m128, m12, m8), previously linked to cognitive trajectory and tangle burden.
• Transcriptome DE related to cognitive trajectory identified 1,087 downregulated and 797 upregulated transcripts (FDR<0.05). Intersecting with TargetScan predictions yielded putative transcript-level targets: 64 for miR-132-3p and 177 for miR-29a-3p.
• Proteome meta-analysis (independent cohorts) identified 229 proteins decreased and 350 increased in association with trajectory (FDR<0.05). Intersecting with TargetScan yielded putative protein-level targets: 22 for miR-132-3p and 41 for miR-29a-3p. Overlaps across transcript and protein levels included MECP2 and RPH3A (miR-132-3p) and SLC30A3, PDHX, HDGF, DIRAS1 (miR-29a-3p). There were 18 transcripts and 2 proteins as common targets of both miR-132-3p and miR-29a-3p.
• Network context: ~45% (miR-132-3p) and ~41% (miR-29a-3p) of protein targets were hub proteins in Banner WGCNA modules. miR-29a-3p targets were enriched in modules related to myelination (M4), mitochondrial function (M3), catabolic/apoptotic processes (M2), and synaptic functions (M1).
• Experimental validation: All 11 tested miR-132-3p targets (MAPT, MECP2, MAPK1, MAPK3, RDX, GMPR, ANKRD29, DPYSL3, EIF4A2, PEA15, DKK3) showed significantly reduced luciferase activity upon miR-132 overexpression; rescue assays supported direct targeting of PEA15 and MAPK3. For miR-29a-3p, 8/10 tested targets (AKAP5, PALM, PURA, GSK3B, SLC25A22, SH3GLB2, SYT7, PDHX) showed reduced activity; rescue assays supported PURA and GSK3B as direct targets.
• Secondary analyses: miR-132-3p associated with β-amyloid and tangles; no significant miRNAs for Lewy bodies, infarcts, atherosclerosis, amyloid angiopathy, or hippocampal sclerosis, suggesting miR-132-3p influences trajectory partly via amyloid/tau and partly via pathology-independent mechanisms.

The study demonstrates that specific brain miRNAs, particularly miR-132-3p and miR-29a-3p, are robustly associated with longitudinal cognitive trajectories in advanced age, largely independent of eight common cerebral pathologies. These findings extend prior observations linking miRNAs to AD pathologies by showing direct relevance to person-specific rates of cognitive change. miR-132-3p exhibits widespread associations across gene co-expression modules, consistent with a master-regulator role influencing multiple biological pathways related to cognition. Integrative transcriptomic and proteomic analyses identify plausible downstream effectors, many of which are hub proteins in brain protein networks, highlighting potential mechanistic routes through synaptic function, myelination, mitochondrial processes, and apoptosis. Experimental validation supports direct targeting of select genes by miR-132-3p and miR-29a-3p. Together, these results suggest that modulation of these miRNAs or their key targets could represent avenues to enhance cognitive resilience independent of classical neuropathologies.

This first global miRNA study of cognitive trajectory identifies miR-132-3p and miR-29a-3p as major, pathology-independent contributors to inter-individual differences in late-life cognitive change, explaining a substantial fraction of variance. Multi-omics integration maps extensive downstream networks at transcript and protein levels, with experimental support for direct targets. These findings provide a framework for mechanistic studies of cognitive resilience and potential therapeutic strategies aimed at miRNA pathways. Future work should employ small RNA-sequencing to broaden miRNA coverage, longitudinal sampling where possible, causal inference approaches (e.g., perturbation studies, Mendelian randomization), and validation of protein targets in brain-relevant systems and cohorts.

• Observational association study: no causal inference can be made.
• Postmortem cross-sectional profiling may not reflect temporal dynamics of miRNA–mRNA–protein relationships during cognitive change.
• Discordance between mRNA and protein levels limits overlap of identified targets across modalities.
• Proteomic datasets (Banner, BLSA) were independent of the ROS/MAP miRNA cohorts, making cross-omic target identification tentative and requiring further validation.
• Validation assays did not directly quantify endogenous protein level changes in brain tissue.
• Nanostring platform captures fewer miRNAs than small RNA-seq, potentially missing additional relevant miRNAs.
• Complex, potentially bidirectional interactions among miRNAs and targets may obscure canonical repression patterns.