Context-dependent regulatory variants in Alzheimer’s disease

Ziheng Chen, Yaxuan Liu, Ashley R. Brown, Heather H. Sestili, Easwaran Ramamurthy, Xushen Xiong, Dmitry Prokopenko, BaDoi N. Phan, Lahari Gadey, Peinan Hu, Li‐Huei Tsai, Lars Bertram, Yoshihide Hayashizaki, Rudolph E. Tanzi, Manolis Kellis, Andreas R. PfenningbioRxiv 2025 · DOI

TL;DR

This paper tackles the challenge of understanding how non-coding genetic variants contribute to Alzheimer's disease (AD). By combining advanced reporter assays and machine learning, the researchers reveal that the impact of these variants is highly dependent on the specific cell type and its state, offering a more nuanced view of AD's genetic underpinnings.

What Problem Does This Paper Address?

Late-onset Alzheimer's disease (LOAD) is a complex genetic disorder, with a significant portion of its genetic risk attributed to non-coding regions of the genome. These non-coding variants are thought to influence gene regulation by affecting how transcription factors (TFs) bind to DNA. However, understanding their precise role is difficult because regulatory elements (like enhancers) can behave differently depending on the cell type (e.g., microglia vs. neurons) and the cell's current state (e.g., resting vs. inflamed). This context-dependency makes it hard to pinpoint which variants are truly causal and how they contribute to disease. The paper aims to systematically map these context-dependent effects of both common and rare non-coding variants associated with LOAD.

Methods

The researchers employed a multi-pronged approach:

  • Massively Parallel Reporter Assays (MPRAs): This technique allows for the high-throughput testing of thousands of DNA sequences (cis-regulatory elements or CREs) to measure their regulatory activity. They used both in vitro (in cell cultures) and in vivo (in mouse brains) MPRAs.
  • Context-Specific MPRA: They tested CREs in various immune states (resting, IFN-β, IFN-γ, LPS+IFN-γ) in THP-1 macrophages and in different brain regions (cortex, hippocampus, striatum) of mice.
  • Machine Learning Models: They trained convolutional neural networks (CNNs) on ATAC-seq data (which maps open chromatin regions, indicating regulatory potential) from multiple cell types and states. These models predict chromatin accessibility and variant effects.
  • Interpretable AI (SHAP values): To understand why the models made certain predictions, they used SHAP values to identify the contribution of specific DNA sequences and TF motifs to regulatory activity.
  • CRISPR Interference (CRISPRi): To functionally validate the impact of a specific regulatory element and its variants, they used CRISPRi to silence an enhancer in the SEC63-OSTM1 locus and then analyzed the downstream gene expression changes using RNA-seq.

Datasets

  • ATAC-seq data: From various immune cell types (THP-1 macrophages in different states, HMC3 microglia) and mouse brain regions (cortex, hippocampus, striatum).
  • RNA-seq data: From THP-1 macrophages in different states and after CRISPRi perturbations.
  • LOAD-associated variants: A curated set of 599 variants from GWAS and WGS studies, selected based on statistical association strength and overlap with open chromatin regions.
  • Publicly available epigenomic data: From the Epigenome Roadmap database.

Key Insights and Ideas

  • Context is King: The same genetic variant can have opposite effects on gene regulation depending on whether it's in an immune cell or a neuron, or even in different states of the same cell type.
  • Common vs. Rare Variants: Common variants associated with LOAD seem to primarily influence immune cells, while rare variants have a more pronounced impact in brain contexts.
  • Beyond Simple Motif Disruption: The impact of variants isn't just about disrupting a single TF binding site; it also involves subtle tuning of TF interactions and the broader "motif context" within regulatory elements.
  • Enhancers as Gatekeepers: Specific enhancers can act as crucial regulators, controlling complex pathways like inflammation and amyloid processing, and rare variants within them can have significant effects.

Key Results

  • Differential Effects: They found that individual variants could differentially and even oppositely modulate regulatory function across cell types and states.
  • Immune vs. Brain Bias: Common variants showed stronger effects in immune contexts, while rare variants had more impact in brain contexts.
  • TF Motif Importance: Machine learning models identified cell-type- and state-specific TF motifs that collectively shape regulatory signatures.
  • SEC63-OSTM1 Enhancer Function: CRISPRi knockdown of an enhancer in the SEC63-OSTM1 locus unexpectedly increased expression of distal genes (GSAP and SSH1), highlighting its role as a gatekeeper for innate immune signaling and potentially influencing LOAD pathogenesis.
  • High Reproducibility: In vitro and in vivo MPRA experiments showed high reproducibility, and machine learning predictions largely agreed with experimental data.

Limitations and Future Work

  • Cell Line Limitations: THP-1 and HMC3 cells are surrogates for primary microglia and macrophages and may not fully capture the complexity of the brain's immune milieu.
  • In Vivo MPRA Nuances: In vivo MPRAs in mice might introduce cross-species regulatory differences.
  • Long-Range Interactions: Synthetic CREs might miss contributions from long-range chromatin contacts or topological features.
  • Splicing Effects: Some variants might alter splicing without affecting MPRA output, requiring complementary assays.
  • Brain Region Specificity: While they identified variants with brain-specific effects, they lacked the statistical power to resolve region-specific effects within the brain.
  • Single-Cell Resolution: Future work could benefit from single-cell resolution MPRA platforms to capture context-specific effects in neural subtypes or immune activation states not represented in their current in vitro systems.
  • Future Directions: The authors suggest systematic enhancer-gene mapping, investigating motif-motif interactions, functional studies of hotspot genes, and defining cellular states where variants become active to guide therapeutic strategies.

Summary updated

Abstract

Noncoding genetic variants underlie many complex diseases, yet identifying and interpreting their functional impacts remains challenging. Late-onset Alzheimer's disease (LOAD), a polygenic neurodegenerative disorder, exemplifies this challenge. The disease is strongly associated with noncoding variation, including common variants enriched in microglial enhancers and rare variants that are hypothesized to influence neurodevelopment and synaptic plasticity. These variants often perturb regulatory sequences by disrupting transcription factor (TF) motifs or altering local TF interactions, thereby reshaping gene expression and chromatin accessibility. However, assessing their impact is complicated by the context-dependent functions of regulatory sequences, underscoring the need to systematically examine variant effects across diverse tissues, cell types, and cellular states. Here, we combined <i>in vitro</i> and <i>in vivo</i> massively parallel reporter assays (MPRAs) with interpretable machine-learning models to systematically characterize common and rare variants across myeloid and neural contexts. Parallel profiling of variants in four immune states <i>in vitro</i> and three mouse brain regions <i>in vivo</i> revealed that individual variants can differentially and even oppositely modulate regulatory function depending on cell-type and cell-state contexts. Common variants associated with LOAD tended to exert stronger effects in immune contexts, whereas rare variants showed more pronounced impacts in brain contexts. Interpretable sequence-to-function deep-learning models elucidated how genetic variation leads to cell-type-specific differences in regulatory activity, pinpointing both direct transcription-factor motif disruptions and subtler tuning of motif context. To probe the broader functional consequences of a locus prioritized by our reporter assays and models, we used CRISPR interference to silence an enhancer within the <i>SEC63-OSTM1</i> locus that harbors four functional rare variants, revealing its gatekeeper role in inflammation and amyloidogenesis. These findings underscore the context-dependent nature of noncoding variant effects in LOAD and provide a generalizable framework for the mechanistic interpretation of risk alleles in complex diseases.
BibTeX
@misc{chen2025context,
  author = {Ziheng Chen and Yaxuan Liu and Ashley R. Brown and Heather H. Sestili and Easwaran Ramamurthy and Xushen Xiong and Dmitry Prokopenko and BaDoi N. Phan and Lahari Gadey and Peinan Hu and Li‐Huei Tsai and Lars Bertram and Yoshihide Hayashizaki and Rudolph E. Tanzi and Manolis Kellis and Andreas R. Pfenning},
  title = {{Context-dependent regulatory variants in Alzheimer’s disease}},
  year = {2025},
  doi = {10.1101/2025.07.11.659973},
  howpublished = {bioRxiv},
}
Figure