🧬 biology

Disease-Specific Perturbational Convergence Between Genetic-Risk and Disease-State Transcriptomic Signatures in Alzheimer's Disease

This study reveals that while Alzheimer's disease genetic risk and end-stage transcriptional changes are largely decoupled at the gene and pathway levels, they converge on a shared, pharmacologically modifiable response landscape enriched for vascular and metabolic agents, suggesting perturbational profiling as a promising framework for future therapeutic discovery.

Original authors: Gareth Williams

Published 2026-10-12

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Gareth Williams

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Alzheimer's disease is a condition that slowly erodes memory and thinking, leaving millions of families searching for answers. For decades, scientists have tried to find the root cause by looking at two different sides of the same coin. On one side is the genetic blueprint we are born with, the inherited risks that make some people more likely to develop the disease. On the other side is the actual state of the brain when the disease is advanced, a landscape of damaged cells and altered chemical signals found in tissue samples taken after death. Usually, researchers study these two sides separately, hoping that understanding the genetic risk will reveal the biological machinery that goes wrong in the sick brain. The big question has been whether these two views actually match up, or if the genetic warning signs we carry are disconnected from the physical reality of the disease itself.

A new study from King's College London takes a fresh look at this problem by bringing these two perspectives together in a way that had not been done before. The researchers, led by Gareth Williams, wanted to see if the molecular changes predicted by our genes align with the changes actually seen in the brains of people with Alzheimer's. To do this, they built two distinct maps of the disease. The first map was drawn from the brains of people who had passed away with the disease. They analyzed over two million individual brain cells from five different studies to create a precise picture of how specific cell types, like neurons and immune cells, change their behavior when the disease is fully established. This represents the "disease state," the actual biological damage.

The second map was entirely theoretical, built from the genetic code of thousands of healthy people. Using known genetic variations associated with Alzheimer's risk, the team calculated what the gene activity in a brain should look like if it were driven purely by inherited risk, long before any disease symptoms appeared. This represents the "genetic risk" profile. When the scientists compared these two maps directly, looking at which genes were turned on or off, they found almost no overlap. The genes that changed in the sick brains were largely different from the genes that the genetic risk predicted would change. Even when they grouped these genes into biological pathways, the two maps remained disconnected. It appeared that the genetic warning signs we carry and the final state of the disease are speaking different languages at the level of individual genes.

However, the story changed when the researchers stopped looking at the genes themselves and started looking at how drugs affect them. They took both their genetic risk map and their disease-state map and tested them against a massive library of drug effects. This library contained data on how thousands of different compounds change gene activity in cells. The goal was to see if any drug could reverse the patterns found in either map. Surprisingly, while the two maps did not look alike on their own, they pointed toward the same set of drugs. Both the genetic risk profile and the actual disease profile suggested that certain types of medications would have a similar effect on the brain's biology.

The drugs that emerged as the best candidates for reversing these signatures were not the usual suspects for brain diseases. Instead, the analysis highlighted medicines used to treat high blood pressure, dilate blood vessels, and manage diabetes. These are treatments for conditions that are already known to be risk factors for dementia. The study suggests that while the specific genetic instructions and the final disease damage might look different, they both converge on a shared vulnerability: the body's vascular and metabolic health. The researchers found that drugs targeting these systems could theoretically push the brain back toward a healthier state, regardless of whether you started from the genetic risk side or the disease damage side.

This finding offers a new way to think about finding treatments. It suggests that looking for a single "magic bullet" gene to fix might be the wrong approach. Instead, the path forward may lie in understanding how different biological systems interact. The study indicates that the genetic risk for Alzheimer's and the actual disease pathology are largely decoupled at the molecular level, yet they meet again in the realm of how the body responds to medication. The convergence on vascular and metabolic drugs supports the idea that managing heart health and blood sugar might be more than just general wellness advice; they could be direct levers for influencing the brain's resilience.

The researchers were careful to note that this does not mean these drugs are a cure. The study used computer models and existing data to find these patterns, and the results need to be tested in real-world clinical trials. Furthermore, the study explicitly ruled out the idea that the genetic risk and the disease state are simply two views of the exact same molecular process. The lack of overlap at the gene level was clear. Instead, the study proposes a more complex model where different biological forces, starting from genetics and ending in disease, eventually influence the same broad systems in the body. By focusing on these systems rather than individual genes, scientists may find new ways to intervene in Alzheimer's, turning the search for a cure toward the management of the body's overall health.

Technical Summary: Disease-Specific Perturbational Convergence Between Genetic-Risk and Disease-State Transcriptomic Signatures in Alzheimer's Disease

Problem Statement
Alzheimer's disease (AD) research has traditionally treated genetic susceptibility and disease-associated transcriptional changes as separate domains. While genome-wide association studies (GWAS) have identified inherited risk factors, and post-mortem transcriptomics have characterized end-stage pathology, it remains unclear whether genetically inferred molecular mechanisms converge with the transcriptional changes observed in diseased brain tissue. Furthermore, target-based therapeutic approaches have yielded only modest relief, prompting a shift toward large-scale quantitative data analysis and drug repurposing. A critical gap exists in understanding if the molecular processes driving inherited risk are recapitulated in established disease, or if these distinct upstream and downstream processes converge only at a higher-order level, such as pharmacological response.

Methodology
The study employed a hierarchical comparison of two conceptually distinct representations of AD biology:

Disease-State Signature ( $P_d$ ): Cell-type-specific differential expression signatures were generated from five independent single-cell and single-nucleus RNA-sequencing datasets comprising over two million brain cells. These signatures were derived by comparing AD donors to controls across major brain cell classes (neurons, microglia, astrocytes, oligodendrocytes, etc.). The author validated the robustness of these signatures by comparing them against bulk-tissue pathology signatures from a large independent cohort (GSE84422), confirming that cell-autonomous changes are preserved despite cellular composition shifts.
Genetic-Risk Signature ( $P_r$ ): A genetically inferred transcriptional profile was constructed using polygenic risk scores (PRS) for AD and expression quantitative trait loci (eQTL) effect estimates from the GTEx project. These were applied to genomic profiles from 3,202 individuals in the 1000 Genomes cohort to estimate the component of gene expression driven by inherited genetic risk load.
Comparative Analysis: The $P_d$ $P_{d}$ and $P_r$ $P_{r}$ profiles were compared at three levels:
- Gene-level: Direct correlation of differential expression statistics.
- Pathway-level: Gene Set Enrichment Analysis (GSEA) using Reactome pathways.
- Perturbational-level: Projection of both signatures onto drug-response spaces derived from the Connectivity Map (CMap) and LINCS resources. Drug candidates were ranked based on their ability to reverse the disease or risk signatures (anti-correlation).

Key Results

Decoupling at Gene and Pathway Levels: There was minimal concordance between the genetic-risk ( $P_r$ ) and disease-state ( $P_d$ ) signatures at the individual gene level. Only 5 genes were shared among the top 500 ranked genes from each analysis. Similarly, pathway enrichment analysis revealed essentially no overlap in significantly dysregulated Reactome pathways between the two profiles.
Convergence in Perturbational Space: In contrast to the gene and pathway levels, a significant positive correlation ( $r = 0.39$ ) was observed when projecting both signatures onto drug-response space. This indicates that while the specific molecular drivers of risk and pathology differ, they identify overlapping regions of pharmacologically modifiable biology.
Specificity of the Signal: This perturbational alignment was specific to AD. When similar analyses were performed for Parkinson's disease, major depressive disorder, and schizophrenia, no significant positive correlation was found between their genetic-risk and disease-state drug scores.
Enrichment of Vascular and Metabolic Agents: The compounds showing consistent reversal signatures across both $P_d$ and $P_r$ were enriched for vasodilators, antihypertensives, and metabolic agents (e.g., tolazamide). These drug classes correspond to established epidemiological risk factors for dementia (vascular health, type 2 diabetes).
Robustness: The perturbational correlation persisted ( $r = 0.29$ ) when the analysis was restricted to individuals of European ancestry, suggesting the finding is not an artifact of population stratification.
Validation of Framework: The framework correctly identified compounds known to induce neurotoxicity (e.g., Hsp90 inhibitors, microtubule disruptors) as having signatures that drive gene expression in the same direction as AD pathology, supporting the biological validity of the bidirectional ranking.

Significance and Claims
The author concludes that genetic risk and end-stage AD pathology are largely decoupled at the level of individual genes and biological pathways but converge on a shared, pharmacologically modifiable response landscape. This convergence aligns with biological mechanisms linked to established vascular and metabolic dementia risk factors.

The paper posits that perturbational profiling offers a potential framework for future therapeutic investigation by bridging genetically informed disease risk and complex molecular phenotypes. However, the author maintains a modest stance, explicitly stating that these findings should not be interpreted as evidence for immediate therapeutic efficacy or as a basis for direct drug prioritization without further validation. Instead, the study suggests that the relationship between genetic susceptibility and late-stage disease may not be adequately captured by direct transcriptional state comparisons, but rather by distinct upstream and downstream processes that converge on cellular states similarly modifiable by pharmacological perturbation. The work highlights the utility of expression-based drug repurposing as a functional bridge in complex neurodegenerative disorders, while acknowledging limitations regarding the use of immortalized cell lines in reference databases and the inherent noise in eQTL-based expression prediction.

More like this