For decades, much of the human genome sat under an unflattering label: “junk DNA”. The term was always more shorthand than a scientific verdict—researchers have long known that many non-coding stretches help regulate when and where genes turn on. However, the sheer scale of the non-coding genome, and the difficulty of proving what each fragment does in a particular cell type, has kept many of these elements effectively invisible.
New work led by researchers at the University of New South Wales (UNSW) shifts that picture for Alzheimer’s disease. In a study highlighted by ScienceDaily, the team experimentally tested nearly 1,000 candidate regulatory DNA “switches” in human astrocytes—support cells in the brain that are increasingly studied in neurodegeneration. Around 150 sequences showed enhancer activity (that is, they measurably changed gene expression in the assay), and many of the functional enhancers controlled genes implicated or dysregulated in Alzheimer’s, though linking a regulatory element to its true target still requires careful follow-up.
The headline message is simple but consequential: many genetic changes linked to Alzheimer’s are not altering proteins directly. Instead, they may be affecting the controls that regulate gene activity—often in specific brain cell types.
Why Alzheimer’s genetics keeps pointing outside genes
Large genetic studies have repeatedly found that common variants associated with complex diseases often fall in non-coding DNA, rather than in protein-coding exons. The US National Human Genome Research Institute explains why this is common—and why it is hard to interpret—on its Genome-Wide Association Studies (GWAS) fact sheet: GWAS identifies genomic neighbourhoods linked to disease, but the causal variant and mechanism are often unclear, particularly when the signal sits in regulatory DNA.
Alzheimer’s disease is a clear example. Curated results in the GWAS Catalog’s Alzheimer’s disease page show many reported associations landing in introns or intergenic regions—places that do not encode proteins but can host enhancers, silencers and other regulatory elements.
That creates a bottleneck. If a variant does not change a protein sequence, you cannot easily infer its effect. You have to measure whether it changes gene regulation—and, crucially, in the right cell type. Brain tissue is not a single entity; it is an ecosystem of neurons, astrocytes, microglia, oligodendrocytes and vascular cells, each with distinct regulatory wiring. A switch that matters in one cell type may do little in another.
Astrocytes: support cells with a growing Alzheimer’s profile
Astrocytes have traditionally been cast as the brain’s caretakers—maintaining chemical balance, recycling neurotransmitters and supporting synapses. In Alzheimer’s research, attention has often centred on neurons (where memory circuitry lives) and microglia (immune-like cells linked to inflammation and plaque responses). Yet astrocytes are also increasingly investigated as active participants in disease biology, including through shifts into inflammatory or dysfunctional states that may affect neuronal health.
A detailed overview in Nature Reviews Neurology—“Astrocytes in Alzheimer’s disease”—summarises evidence that astrocytes can contribute to synaptic dysfunction, metabolic stress and inflammatory signalling in ways that intersect with hallmark Alzheimer’s processes. This makes them a compelling place to look for regulatory switches that might connect genetic risk to downstream biology.
The UNSW team’s approach aligns with a broader move in genomics: instead of asking “which gene is near this risk variant?”, researchers increasingly ask “in which cell type does this DNA region act as a switch, and what does it switch on?”
Stress-testing ~1,000 switches with CRISPRi + single-cell RNA-seq
The UNSW-led team used CRISPR interference (CRISPRi) to silence hundreds of candidate enhancers in cultured human astrocytes and then measured gene-expression changes with single-cell RNA sequencing, allowing them to test nearly 1,000 enhancers at once. In the Nature Neuroscience paper, the authors report functionally testing 979 candidate enhancers and identifying more than 150 enhancer–gene regulatory interactions, including enhancers controlling genes implicated in Alzheimer’s disease.
In research published in Nature Neuroscience, the team reports a large-scale CRISPRi enhancer screen in human astrocytes.
According to the UNSW summary, close to 1,000 candidate switches were assayed, and about 150 showed measurable regulatory activity in astrocytes. That ratio—only a minority of plausible-looking sequences proving functional—serves as a reality check for the field. Computational predictions and epigenomic maps can nominate potential enhancers, but experimental validation remains the standard for demonstrating that a given sequence can drive gene expression in a given cell type under test conditions.
The study also points to a more nuanced outcome than a simple on/off list. Some sequences may only work under certain cellular states, developmental stages or inflammatory conditions. MPRA typically captures activity in a controlled experimental setting, so translating results to the complexity of an ageing human brain still requires care. Even so, establishing a validated set of astrocyte-active enhancers at Alzheimer’s-associated loci is a useful step.
Linking active switches to known risk genes—carefully
One of the most intriguing findings, as described by UNSW, is that many of the validated switches sit near genes already tied to Alzheimer’s risk. This supports a plausible regulatory mechanism: a non-coding variant could subtly change enhancer strength, shifting expression of a nearby (or sometimes more distant) gene, and over years contribute to disease vulnerability.
It is also where precision matters. Proximity does not guarantee causality: enhancers can regulate genes tens to hundreds of kilobases away, sometimes skipping over the nearest gene. As genomics researchers often note, connecting an enhancer to its true target can require additional evidence, such as chromatin-conformation data (which maps physical DNA contacts) or perturbation experiments (such as CRISPR interference).
The Nature Genetics paper provides the more formal framing: MPRA identifies regulatory activity and can detect allele-specific effects for variants embedded in tested sequences, but it does not automatically prove that a particular enhancer–gene–disease chain is causal in people. Even so, the Alzheimer’s field has increasingly investigated the idea that regulatory changes—especially those acting in specific brain cell types—help explain why GWAS variants cluster in non-coding regions.
In other words, this work does not “solve” Alzheimer’s genetics. It refines the map by turning swathes of association signals into experimentally supported regulatory elements that can be prioritised for deeper functional follow-up.
A dataset designed to make AI less wrong about gene control
A practical outcome of the project is a high-quality training resource for computational models. The UNSW team notes that their dataset is already being used to train AI systems to better predict how DNA sequences control gene activity, particularly in astrocytes.
This matters because AI models in regulatory genomics are only as good as their ground truth. Vast amounts of genomic sequence exist, but experimentally validated labels—“this sequence acts as an enhancer in this cell type”—are scarce and expensive to generate. MPRA produces precisely the kind of labelled examples machine-learning models need: many sequences tested under standardised conditions, with quantitative readouts.
Better models could help researchers triage which of the millions of non-coding variants in the human population are most likely to matter, and in which cell types. Even then, predictions are not a substitute for bench work; they are intended to narrow the search space so experimentalists can focus on the most promising candidates.
As with any AI-assisted science, there are limits. A model trained on astrocyte MPRA data may not generalise well to neurons, microglia or vascular cells, and MPRA itself tests DNA outside its native chromatin context. Still, validated functional datasets are an important input for making genome interpretation faster and less speculative.
What this changes—and what it doesn’t (yet)
It is tempting to translate “we found 150 functional switches” into “we found 150 new drug targets”. That would be premature. Enhancers are not straightforward targets in the way enzymes or receptors can be, and Alzheimer’s is a multifactorial disease with long timelines and multiple intersecting pathways.
What the study does change is the plausibility structure of Alzheimer’s risk genetics. It strengthens the case that at least some risk variants may act by shifting gene regulation in astrocytes, and it provides an experimentally grounded shortlist of regulatory elements for the field to interrogate. Future work can now ask sharper questions: Which variants alter enhancer activity? Which genes do those enhancers regulate in native chromatin? Under what astrocyte states—resting, inflammatory, ageing—do the switches matter most?
More broadly, recent Alzheimer’s research has increasingly emphasised cell-type-specific functional genomics, partly because “brain tissue” averages away signals that may be specific to particular cell populations. The UNSW study, anchored by MPRA validation in human astrocytes, fits within that direction of travel.
Neutral wrap-up? “Junk DNA” is not junk so much as circuitry—dense, context-dependent and hard to decode. By experimentally verifying which non-coding switches work in human astrocytes, UNSW researchers have provided a clearer path from genetic association to biological mechanism, while also generating a dataset that could improve AI-assisted genome interpretation.