The amyloid era
Sweep up the rubbish. The disease will follow.
- Amyloid plaques assumed to cause the disease
- 30 years of plaque-clearing drug development
- >99% of trials failed in late-stage testing
- Neurons were treated as the main characters
PhD proposal · 2026–2030
In a healthy brain, your immune cells are gardeners.
In Alzheimer's, a broken switch turns them into
DAM cells.
We are looking for that switch — and the button that could turn it off again. This site is the public-facing companion to a 4-year PhD at the UK Dementia Research Institute at Imperial.
First — a quick primer
Meet the gardener.
Your brain isn't only neurons. About one in every ten cells is a microglia — a tiny immune cell that, in a healthy brain, behaves like a gardener: pruning weak connections, clearing waste, keeping the place tidy. There are roughly 8 billion of them patrolling your skull right now.
In Alzheimer's, something flips them out of "gardener" mode and into the disease-associated state — aggressive, inflamed, and attacking the very cells they should be protecting. See the switch ↓
What's happening here?
We zoom from a whole brain into a single patch of tissue. Most of the cells you can see are neurons — the long, spiky cells that fire signals. The pink branching cell is a microglia: a tiny immune cell that crawls between neurons, prunes weak connections, and clears debris.
About 1 in every 10 brain cells is a microglia. You have ~8 billion of them right now. In a healthy brain they're gardeners; in Alzheimer's, they switch into disease-associated microglia (DAM) — and that's the switch this PhD is mapping.
The evidence
Three different methods. The same answer.
Three completely separate scientific tools have, independently, pointed to the same suspect: microglia. None of them was looking for microglia when they started.
1 · Genetics
80+ loci
The DNA changes that raise your risk land near microglia genes.
Comparing the genomes of half a million people, scientists have found 80+ regions of DNA that change your odds of getting Alzheimer's. The genes nearest those regions — TREM2, CD33, INPP5D, MS4A6A, MEF2C — are almost all switched on in microglia, not neurons.
Explore the genetics →
2 · Cell biology
96% in microglia
The genes that misbehave in Alzheimer's brains misbehave in microglia.
Modern lab tools can ask one cell at a time which genes are on or off. When researchers re-analysed Alzheimer's brains carefully, only 26 genes truly behaved differently from healthy brains — and 25 of those 26 changes happened in microglia.
Explore the cell data →
3 · Statistics
4.1σ
Statistically, microglia top every Alzheimer's gene-list we test.
Our supervisor built a tool called EWCE that ranks which brain cell types a disease is hitting. For Alzheimer's, the microglia score is more than 4 standard deviations above any other cell — the strongest, cleanest signal in the data.
Explore the statistics →
The synthesis
Three roads. One destination.
Genetics asked "where does the risk live in DNA?". Cell biology asked "which cells misbehave?". Statistics asked "which cells carry the gene-list?". All three roads end at microglia.
What's happening here?
Three completely independent scientific approaches were each asked a different question about Alzheimer's:
GWAS (genome-wide association studies) compared the DNA of half a million people and found 80+ regions of risk — the 29 strongest all sit near genes that are switched on in microglia. snRNA-seq (single-nucleus RNA sequencing) looked at one cell at a time and found that 26 genes misbehave in Alzheimer brains; 25 of those misbehave inside microglia. EWCE is a statistics tool that ranks which cell type a disease is hitting; for Alzheimer's, microglia score 4.1 standard deviations higher than any other brain cell.
Three different methods. Three different kinds of evidence. They all converge on the same suspect.
From DNA letter to disease risk
How does a single letter of DNA raise your risk?
Most Alzheimer's risk variants don't sit inside genes themselves. They sit in nearby regulatory DNA — the dimmer switches that decide how loudly each gene speaks. Tweak the dimmer, and a microglia starts behaving differently. Repeat that across many genes and the gardener slides into the disease-associated state.
What's happening here?
This is the chain that turns a tiny inherited DNA difference into measurable disease risk — five steps:
1. One letter of your three-billion-letter genome is different from the average person. We call that a SNP. Most SNPs sit in non-coding DNA, not inside genes. 2. The SNP lands inside a regulatory enhancer — a stretch of DNA that decides how loudly a nearby gene speaks. Enhancers are like dimmer switches. 3. The dimmer is now slightly tweaked, so the gene it controls is expressed at a higher (or lower) level. 4. Repeat for hundreds of dimmers across the microglia's DNA. Together they nudge the cell out of its calm gardener state and into the disease-associated (DAM) state. 5. Multiplied across millions of microglia, the cumulative tilt is what we measure as inherited Alzheimer's risk.
That's why mapping where IRF1 binds matters: it tells us which dimmers are wired into the gardener-to-DAM flip.
The paradigm shift
From clearing plaques… to flipping switches.
For thirty years drug companies aimed at the rubbish microglia leave behind — the amyloid plaques. Almost every trial failed. Genetics had been quietly pointing somewhere else: at the cells that make the rubbish in the first place.
The microglia era
The risk lives in the immune cells. Find the switch.
- Genetic risk lives in microglia — not neurons
- One protein, IRF1, flips them from gardener to disease-associated state
- IRF1 sits upstream of ApoE4 — the strongest known risk gene
- The switches around it are druggable targets
What's happening here?
The same microglia cell can sit in three very different states, depending on what's around it:
Gardener (homeostatic). Long, fine branches constantly probe the neighbourhood. IRF1 activity is low.
Alarmed (activated). Hit with a bacterial signal (LPS), the branches retract and the cell ramps up inflammatory signals. This is a normal and necessary response — it's how microglia fight off infection or clear up after injury. IRF1 rises, but the cell can return to the gardener state once the threat passes.
Disease-associated (DAM). In Alzheimer's, microglia face a very different challenge: a sustained, overwhelming environment of amyloid-β plaques and dying neurons. Prolonged exposure to this signal doesn't just activate the cell — it fundamentally reprograms it. The genome is rewired, the cell loses its ability to return to the gardener state, and it shifts into a chronically aggressive mode that damages rather than protects. IRF1 is high and stays high.
The PhD profiles where IRF1 binds in all three states. Which dimmers does it touch in the gardener? Which new ones lock in during the disease-associated reprogramming? Comparing the maps reveals the molecular wiring of that irreversible switch.
The master switch
One protein. Dozens of dimmers. Including the one that controls ApoE4.
IRF1 is a transcription factor — a protein whose job is to land on DNA and turn whole groups of genes up or down at once. A recent paper (ME-seq, 2026) showed it sits upstream of every known feature of the disease-associated microglia, including ApoE4, the single strongest genetic risk factor for Alzheimer's. Map where IRF1 lands — and you map the whole network.
What's happening here?
IRF1 is the hub at the centre. Each spoke is an enhancer (the enh badges) that IRF1 binds to, and each enhancer in turn controls a single named gene at the rim.
The pink genes are ones where genetic studies have already flagged Alzheimer's risk variants — ApoE4, TREM2, CD33, PLCG2, SPI1 — these are the strongest inherited risk genes the field knows about. IRF1 appears to sit upstream of all of them.
Map where IRF1 lands across the genome and you map the master regulator that points at every one of these risk genes at once.
Why it still matters
The bottleneck — and the trick to break it.
To find the buttons that calm an aggressive microglia back into a gardener, we need to see exactly where IRF1 lands on DNA. That map doesn't yet exist, because the usual way of finding it doesn't work.
Mapping where a protein touches DNA usually relies on an antibody — a sticky molecular hook that grabs the protein from a soup of cells. The catch: IRF1 isn't alone. It's one of a 9-member family (IRF1–9) whose members share a near-identical DNA-binding domain, and every off-the-shelf "anti-IRF1" antibody we've tried grabs the whole family at once. The map you'd get is a blur of every IRF, not a fingerprint of IRF1.
Our solution is a clever workaround called CETCh-seq: instead of fishing for IRF1 directly, we add a tiny molecular barcode to it — a FLAG tag. Then we use the world's best off-the-shelf antibody for that barcode. To attach the barcode without breaking the DNA, we use Prime Editing, the most precise gene-editing tool yet built. The result is a sharp, reliable IRF1 binding map — the first of its kind in human microglia.
What's happening here?
Side by side — why standard ChIP-seq fails and how CETCh-seq fixes it.
Left: The classic technique. You need a custom antibody (the Y-shape) that grips the protein you care about. For IRF1, every off-the-shelf antibody binds the whole IRF family — IRF1, IRF2, IRF8 and the rest share a near-identical DNA-binding domain, and the antibody can't tell them apart. So your "IRF1 ChIP" is really a blur of every IRF. Right: The trick. We use Prime Editing to glue a tiny generic barcode (FLAG) onto IRF1 itself — and only IRF1. We then use the world's best off-the-shelf anti-FLAG antibody, which only recognises the barcode. The antibody is no longer asked to tell siblings apart, so it doesn't have to.
The protein hasn't changed. Only how we grab hold of it has. That's why CETCh-seq can work where ChIP-seq can't.
What's happening here?
Five lab steps to go from "we can't see IRF1" to a clean genome-wide binding map:
1. Design pegRNA. A short guide molecule (PRIDICT 2.0 picks the best one) that tells Prime Editing exactly where to install the FLAG tag at the end of the IRF1 gene. 2. Prime Edit. We zap the cells (PE3 electroporation in iPSC microglia) with the editing machinery. The tag is now physically part of the IRF1 gene. 3. Verify. Sanger sequencing and NGS confirm the edit. Single edited cells are sorted by FACS into clones. 4. CETCh-seq. Same protocol as ChIP-seq, but using anti-FLAG instead of anti-IRF1 — baseline and after LPS/IFNγ stimulation. 5. Binding map. Overlay the IRF1 peaks with the 29 GWAS risk loci. Where they coincide is a candidate drug target.
Why this lab can pull it off
The Skene Lab already maps DNA-binding proteins for a living.
Reading where proteins land on DNA is what this group does every week — using a method called TIP-seq. Before we get to what this PhD adds, here's how TIP-seq actually works.
What's happening here?
The full TIP-seq pipeline in five steps:
1. Find. An antibody locks onto its target protein sitting on DNA — the protein could be any transcription factor we want to map. 2. Cut & tag. A molecular cutter (Tn5) is fused to the antibody. It clips the DNA on either side of the binding site and stamps each cut end with a synthetic tag containing a T7 promoter sequence. 3. Amplify. Here's the secret sauce. T7 RNA polymerase reads the T7 promoter and produces a fixed number of RNA copies of each tagged fragment — linear amplification. PCR doubles every cycle (exponential), so a few fragments end up wildly over-represented; T7 doesn't, so peak heights stay quantitative. 4. Sequence. Convert the RNA copies back to DNA, load them onto a sequencer, read both ends of every fragment. 5. Map. Pile the reads back onto the genome. Wherever many reads stack up — a peak — is somewhere the protein was bound. Do this for every chromosome and you have a genome-wide binding map.
Think of TIP-seq as a microscopic stamp. An antibody — a Y-shaped protein that recognises one specific target — is fused to a tiny molecular cutter called Tn5. When the antibody finds its protein sitting on DNA, the cutter clips and labels the DNA on either side of that exact spot. The bookmark Tn5 drops in includes a T7 promoter — a short DNA sequence that the next step needs.
The secret sauce Linear amplification with T7 RNA polymerase. Older protocols copy the labelled fragments by PCR — which doubles the count every cycle, so a few lucky molecules end up massively over-represented and the proportions become noise. TIP-seq skips PCR. Instead, T7 RNA polymerase latches onto every T7 promoter and makes a fixed number of RNA copies of each tagged fragment — about a hundred. A fragment present once in the cell becomes ~100 copies; a fragment present ten times becomes ~1,000. The scaling is linear, so the peak heights you measure at the end actually reflect how many times the protein was bound. PCR-based methods can't promise that. This is why TIP-seq works on tiny cell numbers, runs cleaner than ChIP-seq, and is what the Skene Lab uses every week.
The catch: TIP-seq still relies on a specific antibody to find the target in the first place. For well-studied proteins (CTCF, RNA Pol II, histone marks) that's fine. For proteins like IRF1 — whose antibodies bind every member of the IRF family at once — TIP-seq inherits the same specificity problem. That's the gap this PhD fills.
What's happening here?
Same goal — "where on the genome did this protein land?" — but two different ways of grabbing the protein:
Left (TIP-seq, lab's existing pipeline): the antibody recognises the protein directly. Works for ~80% of proteins where a paralog-specific antibody exists. Right (this PhD's CETCh-seq + Prime Editing): the antibody recognises a FLAG barcode that we install onto the protein with Prime Editing. Because we tag only IRF1, an off-the-shelf anti-FLAG antibody now pulls down IRF1 and nothing else. Works for the remaining 20% — proteins like IRF1 whose own antibodies can't tell family members apart.
The two methods complement each other. The lab keeps using TIP-seq where it's already excellent, and adds CETCh-seq for the proteins TIP-seq can't reach.
The proposal
Four aims, four years, three cell systems.
We start with a well-behaved human cell line, prove the trick works, carry it into mouse microglia, then into human microglia grown from stem cells, and finally connect what we find back to the genetics of real Alzheimer's patients.
What's happening here?
The four-year PhD plotted on a timeline. Each bar is one aim — one chunk of work with a clear deliverable. The dashed line marked today shows where in the plan we currently are.
The pink bar is the current aim; grey bars are upstream work that follows. The aims overlap a little because once a protocol works in one cell type we start porting it into the next while finishing analyses on the first.
The order is deliberate: each aim de-risks the next. We only attempt the difficult human iPSC-microglia editing once we've shown the trick works in an easy cell line, and we only attempt the disease-genetics overlay once we have real binding maps to overlay.
Aim 1 · Year 1
2026 Q4 — 2027 Q3
Prove the trick works in a friendly cell line.
We use HCT116 — a human cell line scientists already understand inside-out — to nail down how to attach a FLAG barcode to IRF1 and IRF8 using Prime Editing. We confirm the tagged protein still works, and that the barcode-antibody can pull it from chromatin (CTCF as positive control).
Aim 2 · Year 2
2027 Q4 — 2028 Q3
Carry the recipe into microglia.
We adapt the protocol for two microglial systems: BV2 mouse microglia (faster, an ethical step that reduces animal use) and human iPSC-derived microglia (KOLF2.1) — microglia grown from stem cells in a dish. Microglia are notoriously hard to edit; this is where the technical novelty lives.
Aim 3 · Years 2–3
2028 Q4 — 2029 Q3
Map IRF binding in three microglial states.
Where does IRF1 (and its cousin IRF8) actually land on DNA when a microglia is the gardener, when it's been alarmed (LPS), and when it's been turned into the disease-associated state (apoptotic neurons + amyloid-β)? This is the first ever map of those three landscapes side by side.
Aim 4 · Years 3–4
2029 Q4 — 2030 Q3
Connect the maps back to real human risk.
We overlay the IRF binding sites with Alzheimer's genetic data using two statistical tools (SLDP regression and SuSiE fine-mapping). Sites where IRF1 lands AND where DNA changes raise disease risk are the prime drug-target shortlist that comes out of this PhD.
Why this lab, why now
Five reasons this PhD belongs in the Skene Lab.
The same group whose work first put microglia at the centre of Alzheimer's now has the tools, the people, and the infrastructure to finish the job.
Built the tool that proved microglia matter.
Dr Skene's EWCE and MAGMA_Celltyping are the methods the field uses to ask "which cell type carries this disease's risk?". They are how we know, statistically, that Alzheimer's lives in microglia. Cited 1,000+ times.
Built to follow up on findings like this.
When external work — like the recent ME-seq paper (2026) flagging IRF1 upstream of the disease-associated state — points at a candidate, this is the lab positioned to test where it actually lands on DNA. Cell-type genetics, iPSC microglia, and chromatin profiling all live under one roof here.
Already maps proteins on DNA every week.
The lab runs TIP-seq and the new SNAP-Tn5 chromatin-profiling pipelines on UKDRI grants. Adding antibody-independent CETCh-seq for the proteins TIP-seq can't read is the natural next step — not a leap.
The right kit, in the right building.
UK DRI Imperial White City: a Lonza 4D-Nucleofector (the only practical way to gene-edit microglia at scale), 24× NVIDIA A100 GPUs for the genomics, BSL-2 stem-cell suites, and an in-house bioinformatics core for code review.
A network of collaborators across three continents.
The project draws on partners at the Francis Crick Institute (UK), the NIH Center for Alzheimer's and Related Dementias (USA), the Wyss-Coray Lab at Stanford (USA), and the Florey Institute (Australia) — pulling in techniques, datasets, and reagents the lab can't generate alone.
Built with patients, not just for them.
The Microglial Project Advisory Group (people living with or affected by dementia) reviews this PhD's direction twice a year. This very website is part of that two-way channel — come back, ask hard questions, push back.
Share your thoughts →
Who is running this
The team behind FLAIR-AD.
Mrinal Joshi
Doctoral researcher · UK Dementia Research Institute at Imperial
Molecular biologist (MSc Imperial, BSc King's College London). Already on the team as research assistant: built the iPSC-microglia differentiation pipeline and the nucleofection setup that this PhD is launched from.
Dr Nathan Skene
Group leader · UKRI Future Leaders Fellow · UK DRI at Imperial
Computational neurogenomicist whose tools (EWCE, MAGMA_Celltyping, MungeSumstats, Enformer Celltyping) underpin the cell-type enrichment field. Eight PhDs supervised, four already graduated.
Community
Leave a public comment.
If you want to reply publicly — reviewers, prospective collaborators — the thread is here. Top-voted posts surface first.
Forum · loading…