On 26 August 2026, a team led by Won-Suk Chung at the Institute for Basic Science (IBS) Center for Vascular Research and KAIST reported in Nature that a receptor normally found on inhibitory neurons — ERBB4 — turns on in the wrong cells early in Alzheimer’s-like disease and helps drive plaques, circuit noise, gliosis, and memory loss together. In 5×FAD mice, selective CRISPR deletion of Erbb4 in hippocampal excitatory neurons cut amyloid plaque area and number by about half, according to Korea’s Ministry of Science and ICT / DongA Science summary of the work, and rescued several cognition tests. In humans, directed mediation analysis on transcriptomes from 446 ROSMAP Alzheimer’s brains linked excitatory-neuron ERBB4 to plaque severity, tau, and cognitive decline. This is not a cure, not an approved drug, and not proof that deleting ERBB4 would be safe in patients.

The paper’s title is blunt: “Aberrant excitatory neuronal ERBB4 promotes Alzheimer’s disease pathology.” First author Se Young Lee and colleagues argue the misplaced receptor is an early driver, not a late bystander.

Why it matters

Alzheimer’s still has no disease-modifying “break the cascade” drug that most patients can count on. Anti-amyloid antibodies can clear plaques; cognitive gains remain modest. The open question is what sits downstream of amyloid and keeps synapses, glia, and circuits failing even after plaques fall.

Chung’s group started with a different puzzle. Astrocytes and microglia in APP/PS1 and 5×FAD mice increasingly engulfed excitatory synapses while removing fewer inhibitory ones — not a simple “glia eat everything” story. When the team cranked neuronal activity up or down with chemogenetics, glial engulfment rose and fell with it. Something inside the neurons was coaching the glia.

Single-nucleus RNA sequencing then found a new early population the authors call early-responsive excitatory neurons (ERENs). Those cells turn on ectopic Erbb4. In the healthy brain, ERBB4 is mainly an inhibitory-neuron receptor. In the disease models, it appears in a subset of the cells that drive circuit activity. Amyloid-β oligomers injected into CA1 induced excitatory-neuron ERBB4 within two days, and EREN abundance did not require TREM2-dependent disease-associated microglia — so Aβ can push the switch at least partly without that microglial path.

For a stranger watching dementia research, the stake is simple: a named receptor that turns several hallmarks on and off in mice is a concrete target people can follow into toxicity studies and, if it survives them, early trials.

Key numbers

MetricValueSource
Publication26 Aug 2026, NatureDOI 10.1038/s41586-026-10964-z
InstitutionsIBS Center for Vascular Research + KAIST Biological SciencesNature / MSIT
Mouse model (main CRISPR)5×FAD, CA1 excitatory neuronsNature
Aβ plaque area & number after Erbb4 deletion~halfDongA / Ministry of Science and ICT
Human transcriptomic cohort446 ROSMAP AD participantsNature
Downstream path required for many phenotypesmTOR / mTORC1 (Rptor deletion)Nature
Timing of benefitDeletion at 3 mo helped at 7 mo; at 8 mo helped at 10 moNature Extended Data
Illimis collaboration goal (quote)Toxicity 1–2 years; clinical goal “within five years” if efficacy holdsDongA (Jung)

Do not read “~half” as a human outcome. It is a mouse plaque-burden fold change reported in Korean official coverage of the Nature figures. The paper itself describes plaque number and area as “strongly reduced” after excitatory-neuron Erbb4 deletion (Fig. 3n,o).

How they showed it

Loss of function. An AAV carrying CaMKIIα-driven SaCas9 plus an Erbb4 guide into CA1 of 4-month-old 5×FAD mice lowered ERBB4 in pyramidal neurons without stripping it from parvalbumin interneurons. That single move:

  • restored glial synapse-engulfment patterns toward wild-type,
  • normalized excitatory hyperactivity and somatostatin hypoactivity (FOS counts),
  • reduced S100β, GFAP, and IBA1 reactive-gliosis areas and AXL+ disease-associated microglia,
  • cut plaque burden,
  • and rescued spontaneous alternation, novel-object location/recognition, and Barnes-maze performance.

The same deletion in wild-type CA1 did not scramble synapses or gliosis — a useful safety signal in mice, not a human green light.

Gain of function. Overexpressing Erbb4 in only 5–15% of wild-type CA1 pyramidal neurons — matching the diseased fraction — was enough to recreate hyperactivity, synaptic imbalance, gliosis, and cognitive deficits without amyloid plaques. A kinase-dead K751M mutant did not. Cortical overexpression produced similar activity and gliosis shifts.

Mechanism. Deleting Rptor (mTORC1) in 5×FAD excitatory neurons phenocopied much of the Erbb4 rescue. Co-deleting Rptor while overexpressing Erbb4 in wild-type mice blocked the induced pathology. The authors place an ERBB4–mTOR axis upstream of excitability programs and the glial cascade. Acute chemogenetic silencing of excitatory neurons fixed activity and synapse phagocytosis but not gliosis or plaques — so ERBB4 is doing more than short-term circuit noise.

Humans. FISH on AD tissue showed elevated ERBB4 in SLC17A7+ excitatory neurons. In 446 ROSMAP dorsolateral-prefrontal transcriptomes, ERBB4-high excitatory clusters correlated with worse CERAD plaque scores and lower MMSE. Structural equation modeling favored a path in which excitatory ERBB4 sits in a vicious cycle with amyloid, then tau and cognition (comparative fit index 1.0; χ² P = 0.9863). Swapping in ERBB2 broke the fit. That is statistical support for a cascade model, not causal proof in living patients.

What this is not

  • Not a claim that amyloid is “debunked.” The authors say Aβ can induce ERBB4, then ERBB4 amplifies plaques — a feedback loop, not a replacement for amyloid biology.
  • Not an off-the-shelf inhibitor. First author Lee warned that a systemic drug might hit inhibitory neurons too; those cells need ERBB4. Chung’s biotech, Illimis Therapeutics, is collaborating on excitatory-selective approaches; patents are filed. Toxicity studies are planned for one to two years, with a clinical goal inside five years if efficacy holds — aspirational, not scheduled.
  • Not a guarantee for tauopathies. Preliminary public snRNA-seq hints of EREN-like ERBB4-high states in progressive supranuclear palsy and a correlation in Huntington’s disease are early and secondary to the AD mouse work.

What to watch

  1. Excitatory-selective ERBB4 inhibitors from the Illimis–Chung collaboration — toxicity readouts first.
  2. Whether other labs reproduce plaque and cognition rescue outside 5×FAD / APP/PS1.
  3. Human validation beyond ROSMAP mediation: protein-level ERBB4 in excitatory neurons across more banks, and whether CSF or imaging proxies can track the state.
  4. How any ERBB4 path would combine with anti-amyloid antibodies rather than replace them — Chung’s long-term framing in the DongA interview.

The molecular switch has a name, a paper in Nature, a half-plaque mouse result, and a 446-brain human correlation. The drug does not exist yet. That gap is the story’s fence — and the next measurable step.

Sources