The Perineuronal Net, Parvalbumin Interneurons and Gamma Rhythm in Late Alzheimer's Disease
What breaks when the last brake — the perineuronal net — is digested?
The Cage and the Clock
Prepared for James Truchard and Benjamin Aaron Gustafsson by Claude Opus 5
How to read this paper
This paper follows What the Brainstem Stops Sending and The Hands That Let Go, and assumes their conclusions, which are summarised in Section 1. As in those papers, each section below condenses a longer corpus volume, and those volumes carry the full argument and their own ledgers.
The grades are the corpus's four. Established means replicated in human material by more than one group. Probable means good evidence with a gap. Inference means a conclusion drawn here, or in the corpus, from evidence gathered for another purpose. Speculative means coherent and untested.
Two cautions govern this paper in particular. First, the unit throughout the series is the connection, not the cell, because synapse loss is the structural change that tracks cognitive decline most closely — the finding that opened this literature and has never been overturned (Terry and colleagues, 1991; Selkoe, 2002). Second, this step contains the field's most active therapeutic programme and its sharpest replication dispute, and both are reported here in full.
1. Where the Series Has Got To
The first step begins in the third decade, in a handful of small nuclei that each supply one chemical messenger to the whole forebrain. Their axons fail long before their cell bodies, and their supplies fade with them. Three of those supplies hold amyloid back: serotonin brakes its production, somatostatin licenses the enzyme that disposes of it, noradrenaline enables both its cellular clearance and its drainage during sleep. Two of them — noradrenaline and acetylcholine — also hold the brain's immune cell at rest.
The second step begins around the sixth decade, on top of the first. The microglion is not switched on; it is let go, as nine restraints fail in a particular order. Once released it stops clearing amyloid, assembles the NLRP3 inflammasome, which suppresses clearance further and seeds more amyloid, and secretes the proteases that digest the perineuronal net.
Neither step has ended. That is the point of drawing the model as stacked layers rather than a relay. A person in the eighth decade carries all three at once: the brainstem still failing underneath, the microglion still released, and now a third thing beginning on top of both.
The third step is the one in which the disease becomes visible to other people. It is also the one where almost every drug ever trialled has been given.
2. The Cage: What the Perineuronal Net Is and Does
The perineuronal net is a condensed lattice of extracellular matrix wrapped around the cell body and nearest dendrites of certain neurons. It is built of hyaluronan as a backbone, lecticans — aggrecan and brevican above all — as the load-bearing proteins, link proteins that fasten them to the backbone, and tenascin-R as a cross-link. It is not a shapeless coating. It is a structure, with holes left open at synaptic contacts, and it does several jobs at once (Fawcett and colleagues, 2019).
- It excludes. The net physically restricts access to the neuronal surface, which is the office that matters for this series: it is the brake that keeps the released microglion of the second step away from the cell it would otherwise strip.
- It protects against the cell's own chemistry. Fast-spiking neurons pay a high metabolic price, and that price is paid in oxidative stress. Nets protect these cells against it, and degrading the net makes them, and the fast rhythms they generate, more vulnerable (Cabungcal and colleagues, 2013). Established.
- It protects against tau. In subcortical regions of the Alzheimer brain, neurons wearing aggrecan-based nets are protected from tau pathology (Morawski and colleagues, 2010). Established.
- It sets the cell's firing properties. The net is not inert scaffolding: its brevican component gates the parvalbumin cell's function in an activity-dependent way, organising the ion channels and the inputs the cell needs to fire fast (Favuzzi and colleagues, 2017). Established.
- It closes the critical period. Nets assemble as juvenile plasticity ends, and removing them in an adult reopens it — in part by decreasing inhibition and increasing gamma activity (Lensjø and colleagues, 2017). That result is worth holding onto, because it is the clearest demonstration that what the net does to the network is not simply protective: it is regulatory.
And it is a brake its own target attacks. Microglia facilitate the loss of perineuronal nets in the Alzheimer brain, and chronic depletion of microglia prevents the loss, despite plaques persisting (Crapser and colleagues, 2020). Established in models. In human tissue, resilience to Alzheimer's disease — intact cognition despite the pathology — is associated with distinct alterations in these nets, and in people with the disease the genes for net-degrading enzymes are raised (de Vries and colleagues, 2025). Probable, from one well-designed human series.
3. How the Cage Is Opened
The corpus calls this transition the proteolytic turn, and its argument is that three chemically distinct arms converge on the same structure at the same time.
- The protease arm. The inflammasome of the second step matures interleukin-1β, and interleukin-1β is the principal inducer of the aggrecan-degrading enzymes — the ADAMTS aggrecanases and matrix metalloproteinases, MMP-9 chief among them — that cleave the net's core protein. This is the best-characterised route from a cytokine to a protease in matrix biology, but it was characterised in cartilage. The corpus grades it established in the joint and conjecture in the brain, where the intermediate steps have not been measured (One Sensor, Two Collapses).
- The iron arm. Iron liberated from failing oligodendrocytes catalyses oxidative damage to the same structure, which is already the cell's defence against oxidative stress (The Proteolytic Turn).
- The complement arm. Complement proteins tag what is to be removed, and the tagging reaches the same neighbourhood.
Why three arms matter therapeutically. The corpus's conclusion from this is unusually practical and unusually firm: because three mechanistically distinct arms converge, interdicting any one of them produces partial protection while the other two continue. This window, on the corpus's reading, admits no single-target intervention (The Proteolytic Turn). Inference, and the strongest prediction the third step makes.
4. The Cell: Why the Parvalbumin Interneuron Is the One That Matters
Inside the cage is a particular cell, and its job explains the shape of the whole step.
What it does. The parvalbumin interneuron is a fast-spiking inhibitory cell that synapses on the cell bodies and initial segments of excitatory neurons — the places from which it can veto an output rather than merely shade it. It does not carry content. It supplies timing: it inhibits its neighbours in fast, precise rhythm, keeping cortical activity from running away with itself and holding assemblies of excitatory cells in phase.
Why it is protected for so long, and then is not. It wears the densest net in the cortex. On the model's accounting of connection loss, it is the best-protected of the five cell types counted for six decades — and then, when the lattice is digested, it becomes the worst-affected, losing some 84 per cent of its connections against 54 per cent for the excitatory cells around it. That crossing point is the model's explanation for why the last phase arrives so much faster than the first (The Shape of the Disease). The curves are the model's own and are not a measurement; the ordering, not the percentages, is the claim.
One correction the corpus has already made, and keeps. A blind re-reading of the evidence found parvalbumin involvement to be late and functional, with somatostatin interneurons failing earlier. The ordering claim survives; the identity of the terminal cell is held more loosely than it once was. The somatostatin cell, as the first paper described, largely has no net at all — which is consistent with it failing earlier and for a different reason.
The decisive causal experiment is from the other direction. Restoring Nav1.1 — the sodium channel the cell needs in order to fire fast — specifically in parvalbumin interneurons corrects network dysfunction and cognitive deficits in an amyloid model (Verret and colleagues, 2012). Established in models. That is the cleanest demonstration in the literature that the inhibitory cell's loss of capacity, rather than its death, is doing the damage.
5. The Clock: Forty Hertz, and What Stops
What the rhythm is. When parvalbumin cells fire in synchrony they impose a cortical oscillation at roughly forty cycles a second. Gamma rhythm is not an epiphenomenon of busy cortex; it is the mechanism by which populations of excitatory neurons are held in a common timeframe, so that signals arriving together are read together.
It falls early, before plaques. Driving fast-spiking parvalbumin interneurons at gamma frequency reduced amyloid in a mouse model, and the behaviourally driven rhythm was already reduced before plaques formed or cognition declined (Iaccarino and colleagues, 2016). Probable, in models.
What driving it appears to do. The programme that followed is the most ambitious therapeutic effort aimed at this step. Multisensory forty-hertz stimulation reduced pathology and improved cognition across models (Martorell and colleagues, 2019); it entrained the rhythm in deeper structures and preserved neurons and synapses in two models of neurodegeneration (Adaikkan and colleagues, 2019); and it promoted the glymphatic clearance of amyloid, the same drainage route the first paper's noradrenaline drives at night (Murdock and colleagues, 2024).
And the replication that the corpus insists on reporting. A separate laboratory, recording with multisite probes in the visual cortex, entorhinal cortex and hippocampus of two amyloid models, found that forty-hertz flickering light did not engage native gamma oscillations in those regions and did not suppress amyloid (Soula and colleagues, 2023). The corpus's position is that this does not refute the clinical programme, which is being tested on its own endpoints, but that it bears directly on the mechanism the programme claims. The dispute is live and should be presented as live.
The corpus's own warning, which is sharper than either. It is tempting to pair the two interventions of this step: preserve the lattice, then exercise the cell it protects. The Price of the Net argues that this pairing does not survive the evidence. The manipulation that preserves the net — suppressing microglia — worsens gamma disruption, increases network hyperactivity, and increases epileptiform discharge in the same animals. Matrix integrity, the intrinsic excitability of the fast-spiking cell, and the network's excitatory–inhibitory set-point are three variables, and only the first is improved by removing microglia. The two interventions act on different variables with different signs, and were joined by an inference about shared substrate rather than by any experiment that ran them together. Inference, and the most useful negative result in this part of the corpus.
6. Noise, Not Silence: Why Losing a Brake Produces Hyperactivity
The intuitive picture of a failing brain is one that goes quiet. This step does the opposite, and the evidence is unusually direct.
What is observed. Clusters of hyperactive neurons appear near amyloid plaques in a mouse model, and the measured cause is a relative decrease in synaptic inhibition (Busche and colleagues, 2008). Aberrant excitatory activity, non-convulsive seizure activity and compensatory remodelling of inhibitory circuits are documented across amyloid models (Palop and colleagues, 2007), and interneuron dysfunction is the organising theme of the field's principal review of the subject (Palop and Mucke, 2016). Established in models.
In patients. Extended monitoring found subclinical epileptiform activity in 42.4 per cent of Alzheimer patients with no history of seizures, against 10.5 per cent of age-matched controls — and the patients with it declined roughly twice as fast on a global cognitive measure, 3.9 points a year against 1.6 (Vossel and colleagues, 2016). Established, in one well-conducted prospective series.
The direction of the effective drug tells you the direction of the lesion. Of a panel of antiepileptic drugs, only levetiracetam reduced abnormal spike activity in an amyloid model, and chronic treatment reversed the circuit remodelling, the synaptic deficits and the learning failure (Sanchez and colleagues, 2012). A low dose reduced hippocampal hyperactivation and improved memory in people with amnestic mild cognitive impairment (Bakker and colleagues, 2012). In a randomised crossover trial in Alzheimer's disease the drug did not improve the primary endpoint in the group as a whole, but improved performance on spatial memory and executive tasks in the prespecified subgroup with epileptiform activity (Vossel and colleagues, 2021). Established that the effect is confined to the hyperexcitable subgroup; inference that this is the correct dosing rather than a failed drug.
The rival account, taken seriously. The corpus's adjudication volume gives the opposite reading a fair hearing: that the disease is at bottom a failure of excitatory transmission, and the seizures are what an over-inhibited network does. That is not a fringe position — seizures genuinely can arise from too much inhibition, as in absence epilepsy. But in absence epilepsy the rational and demonstrated therapy is to reduce inhibition, and in Alzheimer's disease every intervention that works runs the other way. The adjudication goes to disinhibition (Low Signal, High Noise). Inference, with the rival's strongest leg conceded.
And the consequence for the excitatory neuron. On this reading the excitatory cell fails not for want of drive but for want of coordination. It is left firing without restraint, in a network whose timing has gone, which is both excitotoxic and uninformative. That is why the model counts connections rather than cells, and why the loss of an inhibitory cell shows up as a loss of excitatory function.
7. At the Synapse: Four Receptors and One Instruction
The preceding sections are about a circuit. This one is about a single synapse, and about what amyloid does there. The corpus's reading, assembled across two volumes, is that nothing at the synapse is broken. The molecules are present, often in normal amounts. What has changed is where they are, what covalent marks they carry, and therefore which instruction they are executing. The corpus's phrase is misdirection, not damage (The Hand and the Address).
Amyloid at the glutamate synapse
Soluble amyloid oligomers, rather than the deposit, are the species that act here. They alter the trafficking of NMDA-type glutamate receptors, promoting their removal from the surface (Snyder and colleagues, 2005). Oligomers extracted directly from human Alzheimer cortex inhibit long-term potentiation, enhance its reversal, and reduce dendritic spine density in normal rodent hippocampus — the demonstration that the agent in the human brain confers the deficit (Shankar and colleagues, 2008). Established. A receptor complex for this action has been identified: oligomers bound to cellular prion protein couple to Fyn kinase through metabotropic glutamate receptor 5 (Um and colleagues, 2013). Probable.
The restraint receptor: PirB, LilrB2, and C4d
This is the receptor set Dr. Truchard asked for, and it is the most developed account of how a synapse is instructed to disassemble.
What the receptor normally does. PirB in the mouse, LilrB2 in the human, is a neuronal receptor that restrains plasticity throughout life. Deleting it leaves ocular-dominance plasticity in the visual cortex unrestricted (Syken and colleagues, 2006), and it regulates dendritic spine density cell-autonomously (Vidal and colleagues, 2016). Animals lacking it form more spines and stabilise new ones. Established.
What amyloid does to it. PirB and LilrB2 are receptors for soluble amyloid oligomers, binding at nanomolar affinity through the same two domains that bind their normal ligands; engagement enhances cofilin signalling — the same signalling seen in human Alzheimer brain — and deleting the receptor rescues both long-term potentiation and memory in an amyloid model (Kim and colleagues, 2013). Established in models; the human arm is a tissue measurement.
And the ligand that was found later. C4d, a durable cleavage fragment of complement component 4, binds LilrB2 and PirB with nanomolar affinity, colocalises with the receptor at excitatory synapses in human cortex, is elevated in Alzheimer's disease, and mediates synapse pruning (Brott and colleagues, 2025). Probable: a recent result from one laboratory. Its importance for this series is that it joins the complement arm of Section 3 to the receptor arm of this one — the same protein family that tags a synapse for a microglion also speaks directly to the neuron's own restraint receptor. The neuron, on this reading, disassembles its own spine.
The exchange factor: EphB2 and Ephexin5
A second account describes the machinery inside the spine rather than the receptor on its surface.
The normal arrangement. Ephexin5 is a RhoA exchange factor that acts as a brake on excitatory synapse formation; EphB receptor signalling degrades it, and that degradation is what permits a synapse to be built (Margolis and colleagues, 2010). Established.
What fails. Amyloid oligomers bind EphB2 and trigger its degradation; reducing EphB2 impairs NMDA currents and potentiation, and restoring it in an amyloid model reverses both the plasticity deficit and the memory impairment (Cissé and colleagues, 2011). Established in models. With EphB2 gone, the brake is no longer removed: Ephexin5 accumulates, and reducing its expression ameliorates the Alzheimer-like impairment in mice (Sell and colleagues, 2017). Probable.
So the spine is not demolished. It is forbidden to rebuild — which is why the corpus describes this limb as a ratchet rather than a weapon.
The clasp: reelin and ApoER2
The fourth receptor system is the one that normally opposes all of this at the same membrane.
What it is. Reelin, secreted in the adult cortex by inhibitory interneurons, binds the lipoprotein receptors ApoER2 and VLDLR, which phosphorylate the adaptor Disabled-1 — and mice lacking either the ligand or both receptors show hyperphosphorylation of tau (Hiesberger and colleagues, 1999). Established. The same signal enhances synaptic strength: ApoER2 sits in the postsynaptic density in a complex with the NMDA receptor, reelin markedly enhances long-term potentiation through it, and the enhancement is abolished when the receptors are deleted (Weeber and colleagues, 2002; Beffert and colleagues, 2005). Established.
Why it belongs in this paper. Reelin signalling antagonises amyloid at the synapse. It prevents the suppression of potentiation and of NMDA-receptor function caused by amyloid at concentrations comparable to the diseased brain — but at high amyloid concentrations the protection is overwhelmed and the reelin-dependent phosphorylation is completely blocked (Durakoglugil and colleagues, 2009). Established in slices. The clasp is a brake on this step in the same sense that serotonin is a brake on amyloid production in the first: real, measurable, and finite.
And the genotype term. ApoE4 selectively impairs the recycling of ApoER2, sequestering it along with glutamate receptors inside the cell, so that reelin can no longer enhance synaptic function — and the ability of reelin to protect slices against human Alzheimer brain extract is severely impaired in mice carrying the human ApoE4 isoform (Chen and colleagues, 2010). Probable. This is one of the few places in the series where the strongest common genetic risk factor acts on a named mechanism at a named membrane.
Why this is called misdirection rather than damage
Two features of the receptor account deserve stating separately, because they change what kind of lesion this is.
The receptor's own working limb has never been pointed at its assigned targets. The cytoplasmic tail of LilrB2 carries inhibitory motifs that recruit the phosphatases SHP-1 and SHP-2 — the machinery by which the receptor would actually deliver its instruction inside the cell. That was reported in a single clause of the founding paper, and it has one demonstrated neuronal consequence, on a third-party receptor kinase in the same membrane. Searching the indexed literature for the receptor or its phosphatases paired with EphB2, with Ephexin5 or with Disabled-1 returns nothing on point (The Hand and the Address). The hand is described; what it grips is not. Inference, from an absence, and stated as such.
And the field's standard measurement cannot see the lesion. If the disease-relevant change is where a molecule sits and what covalent mark it carries, rather than how much of it there is, then homogenising a piece of cortex and reporting a mean is systematically blind to it. The corpus's address claim survives in three of four regulators examined and fails in the fourth — and the failure is informative, because in the two cases where abundance was never measured, the method used could not have seen what moved. Inference, and the reason the corpus asks for single-synapse measurements rather than better blots.
What the four accounts say together, and where they collide
Read side by side, the four describe one synapse receiving several instructions at once: a restraint receptor engaged by amyloid and by complement, an exchange-factor brake that can no longer be lifted, and a clasp that would oppose both but is overwhelmed and, in ApoE4 carriers, mislocated.
The corpus does not pretend they agree. The Hand and the Address shows that where the receptor account and the exchange-factor account are joined, they predict opposite signs at the same two residues, and that the single measurement which would separate them — the phosphorylated fraction of Ephexin5 and of Tiam1, reported jointly across single synapses — has not been attempted in disease tissue although the reagent has existed for sixteen years. That is the most specific unperformed experiment in this series. Inference, and the corpus's own.
A supply question runs underneath all four of these receptors and is taken up in the fourth paper. Each is held at the membrane only as long as it is returned there faster than it is removed, which makes a synapse a balance rather than a stock; The Long Way Back argues that the return side of that balance is where this disease does its damage, and marks the extension to these four receptors as untested.
Amyloid at two concentrations, and the receptor we have left out
Everything above describes what amyloid does to a synapse. It says nothing about what the molecule is for, and the omission distorts the picture — it makes the account read as a list of poisonings rather than as a regulator pushed past its range.
The peptide is required, not merely tolerated. Healthy brains make amyloid-β continuously, in proportion to how hard their neurons are working. At low picomolar concentrations a preparation of amyloid-β42 markedly increases long-term potentiation and enhances both reference and contextual fear memory, while high nanomolar concentrations produce the familiar reduction (Puzzo and colleagues, 2008). The dose-response is biphasic.
The stronger version of the result is a subtraction rather than an addition. Removing the animal's own peptide — with an antibody against rodent amyloid-β, or by silencing murine precursor protein — impairs plasticity and memory in a healthy brain, and adding human amyloid-β42 back rescues the deficit (Puzzo and Arancio, 2013). A molecule whose removal breaks memory in a normal animal is not waste. Established in rodent preparations; no human equivalent exists and none is likely.
Where the useful effect happens, and through what. It is not a gentler version of the damage described above. At 200 picomolar, oligomeric amyloid-β acts presynaptically: it raises the frequency of miniature excitatory currents, reduces paired-pulse facilitation and increases the number of docked vesicles, which is to say it augments neurotransmitter release. Postsynaptically it lengthens the postsynaptic density and raises the plasticity proteins — phosphorylated CREB and CaMKII, and brain-derived neurotrophic factor — converting early potentiation into late potentiation and short-term memory into long-term. The effects require the peptide to be outside the cell, and they require α7 nicotinic acetylcholine receptors (Gulisano and colleagues, 2019). Established in models.
That receptor is not one of the four this section is about. It is a fifth, and it is cholinergic.
The four receptors, asked the same question. Setting the positive literature against this section's four produces an asymmetry worth stating plainly rather than smoothing.
| Receptor system | What amyloid does at disease concentrations | Is a physiological role known here? |
|---|---|---|
| NMDA-type glutamate receptors | Removed from the surface; potentiation suppressed | No. The beneficial effect is presynaptic and α7-dependent, and reaches this receptor only downstream |
| PirB / LilrB2, with C4d | Bound at nanomolar affinity; cofilin signalling enhanced | No physiological-concentration role reported |
| EphB2 with Ephexin5 | EphB2 degraded; the brake on rebuilding is never lifted | No physiological-concentration role reported |
| Reelin with ApoER2 | Protection against amyloid, overwhelmed at high concentration | The concentration axis is already here — reelin preserves potentiation at disease-comparable amyloid and is completely blocked above it (Durakoglugil and colleagues, 2009) |
| α7 nicotinic receptors | Not characterised as a damage route in this section | Yes — this is where the useful effect acts |
So the positive role is documented at one receptor, absent from the literature at two, and already implicit at a third without having been named as such. That is a statement about where the field has looked, not a claim that the other three have no physiological arm.
What this implies for the stack, and it is ours. The useful arm of amyloid runs through an acetylcholine receptor. Acetylcholine is what the nucleus basalis supplies, and its withdrawal is a Phase I event — the first paper in this series. The harmful arms run through four postsynaptic receptors that have nothing to do with it, and they are engaged at Phase III concentrations.
Read across the stack, that is a two-stroke failure of a single molecule. Phase I removes the channel through which amyloid does its useful work; Phase III raises amyloid to the concentrations at which it does harm. The beneficial arm is lost first and the harmful arms are gained second, in the order the steps occur.
Inference, and ours. No study has measured the α7 arm and the four-receptor arm in the same preparation, and the cholinergic loss and the amyloid rise have never been staged against each other in one animal. It is stated here because it is the kind of claim the series was built to make visible, and because it names an experiment: raise amyloid to picomolar levels in a preparation with cholinergic input intact and again with it removed, and see whether the benefit survives.
8. What the Third Step Adds Up To
Table 1. The third step, element by element.
| Element | What it normally does | What fails | How firm |
|---|---|---|---|
| Perineuronal net | Excludes microglia; protects against oxidative stress and tau; sets fast-firing properties | Digested by proteases, iron and complement acting together | Net loss in AD: established. The three-arm convergence: inference |
| Parvalbumin interneuron | Perisomatic inhibition; generates the forty-hertz rhythm | Loses the capacity to fire fast, before it dies | Restoring Nav1.1 rescues the network: established in models |
| Forty-hertz rhythm | Holds excitatory assemblies in a common timeframe | Falls before plaques; driving it is therapeutic in some hands and inert in others | Disputed; the replication failure is live |
| Excitatory neurons | Carry the content | Fire without restraint, in a network whose timing has gone | Hyperexcitability established in models and in patients |
| The synapse | Builds, strengthens and prunes under instruction | Receives contradictory instructions through LilrB2, EphB2–Ephexin5 and reelin–ApoER2 | Each limb probable or established; their joint reading is inference |
Three things follow.
Function goes before the cell, again. This has been true at every step of the series. The brainstem axon is removed while its cell body lives; the nucleus basalis surrenders its receptor before its transmitter; the microglion loses its badge set before it does harm; and here the parvalbumin cell loses its capacity to fire fast before it is lost, while the synapse receives wrong instructions rather than suffering breakage. That is the series' single most consistent finding, and it is the ground of whatever hope the model carries.
The step is a convergence, not a chain. Three proteolytic arms meet on one structure, and four receptor systems meet at one membrane. Single-target intervention is predicted to fail here specifically, and the record of single-target immune and anti-amyloid therapy in late disease is consistent with that.
And almost every drug ever trialled has been given here. If the model is even approximately right, the third step is the last few years of a forty-year process, and the target lies further left.
9. What This Step Implies for Treatment
Four things follow for therapy, and they are unusually specific.
No single-target drug in the turn window. Three arms converge on the net. Interdicting one leaves two running. The corpus states this as a flat prediction (The Proteolytic Turn).
Anti-excitability treatment belongs to the subgroup that has the excitability. Levetiracetam helped where epileptiform activity was documented and not otherwise, across an animal model, a mild-cognitive-impairment study and a randomised trial (Sanchez and colleagues, 2012; Bakker and colleagues, 2012; Vossel and colleagues, 2021). That is a dosing instruction, not a failure.
Do not assume the net and the rhythm are one therapy. They act on different variables with opposite signs on excitability, and the experiment that would settle it — microglial suppression against forty-hertz entrainment in one cohort, with net integrity, fast-spiking excitability, gamma power and epileptiform burden as four separate endpoints — has not been run (The Price of the Net).
And the target is capacity, not survival. The Nav1.1 result says that restoring the inhibitory cell's ability to fire corrects the network while the cell is still there (Verret and colleagues, 2012). Every step in this series has the same shape, and this is where it is most directly testable.
10. Graded Ledger
Table 2. Each claim in this paper, with its grade and the evidence it rests on.
| # | Claim | Grade | Basis |
|---|---|---|---|
| 1 | Synapse loss is the structural change that tracks cognition most closely | Established | Terry et al. 1991; Selkoe 2002 |
| 2 | The perineuronal net protects fast-spiking cells against oxidative stress | Established | Cabungcal et al. 2013 |
| 3 | Net-bearing neurons are protected against tau | Established | Morawski et al. 2010 |
| 4 | The net gates parvalbumin-cell function; removing it reopens plasticity and raises gamma | Established | Favuzzi et al. 2017; Lensjø et al. 2017; Fawcett et al. 2019 |
| 5 | Microglia digest the net; depleting them preserves it | Established (models) | Crapser et al. 2020 |
| 6 | Human resilience is associated with altered nets | Probable | de Vries et al. 2025 |
| 7 | Three arms converge on the net, so no single-target drug can work in this window | Inference | The Proteolytic Turn; One Sensor, Two Collapses |
| 8 | The parvalbumin cell is best protected for six decades, then worst affected | Inference; the curves are the model's own | The Shape of the Disease |
| 9 | Somatostatin interneurons fail earlier; the terminal cell's identity is held loosely | Probable | Blind re-reading recorded in the corpus; Gabitto et al. 2024 |
| 10 | Restoring Nav1.1 in parvalbumin cells rescues network and cognition | Established (models) | Verret et al. 2012 |
| 11 | Gamma falls before plaques, and driving it reduces pathology | Probable, and disputed | Iaccarino et al. 2016; Martorell et al. 2019; Adaikkan et al. 2019; Murdock et al. 2024 |
| 12 | Forty-hertz light does not entrain native gamma in two models | Established, as a negative result | Soula et al. 2023 |
| 13 | Preserving the net and driving the rhythm are not complementary | Inference | The Price of the Net |
| 14 | The Alzheimer cortex is hyperexcitable, from reduced inhibition | Established (models) | Busche et al. 2008; Palop et al. 2007; Palop and Mucke 2016 |
| 15 | Subclinical epileptiform activity affects ~42% of patients and doubles the rate of decline | Established, one series | Vossel et al. 2016 |
| 16 | Anti-excitability treatment helps where hyperexcitability is documented, and not otherwise | Established | Sanchez et al. 2012; Bakker et al. 2012; Vossel et al. 2021 |
| 17 | Human-brain-derived amyloid oligomers impair plasticity and spine density | Established | Shankar et al. 2008; Snyder et al. 2005; Um et al. 2013 |
| 18 | LilrB2/PirB restrains plasticity and is a nanomolar amyloid receptor | Established | Syken et al. 2006; Kim et al. 2013; Vidal et al. 2016 |
| 19 | C4d is a high-affinity LilrB2 ligand, elevated in AD, and prunes synapses | Probable; one laboratory | Brott et al. 2025 |
| 20 | Amyloid degrades EphB2; Ephexin5 then accumulates and forbids rebuilding | Established (EphB2); probable (Ephexin5) | Cissé et al. 2011; Margolis et al. 2010; Sell et al. 2017 |
| 21 | Reelin–ApoER2 signalling antagonises amyloid at the synapse and is overwhelmed at high concentrations | Established (slices) | Durakoglugil et al. 2009; Weeber et al. 2002; Beffert et al. 2005; Hiesberger et al. 1999 |
| 22 | ApoE4 impairs ApoER2 recycling and disables the reelin protection | Probable | Chen et al. 2010 |
| 23 | The receptor and exchange-factor accounts predict opposite signs at the same residues | Inference | The Hand and the Address |
11. What Would Change This Account
- A human series showing the perineuronal net intact in late Alzheimer cortex. The whole step would need rebuilding.
- A single-target intervention at one of the three proteolytic arms that fully preserves the net. The convergence argument of Section 3 would fail.
- Evidence that the parvalbumin cell dies before it loses its fast-firing capacity. The therapeutic case for acting on capacity rather than survival would go.
- A properly powered trial of forty-hertz entrainment that succeeds on a clinical endpoint while the replication failure stands. The mechanism would then need an account that does not depend on native entrainment.
- A joint measurement of the Ephexin5 and Tiam1 phosphorylation marks across single synapses showing both rising together. One of the two synaptic accounts would be eliminated.
- Levetiracetam helping a whole unselected Alzheimer population. Hyperexcitability would be general rather than a subgroup property.
12. The Three Steps Together
This series has described a disease that begins in the third decade in a handful of brainstem neurons, loosens a braking system on the brain's immune cell in the sixth, and digests a protective lattice in the eighth — with no step ending when the next begins.
Three conclusions hold across all three papers.
Function fails before cells die, everywhere. At every step the measurable loss is of capacity, connection or instruction, and not of the cell. A neuron that has lost its inputs is in a different condition from a dead one, and the model's claim is that much of what is lost in the late step is the former.
The brakes are the story. Amyloid production is braked by serotonin, its disposal licensed by somatostatin, its clearance enabled by noradrenaline; the microglion is restrained by nine separate mechanisms; the parvalbumin cell is caged; the synapse is held by a reelin clasp. The disease is better described as the sequential failure of restraints than as the accumulation of a toxin.
And the target lies further left than the trials have gone. That is the practical conclusion the three papers share, and the one most worth attacking.
The measurement that would settle the most. Whether a silenced neuron can be recovered is the open question on which the hopeful reading of this whole model rests. It has not been answered, and nothing in these three papers answers it.
Corpus papers drawn on
The Shape of the Disease. The Proteolytic Turn. One Sensor, Two Collapses and The Net and the Speck. The Price of the Net. Low Signal, High Noise: Excitation and Inhibition in Alzheimer's Disease. The Hand and the Address: LilrB2, C4d, Ephexin5 and Reelin in Alzheimer's Disease — Misdirection, Not Damage. No Switch but the Endosome. The Architect's Handshake: The Reelin–ApoER2–Heparan Sulfate Interface. The Matrix of Collapse and The Perineuronal Turn. Convergent Synaptic Collapse. The Unnetted Governor. Papers one and two of this series, What the Brainstem Stops Sending and The Hands That Let Go. All are available in full at AdultCognitiveDisease.com.
References
Adaikkan C, Middleton SJ, Marco A, Pao PC, Mathys H, Kim DN, et al.. Gamma entrainment binds Higher-Order brain regions and offers neuroprotection. Neuron. 2019;102(5):929-943.e8. PMID 31076275. doi:DOI 10.1016/j.neuron.2019.04.011
Bakker A, Krauss GL, Albert MS, Speck CL, Jones LR, Stark CE, et al.. Reduction of hippocampal hyperactivity improves cognition in amnestic mild cognitive impairment. Neuron. 2012;74(3):467-74. PMID 22578498. doi:DOI 10.1016/j.neuron.2012.03.023
Beffert U, Weeber EJ, Durudas A, Qiu S, Masiulis I, Sweatt JD, et al. Modulation of synaptic plasticity and memory by Reelin involves differential splicing of the lipoprotein receptor Apoer2. Neuron. 2005;47(4):567-79. PMID 16102539. doi:DOI 10.1016/j.neuron.2005.07.007
Brott BK, Raissi AJ, Micheva KD, Vielmetter J, Mendes MS, Baccus CJ, et al.. C4d, a high-affinity LilrB2 ligand, is elevated in Alzheimer's disease and mediates synapse pruning. Proceedings of the National Academy of Sciences USA. 2025;122(38):e2519253122. PMID 40966293. doi:DOI 10.1073/pnas.2519253122
Busche MA, Eichhoff G, Adelsberger H, Abramowski D, Wiederhold KH, Haass C, et al.. Clusters of hyperactive neurons near amyloid plaques in a mouse model of Alzheimer's disease. Science (New York, N.Y.). 2008;321(5896):1686-9. PMID 18802001. doi:DOI 10.1126/science.1162844
Cabungcal JH, Steullet P, Morishita H, Kraftsik R, Cuenod M, Hensch TK, et al. Perineuronal nets protect fast-spiking interneurons against oxidative stress. Proceedings of the National Academy of Sciences USA. 2013;110(22):9130-5. PMID 23671099. doi:DOI 10.1073/pnas.1300454110
Chen Y, Durakoglugil MS, Xian X, Herz J. ApoE4 reduces glutamate receptor function and synaptic plasticity by selectively impairing ApoE receptor recycling. Proceedings of the National Academy of Sciences USA. 2010;107(26):12011-6. PMID 20547867. doi:DOI 10.1073/pnas.0914984107
Cissé M, Halabisky B, Harris J, Devidze N, Dubal DB, Sun B, et al. Reversing EphB2 depletion rescues cognitive functions in Alzheimer model. Nature. 2011;469(7328):47-52. PMID 21113149. doi:DOI 10.1038/nature09635
Crapser JD, Spangenberg EE, Barahona RA, Arreola MA, Hohsfield LA, Green KN. Microglia facilitate loss of perineuronal nets in the Alzheimer's disease brain. EBioMedicine. 2020;58:102919. PMID 32745992. doi:DOI 10.1016/j.ebiom.2020.102919
de Vries LE, Bahnerth A, Swaab DF, Verhaagen J, Carulli D. Resilience to Alzheimer's disease associates with alterations in perineuronal nets. Alzheimer's & Dementia. 2025;21(2):e14504. PMID 39737731. doi:DOI 10.1002/alz.14504
Durakoglugil MS, Chen Y, White CL, Kavalali ET, Herz J. Reelin signaling antagonizes β-amyloid at the synapse. Proceedings of the National Academy of Sciences USA. 2009;106(37):15938-43. PMID 19805234. doi:DOI 10.1073/pnas.0908176106
Favuzzi E, Marques-Smith A, Deogracias R, Winterflood CM, Sánchez-Aguilera A, Mantoan L, et al.. Activity-Dependent gating of parvalbumin Interneuron function by the perineuronal net protein brevican. Neuron. 2017;95(3):639-655.e10. PMID 28712654. doi:DOI 10.1016/j.neuron.2017.06.028
Fawcett JW, Oohashi T, Pizzorusso T. The roles of perineuronal nets and the perinodal extracellular matrix in neuronal function. Nature Reviews Neuroscience. 2019;20(8):451-465. PMID 31263252. doi:DOI 10.1038/s41583-019-0196-3
Gabitto MI, Travaglini KJ, Rachleff VM, Kaplan ES, Long B, Ariza J, et al. Integrated multimodal cell atlas of Alzheimer's disease. Nature Neuroscience. 2024;27(12):2366-2383. PMID 39402379. doi:DOI 10.1038/s41593-024-01774-5
Gulisano W, Melone M, Ripoli C, Tropea MR, Li Puma DD, Giunta S, Cocco S, Marcotulli D, Origlia N, Palmeri A, Arancio O, Conti F, Grassi C, Puzzo D. Neuromodulatory action of picomolar extracellular Aβ42 oligomers on presynaptic and postsynaptic mechanisms underlying synaptic function and memory. Journal of Neuroscience. 2019;39(30):5986-6000. PMID 31127002.
Hiesberger T, Trommsdorff M, Howell BW, Goffinet A, Mumby MC, Cooper JA, et al. Direct binding of Reelin to VLDL receptor and ApoE receptor 2 induces tyrosine phosphorylation of disabled-1 and modulates tau phosphorylation. Neuron. 1999;24(2):481-9. PMID 10571241. doi:DOI 10.1016/s0896-6273(00)80861-2
Iaccarino HF, Singer AC, Martorell AJ, Rudenko A, Gao F, Gillingham TZ, et al.. Gamma frequency entrainment attenuates amyloid load and modifies microglia. Nature. 2016;540(7632):230-235. PMID 27929004. doi:DOI 10.1038/nature20587
Kim T, Vidal GS, Djurisic M, William CM, Birnbaum ME, Garcia KC, et al.. Human LilrB2 is a β-amyloid receptor and its murine homolog PirB regulates synaptic plasticity in an Alzheimer's model. Science (New York, N.Y.). 2013;341(6152):1399-404. PMID 24052308. doi:DOI 10.1126/science.1242077
Lensjø KK, Lepperød ME, Dick G, Hafting T, Fyhn M. Removal of perineuronal nets unlocks juvenile plasticity through network mechanisms of decreased inhibition and increased gamma activity. The Journal of Neuroscience. 2017;37(5):1269-1283. PMID 28039374. doi:DOI 10.1523/JNEUROSCI.2504-16.2016
Margolis SS, Salogiannis J, Lipton DM, Mandel-Brehm C, Wills ZP, Mardinly AR, et al.. EphB-mediated degradation of the RhoA GEF Ephexin5 relieves a developmental brake on excitatory synapse formation. Cell. 2010;143(3):442-55. PMID 21029865. doi:DOI 10.1016/j.cell.2010.09.038
Martorell AJ, Paulson AL, Suk HJ, Abdurrob F, Drummond GT, Guan W, et al.. Multi-sensory gamma stimulation ameliorates Alzheimer's-Associated pathology and Improves cognition. Cell. 2019;177(2):256-271.e22. PMID 30879788. doi:DOI 10.1016/j.cell.2019.02.014
Morawski M, Brückner G, Jäger C, Seeger G, Arendt T. Neurons associated with aggrecan-based perineuronal nets are protected against tau pathology in subcortical regions in Alzheimer's disease. Neuroscience. 2010;169(3):1347-63. PMID 20497908. doi:DOI 10.1016/j.neuroscience.2010.05.022
Murdock MH, Yang CY, Sun N, Pao PC, Blanco-Duque C, Kahn MC, et al.. Multisensory gamma stimulation promotes glymphatic clearance of amyloid. Nature. 2024;627(8002):149-156. PMID 38418876. doi:DOI 10.1038/s41586-024-07132-6
Palop JJ, Chin J, Roberson ED, Wang J, Thwin MT, Bien-Ly N, et al.. Aberrant excitatory neuronal activity and compensatory remodeling of inhibitory hippocampal circuits in mouse models of Alzheimer's disease. Neuron. 2007;55(5):697-711. PMID 17785178. doi:DOI 10.1016/j.neuron.2007.07.025
Palop JJ, Mucke L. Network abnormalities and interneuron dysfunction in Alzheimer disease. Nature Reviews Neuroscience. 2016;17(12):777-792. PMID 27829687. doi:DOI 10.1038/nrn.2016.141
Sanchez PE, Zhu L, Verret L, Vossel KA, Orr AG, Cirrito JR, et al. Levetiracetam suppresses neuronal network dysfunction and reverses synaptic and cognitive deficits in an Alzheimer's disease model. Proceedings of the National Academy of Sciences USA. 2012;109(42):E2895-903. PMID 22869752. doi:DOI 10.1073/pnas.1121081109
Selkoe DJ. Alzheimer's disease is a synaptic failure. Science (New York, N.Y.). 2002;298(5594):789-91. PMID 12399581. doi:DOI 10.1126/science.1074069
Sell GL, Schaffer TB, Margolis SS. Reducing expression of synapse-restricting protein Ephexin5 ameliorates Alzheimer's-like impairment in mice. The Journal of Clinical Investigation. 2017;127(5):1646-1650. PMID 28346227. doi:DOI 10.1172/JCI85504
Shankar GM, Li S, Mehta TH, Garcia-Munoz A, Shepardson NE, Smith I, et al.. Amyloid-beta protein dimers isolated directly from Alzheimer's brains impair synaptic plasticity and memory. Nature Medicine. 2008;14(8):837-42. PMID 18568035. doi:DOI 10.1038/nm1782
Puzzo D, Privitera L, Leznik E, Fà M, Staniszewski A, Palmeri A, Arancio O. Picomolar amyloid-beta positively modulates synaptic plasticity and memory in hippocampus. Journal of Neuroscience. 2008;28(53):14537-14545. PMID 19118188.
Puzzo D, Arancio O. Amyloid-β peptide: Dr. Jekyll or Mr. Hyde? Journal of Alzheimer's Disease. 2013;33(Suppl 1):S111-S120. PMID 22735675.
Snyder EM, Nong Y, Almeida CG, Paul S, Moran T, Choi EY, et al.. Regulation of NMDA receptor trafficking by amyloid-beta. Nature Neuroscience. 2005;8(8):1051-8. PMID 16025111. doi:DOI 10.1038/nn1503
Soula M, Martín-Ávila A, Zhang Y, Dhingra A, Nitzan N, Sadowski MJ, et al.. Forty-hertz light stimulation does not entrain native gamma oscillations in Alzheimer's disease model mice. Nature Neuroscience. 2023;26(4):570-578. PMID 36879142. doi:DOI 10.1038/s41593-023-01270-2
Syken J, Grandpre T, Kanold PO, Shatz CJ. PirB restricts ocular-dominance plasticity in visual cortex. Science (New York, N.Y.). 2006;313(5794):1795-800. PMID 16917027. doi:DOI 10.1126/science.1128232
Terry RD, Masliah E, Salmon DP, Butters N, DeTeresa R, Hill R, et al.. Physical basis of cognitive alterations in Alzheimer's disease: synapse loss is the major correlate of cognitive impairment. Annals of Neurology. 1991;30(4):572-80. PMID 1789684. doi:DOI 10.1002/ana.410300410
Um JW, Kaufman AC, Kostylev M, Heiss JK, Stagi M, Takahashi H, et al.. Metabotropic glutamate receptor 5 is a coreceptor for Alzheimer aβ oligomer bound to cellular prion protein. Neuron. 2013;79(5):887-902. PMID 24012003. doi:DOI 10.1016/j.neuron.2013.06.036
Verret L, Mann EO, Hang GB, Barth AM, Cobos I, Ho K, et al.. Inhibitory interneuron deficit links altered network activity and cognitive dysfunction in Alzheimer model. Cell. 2012;149(3):708-21. PMID 22541439. doi:DOI 10.1016/j.cell.2012.02.046
Vidal GS, Djurisic M, Brown K, Sapp RW, Shatz CJ. Cell-Autonomous regulation of dendritic spine density by PirB. eNeuro. 2016;3(5). PMID 27752542. doi:DOI 10.1523/ENEURO.0089-16.2016
Vossel KA, Ranasinghe KG, Beagle AJ, Mizuiri D, Honma SM, Dowling AF, et al. Incidence and impact of subclinical epileptiform activity in Alzheimer's disease. Annals of Neurology. 2016;80(6):858-870. PMID 27696483. doi:DOI 10.1002/ana.24794
Vossel K, Ranasinghe KG, Beagle AJ, La A, Ah Pook K, Castro M, et al. Effect of levetiracetam on cognition in patients with Alzheimer disease with and without epileptiform activity: a randomized clinical trial. JAMA Neurology. 2021;78(11):1345-1354. PMID 34570177. doi:DOI 10.1001/jamaneurol.2021.3310
Weeber EJ, Beffert U, Jones C, Christian JM, Forster E, Sweatt JD, et al. Reelin and ApoE receptors cooperate to enhance hippocampal synaptic plasticity and learning. The Journal of Biological Chemistry. 2002;277(42):39944-52. PMID 12167620. doi:DOI 10.1074/jbc.M205147200
Genes named on this page: LilrB2, PirB; RELN, reelin; PVALB (parvalbumin), parvalbumin; ARHGEF15 (Ephexin5), Ephexin5, Ephexin-5; EphB2; LRP8 (ApoER2), ApoER2; ApoE, apoe4; C4 (complement), C4d; SST (somatostatin), somatostatin; ACAN (aggrecan), aggrecan; DAB1, Disabled-1; BCAN (brevican), brevican; RhoA; PRNP, Prion Protein; NLRP3; MMP9, MMP-9; Fyn; VLDLR; TNR (tenascin-R), tenascin-R, tenascinR; CAMK2A (CaMKII), CaMKII.