THE MATRIX OF COLLAPSE
The Perineuronal Net as the Structural Convergence Node of the Eight Frameworks — and a Unified Timeline of Synaptic Disintegration in Alzheimer's Disease
Abstract
The companion analysis Convergent Synaptic Collapse assembled eight independently developed frameworks — Ramsden's lipid peroxidation and ApoER2–Dab1 disruption, Small's retromer–endosomal traffic jam, Gouras's intraneuronal inside-out paradigm, Moosmann's chronic excitatory insufficiency, Rappoport's lipid-raft allostatic collapse, Margolis's synaptic restriction and proteostatic failure, Huang's competitive synaptic plasticity, and the Shatz–Brott C4d–LilrB2 axis — and showed that they converge on a small number of shared nodes: an endosomal nexus, a cytoskeletal collapse node, a compensatory paradigm, a neuroimmune interface, the ApoE4 hub, and a transcriptional layer. That synthesis described, with great precision, how the synapse is destroyed. It did not name the structure on which the destruction is staged. This paper supplies it.
We propose that the perineuronal net — the condensed extracellular-matrix lattice of aggrecan, brevican, hyaluronan and tenascin-R that ensheathes the parvalbumin-positive fast-spiking interneuron and a minority of excitatory neurons — is the missing structural convergence node of the entire synthesis, and that its degradation is the single physical event that converts the diffuse, intracellular molecular pathology of the eight frameworks into the focal, perisomatic, inhibitory collapse that is experienced as dementia. The net is not one casualty among many; it is the surface on which the convergence happens. It is, at once, the protective shield whose loss exposes the most metabolically extravagant neuron in the cortex; the protease and complement target on which the neuroimmune machinery of the Shatz–Brott and Margolis frameworks deposits C4d and discharges MMP-9 and the ADAMTS proteases; the iron sink on which the lipid-peroxidative chemistry of the Ramsden framework runs its Fenton reactions; and the scaffold surrounding the perisomatic synapses at which the cofilin-actin-rod cascade — the common cytoskeletal endpoint of Ramsden, Margolis, and Shatz–Brott — is physically executed. Strip the net, and the four convergent arms of the synthesis arrive together at one cell.
Reading the disease through the net resolves two of the synthesis's open problems. First, it explains the geometry of inhibitory failure that the single-cell atlases report — the somatostatin interneuron, which carries no net, falls early, while the netted parvalbumin interneuron falls late — as a single variable: vulnerability tracks the absence or removal of the protective matrix. Second, it supplies the timeline. Because the net is stripped before the cell it protects can fail, and because the stripped cell persists for years in a silenced, reawakenable state, the disease unfolds as a stereotyped half-century sequence with a definable latent interval — the gap between the loss of the net and the loss of the neuron — that is the disease's principal and most neglected window of rescue. We close by assembling, from the eight frameworks and the net together, a unified timeline of every load-bearing event from the third decade of life to terminal dementia, and by stating the falsifiable prediction on which the whole construction rests: that in the human disease the aggrecan core of the parvalbumin net is degraded, stage-specifically, before the cells and synapses it protects are lost.
I. The Gap in the Convergent Collapse
The convergent synthesis is, by design, a map of mechanisms rather than of places. It traces the disease from oxidative lipid modification and endosomal trafficking arrest, through the reactivation of developmental pruning programmes and the depletion of protective amyloid monomer, to the complement-driven elimination of synapses, and it shows that these arms meet at shared molecular nodes. The achievement of the synthesis is to demonstrate that no single framework is sufficient and that the disease is a network failure. But a network failure still happens somewhere, on some physical substrate, and on this question the synthesis is — understandably, given its molecular focus — silent. It names the kinases (GSK3β, LIMK1, ROCK), the effectors (cofilin, the actin cytoskeleton), the immune ligands (C4d, LilrB2), and the cell types (the pyramidal neuron, the microglion), but it does not name the structure that organises the zone where the perisomatic synapse is made, protected, and finally destroyed.
This omission has a consequence. Without a structural node, the synthesis's most important convergence — the cytoskeletal collapse node, at which Ramsden's GSK3β→cofilin arm, Margolis's Ephexin5–RhoA–ROCK arm, and Shatz–Brott's C4d–LilrB2→cofilin-actin arm all arrive — appears overdetermined but unlocated. Why do three independent chemistries converge on the same synapses, in the same cells, in the same anatomical order? Why is the parvalbumin interneuron, which the Moosmann framework places at the centre of the excitatory–inhibitory collapse and of memantine's mechanism, both spared for decades and then catastrophically lost? And why does the loss of inhibition arrive late, after the long preclinical accrual of intracellular pathology the other frameworks describe? A purely molecular synthesis cannot answer these questions, because the answer is not a molecule. It is a structure, and the structure is the perineuronal net.
II. The Net as the Seventh Node
What the net is, and whom it protects
The perineuronal net is a reticulated lattice of extracellular matrix condensed around the soma and proximal dendrites of specific neurons: a backbone of hyaluronan, decorated with the lectican chondroitin-sulfate proteoglycans — aggrecan (the obligatory, highest-density component), brevican, neurocan and versican — cross-linked by tenascin-R and stabilised by the hyaluronan-and-proteoglycan link proteins (Fawcett, Oohashi & Pizzorusso, 2019). It is, overwhelmingly, the property of the parvalbumin-positive fast-spiking interneuron (Härtig, Brauer & Brückner, 1992; Celio, 1986), and it is borne in addition by a minority of excitatory neurons in specific layers and regions. The cell it most characteristically wraps is the one the Moosmann framework identifies as the linchpin of excitatory–inhibitory balance and of memantine's paradoxical efficacy: a neuron that fires above two hundred hertz, carries the highest oxidative load of any cortical neuron, and bears calcium-permeable receptors that leave it exposed to the ion its own activity admits (Hu, Gan & Jonas, 2014).
Why the net is a node, not a casualty
The reason the net belongs at the centre of the synthesis, rather than at its periphery, is that it is the one structure that every convergent arm must pass through to reach its target. The net buffers the cation fluxes that sustain fast firing; it functions as an antioxidant shield, such that its experimental removal renders the enclosed neuron measurably more vulnerable to oxidative insult (Cabungcal et al., 2013; Suttkus et al., 2014); it stabilises the perisomatic synapses that terminate within it; it restricts the lateral mobility of glutamate receptors and so disciplines plasticity into stability (Frischknecht et al., 2009); and it restricts the internalisation of pathological species, so that net-bearing neurons resist tangle formation (Morawski et al., 2010). Each of these offices is the obverse of a convergent arm of the synthesis: the antioxidant office is the obverse of Ramsden's lipid peroxidation; the receptor-disciplining office is the obverse of the cytoskeletal collapse node; the exclusion office is the obverse of the inside-out tauopathy. The net is the node because it is the structure whose loss is the convergence — the point at which four molecular arms, each described by a different framework, become one perisomatic catastrophe.
III. How Each Framework Touches the Net
The claim that the net is the structural convergence node is testable against the eight frameworks one at a time: each, on inspection, makes molecular contact with the matrix, and the contacts are not incidental but load-bearing.
| Framework | Point of contact with the net | Consequence at the matrix |
|---|---|---|
| Ramsden (lipid peroxidation, iron) | the net is a dense polyanion and an avid iron sink; 4-HNE and Fenton chemistry | oxidative fragmentation of aggrecan; the net becomes the substrate of the iron bridge |
| Small (retromer / trafficking) | retromer-dependent secretion and surface delivery of matrix proteases and their regulators | dysregulated MMP/ADAMTS output toward the net; lysosomal handling of engulfed matrix |
| Gouras (intraneuronal Aβ) | aggrecan and net fragments co-localise with intraneuronal and plaque Aβ | the net's loss removes the exclusion barrier; the netted neuron becomes tau-permissive |
| Moosmann (excitatory insufficiency) | the net protects the PV interneuron whose tonic NMDA drive and gamma output set E/I balance | net loss disinhibits the cortex; gamma collapses; memantine's PV target is unshielded |
| Rappoport (lipid raft / allostatic) | net integrity couples to membrane and astrocytic ion handling; allostatic oxidative load | chronic stress shifts the net from maintenance to degradation |
| Margolis (Ephexin5–RhoA–ROCK; MMP) | RhoA/ROCK and the matrix proteases act in the perisomatic zone the net occupies | actin-myosin contraction beneath a digested net; MMP-driven matrix loss |
| Huang (monomer depletion; MMP-9) | activated microglia recruit MMP-9 hyperactivation, the canonical aggrecan-degrading protease | the proteolytic switch that strips the net is the same that executes synaptic loss |
| Shatz–Brott (C4d–LilrB2; complement) | C1q and C4d deposit on the aggrecan/brevican net as on the synapse | the net is opsonised for microglial stripping; cofilin-actin rods form beneath it |
Read down the column of contacts, a single mechanism assembles itself. The lipid-peroxidative arm loads the net with iron and fragments it by Fenton chemistry (Ramsden); the trafficking arm dysregulates the secretion of the proteases that will digest it (Small); the immune arm opsonises it with complement and recruits MMP-9 to cleave it (Shatz–Brott, Huang); and the cytoskeletal arm contracts the actin beneath it once the protective lattice is gone (Margolis, Ramsden). The frameworks do not merely co-occur; they are four chemistries pointed at one structure.
IV. The Cytoskeletal Collapse, Re-centred on the Net
The synthesis's cytoskeletal collapse node is the place where the matrix hypothesis does its most useful work, because it explains the node's most puzzling feature: its overdetermination. The synthesis observes, correctly, that the dendritic spine is attacked from several directions at once — by GSK3β-driven, LIMK1-mediated cofilin dysregulation (Ramsden); by Ephexin5–RhoA–ROCK-mediated actin-myosin contraction (Margolis); and by C4d–LilrB2-triggered cofilin-actin-rod formation (Shatz–Brott) — and that this convergence makes the collapse robust and difficult to reverse. What it cannot say is why these three chemistries find the same synapses.
The net supplies the reason. The perisomatic synapses on the parvalbumin interneuron, and on the netted excitatory neuron, are the synapses embedded within the matrix lattice. The net is the structure that holds them in place, buffers their local ionic environment, and restricts the receptor mobility on which their stability depends. While the net is intact, the cofilin-actin machinery beneath it is held in a disciplined, stable configuration; the matrix is, in effect, a brake on the very cytoskeletal dynamics the three arms seek to unleash. When the net is digested — by the iron, protease, and complement chemistries of Section III — that brake is released, and the three convergent arms find, suddenly exposed, exactly the synapses they were always able to attack but could not previously reach. The cofilin-actin rods of the Shatz–Brott and Bamburg literatures form, on this reading, beneath a stripped net, in the perisomatic zone the matrix used to protect. The cytoskeletal collapse node is therefore not merely overdetermined; it is gated by the net. The net's loss is the permission that lets the convergence occur.
V. The Two-Step Lesion and the Latent Interval
The matrix hypothesis adds to the synthesis a feature its molecular nodes cannot supply: a separation in time between the lesion and the death. The net is degraded before the cell it protects is lost — the impairment of the matrix precedes, rather than accompanies, the depletion of the parvalbumin population (Crapser et al., 2020). Between these two events lies a protracted state in which the parvalbumin interneuron, stripped of its shield, does not die but withdraws: it downregulates parvalbumin and its sodium channel Nav1.1, loses the synaptic organiser NPTX2, and falls electrically quiet (Verret et al., 2012; and the human single-cell atlases, which place parvalbumin loss late — Gabitto et al., 2024). This is not the beginning of the cell's death but a strategy of survival in the unshielded condition: a fast-spiking neuron that has lost the net buffering its firing is one whose continued activity would now poison it, and its quiescence reduces the very metabolic and oxidative load it can no longer afford.
This latent interval — the gap between the loss of the net and the loss of the neuron — reconciles the synthesis with its own timeline. The convergent frameworks describe a long preclinical accrual of intracellular pathology followed by a comparatively rapid clinical collapse; the matrix hypothesis locates the hinge between them at the stripping of the net. Before the net is stripped, the disease is the slow, intracellular, compensable process the endosomal and compensatory nodes describe. After the net is stripped, the parvalbumin cell enters its latent state, the cortex begins to disinhibit, and the rapid, self-sustaining phase begins. The clinical transition from preclinical pathology to dementia is, on this account, the crossing of the net — and the latent interval that follows is the disease's window of rescue, because a silenced cell can be reawakened and a dead one cannot.
VI. The Excitatory Cascade and the Closing of the Loop
The synthesis's "self-sustaining loop" — the feed-forward cycle in which synaptic and immune mechanisms reinforce one another until the network is locked into a pathological state — acquires, with the net at its centre, a concrete mechanism and an anatomical address. Stripping the net does not injure only the interneuron it surrounded; it injures the excitatory cortex through three channels at once, and these channels close the loop.
The first channel is disinhibition. A silenced parvalbumin cell no longer paces the pyramidal neurons it innervated, and the result is the hyperexcitable, hypersynchronous, gamma-deficient cortex the Moosmann framework describes — and, critically, the activity-dependent amplifier the synthesis requires, because the synaptic release of amyloid-β and of tau rises with neuronal firing. A disinhibited excitatory neuron is therefore not merely injured; it is conscripted into manufacturing the very species that drive the disease. The second channel is direct: the netted excitatory neurons that lose their own matrix lose the tau-exclusion barrier the net conferred, and in the human cortex these neurons carry markedly lower tau while their nets are intact (de Vries et al., 2024) — so their stripping renders them newly tau-permissive. The third channel is biophysical: the glutamatergic synapses stripped of the surrounding matrix lose the receptor-mobility discipline the net imposed (Frischknecht et al., 2009) and are returned, at the worst possible moment, to a labile, juvenile-like plasticity (Pizzorusso et al., 2002).
The loop now closes upon the net. Net digestion disinhibits and destabilises the excitatory cortex; the hyperexcitable, tau-permissive excitatory neurons release more amyloid and tau; the rising pathology sustains the post-homeostatic, matrix-degrading, complement-discharging microglia of the neuroimmune node; and the intensified microglial activity strips further net — from the next interneuron, and from the next netted excitatory cell. Each turn of the loop removes more matrix, and each removal of matrix drives the loop another turn. This is the synthesis's self-sustaining loop, given a structural engine: the perineuronal net, whose loss begins the loop, whose continued loss propagates it, and whose preservation would break it.
VII. The Inhibitory Geometry
The single-cell atlases supply the synthesis a fact it does not fully explain: the somatostatin interneuron is an early casualty, depleted as the molecular pathology begins to rise, while the parvalbumin interneuron is a late casualty, lost in the disease's final epoch (Gabitto et al., 2024). On the surface this ordering complicates any account that centres the parvalbumin cell. The matrix hypothesis resolves it into a single variable, because the two cells differ in precisely the respect the hypothesis makes decisive: the parvalbumin interneuron is enwrapped by a net, and the somatostatin interneuron, a dendrite-targeting cell, is not.
The disease therefore performs its own controlled experiment. The somatostatin cell shows what becomes of an unprotected fast-inhibitory neuron facing the convergent pathology with no matrix between itself and its environment: it is taken early, because nothing shields it. The parvalbumin cell shows what becomes of one whose protection is removed: it survives into the disease's final phase, and only for as long as its net endures. The ordering of inhibitory failure — somatostatin first, parvalbumin last — is thus not a difficulty but a demonstration, and it is written by a single variable: when each cell loses, or never had, its net.
| Population | Net status | Timing of loss | Reading under the matrix hypothesis |
|---|---|---|---|
| Somatostatin interneuron | netless (dendrite-targeting) | early | unshielded throughout; falls as the convergence arrives |
| Parvalbumin interneuron | aggrecan-enwrapped | late | protected until the net is stripped, then a slow latent failure |
VIII. The Unified Timeline
With the net established as the structural node, the eight frameworks and the matrix can be assembled into a single timeline — a stereotyped half-century sequence in which each framework's load-bearing event is placed at its epoch, and the status of the net is tracked alongside. The timeline's organising principle is the one the matrix hypothesis supplies: the disease is intracellular and compensable while the net is intact, and becomes extracellular, circuit-level, and self-sustaining once the net is stripped.
| Phase | Epoch | Load-bearing molecular events (by framework) | Perineuronal-net status | Clinical correlate |
|---|---|---|---|---|
| I — Molecular Initiation | decades 3–5 (~ −30 to −15 yr) | lipid peroxidation of ApoE-borne PUFA and ApoER2–Dab1 disruption (Ramsden); incipient excitatory insufficiency and compensatory Aβ/tau (Moosmann, Rappoport); ApoE4 lowers every threshold | intact and protective; PV cells shielded; antioxidant and cation-buffering offices fully operative | asymptomatic; no pathology on imaging |
| II — Endosomal Dysregulation | ~ −15 to −5 yr | retromer traffic jam and βCTF accumulation (Small); intraneuronal Aβ42 in synaptic MVBs (Gouras); monomer depletion begins (Huang); GSK3β rising | intact but stressed; iron loading and oxidative burden begin to accrue on the matrix; SST (netless) interneurons begin to fail | asymptomatic; amyloid-positive |
| III — The Proteolytic Turn | ~ −5 to 0 yr | GSK3β→LIMK1→cofilin and Ephexin5–RhoA–ROCK engaged (Ramsden, Margolis); MMP-9/ADAMTS hyperactivation (Huang); iron-Fenton fragmentation; complement priming of the matrix | aggrecan core stripped; the brake on the cytoskeleton released; PV cell loses its shield and enters the latent interval (silencing begins) | transition to MCI |
| IV — Neuroimmune Synaptic Elimination | 0 to ~5 yr | C4d–LilrB2 deposition and cofilin-actin rods on perisomatic synapses (Shatz–Brott); microglial engulfment of net and synapse; Nav1.1/NPTX2 loss; gamma collapse; activity-dependent Aβ/tau amplification | digested and engulfed; the loop closes on the matrix; netted excitatory neurons become tau-permissive | mild-to-moderate dementia |
| V — Network Disintegration | ~5 to 15 yr | latent PV cells finally fail (frank loss); widespread synapse loss; tangles mature; extracellular plaques as gravestones | gone in affected fields; no matrix to protect the surviving inhibition | moderate-to-severe dementia |
Two features of the timeline are worth drawing out. First, the hinge of the disease is Phase III, the Proteolytic Turn, at which the net is stripped: everything before it is the slow, intracellular, compensable accrual the endosomal and compensatory frameworks describe, and everything after it is the rapid, extracellular, circuit-level collapse the neuroimmune and cytoskeletal frameworks describe. The matrix hypothesis identifies this hinge as a structural event — the loss of one lattice — and thereby gives the synthesis's two halves a single point of articulation. Second, the latent interval spans Phases III and IV: the net is stripped at the start of Phase III, but the parvalbumin cell is not frankly lost until Phase V, and through the intervening years it persists in a silenced, reawakenable state. That interval — measured in years, not weeks — is the disease's window of rescue, and it is invisible to any framework that does not separate the loss of the net from the loss of the neuron.
IX. The Therapeutic Window, Re-stated
The convergent synthesis concluded that monotherapy must fail, because the disease is a network with redundant arms, and that effective intervention must be early and multi-targeted. The matrix hypothesis sharpens this conclusion into a single, measurable target and a single, definable window. The target is not the dead neuron, nor the upstream protein whose accrual may have closed years before; it is the protection of the neuron — the preservation of an intact net where one remains, and the restoration of net integrity, or of the protective offices the net performed, where it has begun to be lost. Because the net is the structure whose stripping defines the hinge of the disease and whose preservation would arrest the self-sustaining loop, it is also the natural biomarker of whether the transition is being prevented: a measure of perineuronal-net integrity around the parvalbumin population reports which side of the hinge a given brain occupies and whether an intervention is holding it back.
The window is the latent interval. An intervention that restores the net, or its functions, within the interval — after the matrix is stripped but before the parvalbumin cell has frankly failed — can in principle reawaken the silenced cell and re-close the loop; the same intervention after frank loss cannot. This predicts a discontinuity in therapeutic outcome at a definable cellular point, and it nominates, from a single structure, both what to preserve and how to know whether one is succeeding. The synthesis pointed toward multifaceted intervention at endosomal, oxidative, excitatory, and immune nodes; the matrix hypothesis adds that all four of these are, at the perisomatic zone, arms of a single attack on one lattice, and that defending the lattice is the most parsimonious way to interrupt them together.
X. Falsifiable Predictions
The matrix hypothesis, like the synthesis it extends, earns its standing by exposing itself to refutation, and several of its predictions are testable with existing methods.
On the hinge. In the human cortex, the aggrecan core of the parvalbumin perineuronal net will be found progressively degraded with disease stage — by core-directed antibody and biochemical readout, even where the glycan coat is partially retained — and the degradation will be detectable before the depletion of the parvalbumin population and before the loss of the perisomatic synapses the net surrounds. A finding that the aggrecan core is intact through the late disease would falsify the hypothesis.
On the gating of the cytoskeletal node. Cofilin-actin rods and C4d–LilrB2-driven spine collapse will be found preferentially at synapses whose surrounding net has been degraded, and experimental preservation of the net will reduce the formation of those rods despite the presence of the upstream amyloid, tau, and complement signals.
On the inhibitory geometry. The early loss of somatostatin interneurons and the late loss of parvalbumin interneurons will track their net status — the early-lost cells netless, the late-lost cells netted — and conferring net-like protection on the unprotected population, or accelerating net loss on the protected one, will move each cell's timing in the predicted direction.
On the latent interval. Parvalbumin neurons in the interval between net loss and frank loss will be found alive but downregulated, and restoration of net integrity within that interval will recover parvalbumin expression, Nav1.1, and fast-spiking function — whereas the same restoration after frank loss will not.
On the loop. The transition from the compensable to the self-sustaining phase of the disease will coincide with net stripping, and interventions that preserve the net will arrest the disease before that point and fail after it — a discontinuity, not a continuum.
Each prediction is a measurement addressed to one structure. That is the discipline a structural theory imposes on a molecular synthesis: the eight frameworks may each be partly right about mechanism, but the matrix hypothesis must be wrong, if it is wrong, about the net.
XI. Conclusion
The Convergent Synaptic Collapse established that Alzheimer's disease is not the work of a single molecule but the convergent outcome of parallel molecular insults — oxidative, trafficking, excitatory, proteostatic, and immune — that meet at shared nodes and overwhelm the synaptic architecture. That synthesis told us, in unprecedented mechanistic detail, how the synapse dies. It did not tell us where the convergence is staged, and without a place, its most important node — the cytoskeletal collapse on which three frameworks agree — remained overdetermined but unlocated, and its timeline remained a sequence of molecular events without a hinge.
This paper has argued that the place is the perineuronal net, and that the net is the seventh and structural node of the synthesis: the protective shield whose loss exposes the parvalbumin interneuron, the protease and complement target on which the neuroimmune machinery discharges, the iron sink on which the lipid-peroxidative chemistry runs, and the scaffold surrounding the perisomatic synapses at which the cytoskeletal collapse is executed. To strip the net is to let four convergent arms arrive together at one cell. Reading the disease through the net resolves the geometry of inhibitory failure into a single variable, separates the lesion from the death by a latent interval that is the disease's window of rescue, and assembles the eight frameworks into a unified timeline with a definable structural hinge. The synthesis described the convergence; the matrix names its address. And the address is a structure one can measure, defend, and — for a window that the disease itself leaves open — restore.
References
Cabungcal, J.-H., Steullet, P., Morishita, H., Kraftsik, R., Cuenod, M., Hensch, T. K., & Do, K. Q. (2013). Perineuronal nets protect fast-spiking interneurons against oxidative stress. Proceedings of the National Academy of Sciences USA, 110(22), 9130–9135.
Celio, M. R. (1986). Parvalbumin in most γ-aminobutyric acid-containing neurons of the rat cerebral cortex. Science, 231(4741), 995–997.
Crapser, J. D., Spangenberg, E. E., Barahona, R. A., Arreola, M. A., Hohsfield, L. A., & Green, K. N. (2020). Microglia facilitate loss of perineuronal nets in the Alzheimer's disease brain. EBioMedicine, 58, 102919.
de Vries, L. E., Huitinga, I., Kessels, H. W., Swaab, D. F., & Verhaagen, J. (2024). The concept of resilience to Alzheimer's disease: perineuronal nets as a protective factor. Alzheimer's & Dementia, 21(2), e14504.
Fawcett, J. W., Oohashi, T., & Pizzorusso, T. (2019). The roles of perineuronal nets and the perineuronal matrix in memory and disease. Nature Reviews Neuroscience, 20(8), 451–465.
Frischknecht, R., Heine, M., Perrais, D., Seidenbecher, C. I., Choquet, D., & Gundelfinger, E. D. (2009). Brain extracellular matrix affects AMPA receptor lateral mobility and short-term synaptic plasticity. Nature Neuroscience, 12(7), 897–904.
Gabitto, M. I., Travaglini, K. J., Rachleff, V. M., et al. (2024). Integrated multimodal cell atlas of Alzheimer's disease. Nature Neuroscience, 27(12), 2366–2383.
Härtig, W., Brauer, K., & Brückner, G. (1992). Wisteria floribunda agglutinin-labelled nets surround parvalbumin-containing neurons. NeuroReport, 3(10), 869–872.
Hong, S., Beja-Glasser, V. F., Nfonoyim, B. M., et al. (2016). Complement and microglia mediate early synapse loss in Alzheimer mouse models. Science, 352(6286), 712–716.
Hu, H., Gan, J., & Jonas, P. (2014). Fast-spiking, parvalbumin⁺ GABAergic interneurons: from cellular design to microcircuit function. Science, 345(6196), 1255263.
Margolis, S. S., et al. (2017). Reducing expression of the synapse-restricting protein Ephexin5 ameliorates Alzheimer's-like impairment in mice. Journal of Clinical Investigation, 127(5), 1815–1823.
Moosmann, B., & Sohre, S. (2018). The pathological hallmarks of Alzheimer's disease derive from compensatory responses to NMDA receptor insufficiency. bioRxiv / Oskar Fischer Prize framework.
Morawski, M., Brückner, G., Jäger, C., Seeger, G., & Arendt, T. (2010). Neurons associated with aggrecan-based perineuronal nets are protected against tau pathology in subcortical regions in Alzheimer's disease. Neuroscience, 169(3), 1347–1363.
Pizzorusso, T., Medini, P., Berardi, N., Chierzi, S., Fawcett, J. W., & Maffei, L. (2002). Reactivation of ocular dominance plasticity in the adult visual cortex. Science, 298(5596), 1248–1251.
Ramsden, C. E., Keyes, G. S., et al. (2023). ApoER2–Dab1 disruption as the origin of pTau-associated neurodegeneration in sporadic Alzheimer's disease. Acta Neuropathologica Communications / Oskar Fischer Prize framework.
Small, S. A., & Petsko, G. A. (2015). Retromer in Alzheimer disease, Parkinson disease and other neurological disorders. Nature Reviews Neuroscience, 16(3), 126–132.
Suttkus, A., Rohn, S., Weigel, S., Glöckner, P., Arendt, T., & Morawski, M. (2014). Aggrecan, link protein and tenascin-R are essential components of the perineuronal net to protect neurons against iron-induced oxidative stress. Cell Death & Disease, 5(3), e1119.
Terry, R. D., Masliah, E., Salmon, D. P., et al. (1991). Physical basis of cognitive alterations in Alzheimer's disease: synapse loss is the major correlate of cognitive impairment. Annals of Neurology, 30(4), 572–580.
Verret, L., Mann, E. O., Hang, G. B., et al. (2012). Inhibitory interneuron deficit links altered network activity and cognitive dysfunction in Alzheimer model. Cell, 149(3), 708–721.
Werneburg, S., Brott, B. K., Shatz, C. J., et al. (2025). C4d, a high-affinity LilrB2 ligand, is elevated in Alzheimer's disease and mediates synapse pruning. Proceedings of the National Academy of Sciences USA, 122.
Xiao, M.-F., Xu, D., Craig, M. T., et al. (2017). NPTX2 and cognitive dysfunction in Alzheimer's disease. eLife, 6, e23798.
research/matrix-of-collapse/ONS_MatrixOfCollapse_Thesis.md