Issue 31 · Pick 05 Neuroscience ✓ read
Neuropil aggrecan, not perineuronal nets, closes the critical period for visual plasticity
TL;DR. For twenty years the textbook story has been that perineuronal nets (PNNs) — the lattice-like extracellular matrix that wraps parvalbumin interneurons — slam shut the critical period for visual cortex plasticity. This paper takes the story apart genetically. By deleting the gene for aggrecan (the core PNN protein) in specific cell types, the authors show that you can eliminate PNNs entirely and the critical period still closes normally. What actually keeps adult cortex plastic is aggrecan made by excitatory neurons and deposited diffusely in the neuropil — not the nets themselves. The enzyme experiment everyone cited (chondroitinase ABC) was reopening plasticity by chewing up that diffuse aggrecan, and the field misattributed the effect to the visible nets.
The dogma and why it was shaky
A quick refresher on the assay, because everything hinges on it. In juvenile mice (~P21–32), suturing one eye shut for 4–5 days — monocular deprivation, MD — dramatically shifts cortical responses toward the open eye. This is ocular dominance (OD) plasticity, quantified by the contralateral bias index (CBI), where 0.5 means the two eyes drive cortex equally, or by the ocular dominance index (ODI) from calcium imaging, where ~0.3 is the normal contra-biased resting state. After the critical period closes, the same deprivation does almost nothing. The critical period is thus a clean, bidirectional readout of "is the cortex still plastic?"
PNNs became the favored "brake" because their timing is suggestive: they thicken as the critical period closes and plateau in adulthood. And the killer experiment — Pizzorusso et al. 2002 — injected chondroitinase ABC (chABC) into adult visual cortex, degraded the nets, and restored OD plasticity.
Here's the crack the authors drive a wedge into: chABC is a blunt instrument. It doesn't just dissolve PNNs. It cleaves the sugar chains on all chondroitin sulfate proteoglycans, plus dermatan sulfate and hyaluronan, wherever they are — including the aggrecan diffusely spread through the neuropil between cells. The nets are just the visible, condensed tip of a much larger iceberg of matrix. So "chABC reopens plasticity" never actually pinned the effect on the nets. It pinned it on the matrix, somewhere.
The idea: let genetics do what the enzyme can't
The move is elegant. Aggrecan is encoded by one gene, Acan. Use a conditional ("floxed") Acan allele and cross it to different Cre driver lines, each of which deletes the gene only in a particular cell population. Now you can ask, separately: what happens when aggrecan is gone from inhibitory neurons (which is where the PNNs live) versus from excitatory neurons (which contribute to the neuropil matrix)?
The reason this cleanly separates the two hypotheses: PNNs in visual cortex almost exclusively ensheath PV+ interneurons. So an inhibitory deletion should wipe out PNNs. An excitatory deletion should leave the PV nets intact while stripping aggrecan from the surrounding neuropil. If PNNs are the brake, only the first should reopen plasticity.
What they found
First they reproduced the classic result with a cleaner tool: instead of repeatedly injecting purified enzyme, they used AAVs to make neurons constitutively express chABC. This abolished PNNs across V1 and restored adult OD plasticity — deprived mice showed the low CBI of juveniles, significantly different from non-deprived or enzyme-inactive controls (P < .0001). So the phenomenon is real and their measurement is sound.
Then the dissection. Three different Cre lines that all hit PV interneurons — PV-Cre, Dlx5/6-Cre, GAD2-Cre — each "decimated" PNNs in V1. And yet none reopened plasticity: after 4–5 days of MD the CBI was statistically indistinguishable from non-deprived animals (P ranging .47 to >.99). No nets, but the critical period closed right on schedule.
By contrast, deleting Acan in all neurons (BAF53b-Cre) sustained juvenile-like plasticity into adulthood (non-deprived vs. MD, P < .0001). They confirmed this a second way — two-photon calcium imaging in awake mice — where BAF53b ODI shifted to ~0.0 (a real binocular rebalancing) while control and GAD2 mice stayed near the non-deprived ~0.3.
The cleanest single experiment is Camk2a-Cre, which deletes Acan only in excitatory forebrain neurons. Here the PNNs "appeared unaffected" — the nets around PV cells are still there — and yet adult mice were fully plastic, matching BAF53b. That is the dissociation the whole paper needs: nets present, plasticity open. It rules out the escape hatch that some residual net damage was doing the work.
The circuit and molecular correlates
Where does the plasticity come from mechanistically? They mapped excitatory inputs onto L2/3 PV interneurons using laser-scanning photostimulation (glutamate uncaging over a grid, recording the patched PV cell). Removing PNNs alone (GAD2-Cre) changed nothing about the input maps. But BAF53b-Cre — the plastic condition — showed a moderate, significant reduction in excitatory drive onto PV cells, concentrated in the L4→L2/3 inputs. Less excitation of inhibitory cells nudges the excitation/inhibition balance toward plasticity, and notably this mirrors the Ngr1 (Nogo receptor) mutant, one of the few other genetic handles on critical-period closure. Importantly, this didn't degrade cortex: response strength, orientation tuning width, and behavioral visual acuity were all normal.
The spatial transcriptomics (Xenium, ~15,000 cells per section, two mice per genotype) is more of a supporting sketch than a smoking gun, and the authors say so. Very few genes changed. The plastic genotypes (Camk2a, BAF53b) showed modestly elevated immediate-early genes — Npas4, Arc (~1.75×), Dusp1 (2–3.4×), Nr4a1, Ccn1 — which they read as a consequence of ongoing plasticity rather than its cause. They pointedly note the Arc bump is far smaller than the ~5× that a prior study needed to force adult plasticity. So no clean transcriptional mechanism yet; the "how does neuropil aggrecan gate plasticity" question is left open.
They also directly addressed the obvious objection — maybe deleting Acan only in adult inhibitory neurons (rather than from birth) would behave like chABC. It didn't: an adult-onset inhibitory deletion via AAV-hDlx-iCre erased PNNs but left the critical period firmly closed.
Why this matters, and what to doubt
If it holds, this is a genuine correction to a heavily cited model. The perineuronal net has become a load-bearing concept well beyond vision — invoked in fear-memory erasure, addiction, spatial memory, and various therapeutic "reopen the brain" ambitions. A large fraction of that literature leans on chABC and on the intuition that the visible nets are the plasticity brake. This paper says: the nets are a correlate, not the cause, at least for visual OD plasticity; the causal agent is a diffuse pool of the same molecule secreted by excitatory neurons. That reframes what to target and reinterprets a lot of enzyme-injection results.
The methodological subtext is also worth internalizing: the field's strongest positive controls (chABC injection; the Hapln1 link-protein mutant) used tools or readouts — a promiscuous enzyme, or visually-evoked potentials that don't faithfully track spiking — that couldn't distinguish nets from neuropil. Here they insist on multi-unit spiking plus single-cell calcium imaging, the gold standards.
What to be skeptical about. First, "PNNs appeared unaffected" in the Camk2a mice rests on WFA staining, a coarse readout; a subtle change in net composition or in the aggrecan content within nets can't be excluded from images alone, and this is the linchpin of the dissociation. Second, the transcriptomics is n = 2 mice per genotype and finds no mechanistic through-line, so the claim is presently "aggrecan somewhere in the neuropil, made by excitatory cells" rather than a molecular pathway. Third, the effect on PV excitatory drive is described as "moderate," and it's a correlate — they haven't shown that restoring that drive re-closes the period. Finally, this is OD plasticity in mouse V1; whether the same neuropil-vs-net logic governs the other PNN-implicated behaviors is explicitly left for future work.
If you read one thing, read the Results section "PNNs do not close the critical period" alongside Figure 1 — that's where the genetic logic and the two independent plasticity readouts land the core claim. The Discussion is the best place to see how carefully they distinguish their conclusion from a century of circumstantial PNN evidence.