A dendritic mechanism for balancing synaptic flexibility and stability

SUMMARY

Biological and artificial neural networks learn by modifying synaptic weights, but it is unclear how these systems retain previous knowledge and also acquire new information. Here, we show that cortical pyramidal neurons can solve this plasticity-versus-stability dilemma by differentially regulating synaptic plasticity at distinct dendritic compartments. Oblique dendrites of adult mouse layer 5 cortical pyramidal neurons selectively receive monosynaptic thalamic input, integrate linearly, and lack burst-timing synaptic potentiation. In contrast, basal dendrites, which do not receive thalamic input, exhibit conventional NMDA receptor (NMDAR)-mediated supralinear integration and synaptic potentiation. Congruently, spiny synapses on oblique branches show decreased structural plasticity in vivo. The selective decline in NMDAR activity and expression at synapses on oblique dendrites is controlled by a critical period of visual experience. Our results demonstrate a biological mechanism for how single neurons can safeguard a set of inputs from ongoing plasticity by altering synaptic properties at distinct dendritic domains.

In brief

Yaeger et al. report that neurons in the adult visual cortex organize plastic and stable synapses in discrete dendritic regions. Stable synapses are located in oblique dendrites of layer 5 pyramidal neurons and have distinct physiological properties and protein expression. This mechanism helps protect specific inputs from ongoing experience-dependent plasticity.

Graphical Abstract

INTRODUCTION

The brain learns continuously by balancing plasticity and stability, whereby new information is incorporated without disrupting prior knowledge. Synaptic plasticity is the key principle for learning, but neural circuits must also exhibit sufficient synaptic stability to maintain information storage over the long term. This process poses a major unsolved problem for both biological and artificial neural networks, known as the plasticity-stability dilemma.^1–4 In artificial learning systems, this problem manifests as catastrophic forgetting, where previously acquired information is overwritten unless weight changes are limited.^5–7 While half a century of experimental and theoretical work on biological synaptic plasticity has focused on how plasticity is induced^8–17 and maintained within a working range,^18–22 little is currently known about how neurons sustain previously learned information throughout a lifetime of experience-dependent plasticity.

The plasticity-stability dilemma is particularly relevant for the adult mammalian cortex, which flexibly learns many perceptual associations and complex cognitive functions without compromising stable sensory representations. However, mechanistic cortical synaptic plasticity experiments have been almost exclusively conducted in developing animals, when sensory maps are still undergoing robust experience-dependent plasticity^{8,11–14,23–27} (but see reference 17). Adult synaptic plasticity research has been primarily conducted in the highly plastic hippocampus^9,15,28–34 or with indirect methods in cortex, such in vivo dendritic spine imaging,^35–44 extracellular stimulation,^45–50 and/or sensory map plasticity with deprivation or neuromodulation.^51–58 Elucidating the mechanisms and activity requirements for adult synaptic plasticity is critical for understanding how the cortex balances stability with flexibility.

In the sensory cortex, initial sensory experiences must be stored and stabilized for robust sensory processing, but these circuits must also retain flexibility to accommodate subsequent perceptual learning in adulthood.⁵⁹ For example, primary sensory inputs in the visual cortex (V1) from the dorsal lateral geniculate nucleus of the thalamus (dLGN) rapidly change at the onset of vision.^60–63 By adulthood, these inputs are stable and no longer plastic during normal visual experience.^64–67 However, higher-order visual representations remain highly plastic.^68–70

Because single cortical neurons receive spatially targeted inputs within distinct dendritic domains,^71–74 we hypothesized that dendritic compartmentalization could arbitrate plasticity or stability for different kinds of information. Using complimentary physiological and imaging techniques, we identified dLGN-recipient and non-recipient dendritic domains within layer 5b pyramidal neurons (L5b PNs) and characterized synaptic properties and plasticity within these domains across maturation.

RESULTS

L5b PN oblique dendrites selectively receive thalamic input and exhibit distinct integrative properties from basal dendrites

Visual information from dLGN projects primarily to cortical layer 4.^75–79 First-order thalamic inputs are known to synapse on L5 PNs,^80–82 which integrate inputs from all cortical layers, although the location of the synaptic contacts is unclear in V1. Physiological data from juvenile V1⁷³ and anatomical data⁸³ suggest that dLGN inputs target the apical oblique dendrites of L5 PNs, which can reside in layer 4. To directly test where dLGN inputs make functional synapses within the dendritic arbor of L5 PNs in adult mouse V1, we employed subcellular channelrhodopsin (ChR2)-assisted circuit mapping (sCRACM).⁷⁴ After viral expression of ChR2 in dLGN (Figure 1A), we performed whole-cell patch-clamp recordings from V1 L5b PNs in acute slices from adult mice (postnatal day 56+). ChR2-expressing dLGN axons were activated with a 473-nm laser, using a grid of ~50-μm diameter spots that spanned the entire neuron, in the presence of tetrodotoxin to block axonal conduction. Local photostimulation of ChR2-expressing dLGN boutons across the dendritic tree revealed monosynaptic excitatory postsynaptic potentials (EPSPs) tightly restricted to apical oblique branches in layer 4 (Figures 1B and 1C).

Figure 1. Apical oblique dendrites of adult V1 L5b PNs selectively receive dLGN input and do not have NMDAR-mediated integrative properties.

(A) Example confocal image of a coronal brain slice showing dLGN neurons infected with AAV2-hSyn-mCherry-ChR2. Scale bar: 500 μm.

(B) Left: example two-photon image of a V1 L5b PN whole-cell recording with overlaid photostimulation grid (blue). Scale bar, 50 μm. Center: average evoked EPSPs from six repetitions of 473-nm photostimulation for the neuron and grid shown. Soma position is indicated by the triangle. Right: heatmap of average evoked EPSPs. Each square is 50 μm.

(C) Left: normalized heatmap of pooled L5b PNs (n = 9 neurons, 9 slices, 3 P56+ animals), aligned at the soma. Each square is 50 μm. Right: mean normalized input as a function of depth for all neurons. Shaded region indicates the bootstrapped 95% confidence interval.

(D) Example two-photon z stack of a L5b PN with glutamate uncaging sites indicated on an oblique (purple) and a basal (green) branch; scale bar: 50 μm. Uncaging sites (stars) and linescan (yellow) are magnified in insets (scale bars: 10 μm).

(E) Expected EPSPs, measured EPSPs, and local branch Ca²⁺ signals for oblique (top) and basal (bottom) dendrites shown in (D). Left: expected voltages from increasing numbers of linearly summed inputs. Center: measured EPSPs in response to increasing numbers of synchronous uncaging inputs. Stimulation of all inputs led to an action potential (AP, arrow). Right: corresponding changes in local branch OGB-1 signal (ΔF/F). Colored trace indicates the supralinear threshold and significant Ca²⁺ signal.

(F) Measured voltage (left) and branch ΔF/F (right) as a function of expected voltage for the branches shown in (D). Dashed line indicates linearity (left) and threshold Ca²⁺ signal (right).

(G) Adult population gain (measured/expected, left) and branch ΔF/F (right). n = 13 basal branches, 7 animals and 9 oblique branches, 6 animals. All animals were P56+. Basal APs removed for clarity. Error bars are SEM. Basal vs. oblique gain (AP excluded), Mann-Whitney U test: p = 7.63–4; basal vs. oblique ΔF/F (AP excluded), Mann-Whitney U test: p = 1.08–5.

See also Figure S1.

Dendritic domains that receive distinct long-range inputs can also have specific integration modes.⁷¹ In light of this specific targeting of dLGN inputs to L5b PN oblique dendrites, we asked whether these branches exhibited distinct integrative properties compared to nearby basal dendrites (Figure 1D), which receive intracortical inputs.^72,74,84 In acute slices from adult mice, we combined somatic patch-clamp recording with two-photon imaging and glutamate uncaging at either basal or oblique dendritic branches. Uncaging-evoked EPSPs were recorded at increasingly large groups of spines, and these measurements were compared with the arithmetic sum of unitary EPSPs. Local branch calcium, measured with OGB-1, was also recorded simultaneously. Basal dendrites displayed highly supralinear integration and large local Ca²⁺ signals in response to spatiotemporally clustered glutamate uncaging (Figures 1E–1G). These local Ca²⁺ signals and voltage supralinearities are characteristic of N-methyl-D-aspartate (NMDA) spikes in thin cortical dendrites (see Figure S3).^{12,24,85–95} In contrast, oblique dendrites integrated inputs linearly until axosomatic action potential threshold was reached and did not show significant subthreshold Ca²⁺ influx (Figures 1E–1G), similar to what has been previously described in the oblique dendrites of retrosplenial cortex.⁷¹ These differences between oblique and basal dendrites could not be explained by the size of uncaging events, the number of inputs, distance from the soma, action potential threshold, or δV/δt prior to action potential initiation (Figure S1). Thus, oblique dendrites of adult V1 L5b PNs selectively receive dLGN input and exhibit a specialized linear integration mode.

Oblique dendrites lack burst-timing synaptic potentiation in adult V1 L5b PNs

The lack of large subthreshold Ca²⁺ signals in oblique dendrites led us to hypothesize that synapses on these branches may exhibit plasticity differences compared to basal dendrites. It is well established that elevated postsynaptic Ca²⁺ concentration is key for one of the major forms of long-term potentiation at cortical synapses, driven by strong dendritic depolarization and NMDA receptor (NMDAR) activation.^92,96,97 To compare long-term potentiation across dendritic compartments, we sought an induction protocol that was physiologically relevant and effective at both proximal and more distal synapses. Several studies in L5 PNs show that pairing single pre- and postsynaptic action potentials does not induce synaptic plasticity at low frequencies^8,11,23 or at physiological extracellular Ca²⁺ concentrations.^98,99 High-frequency bursts of action potentials, however, can potentiate proximal and distal inputs^{8,11–13,17,23,25,29,100} under physiological [Ca²⁺]^98,99 and are a primary firing mode of L5 PNs.^101–105 Therefore, we tested synaptic potentiation in adult V1 L5b PNs by pairing focal electrical stimulation of presynaptic axons at basal or oblique branches with burst firing via somatic current injection. A bipolar stimulating electrode was positioned within 10 μm of either a basal or an oblique dendritic branch (Figures 2A, 2B, and S2) to generate a 1- to 2-mV EPSP in the presence of a GABA-A antagonist. EPSPs were paired with 20-ms somatic current injections that drove bursts of 2–4 action potentials. The peak of the EPSP preceded the peak of the first action potential by 10 ms for optimal potentiation.^8,11,23 Five pairings at 10 Hz (i.e., theta rhythm) were repeated 30 times, with 10 s between epochs (Figure 2C). This protocol robustly potentiated synapses on basal dendrites (Figures 2D,2F,-and 2G and S2), similar to findings in juvenile cortex.¹² Potentiation at basal branches required pre- and postsynaptic coincidence as well as NMDARs (Figures 2G and S2). In contrast, synapses on oblique dendrites did not exhibit any significant potentiation after pairing (Figures 2E–2G and S2). Doubling the number of pairings did not change this outcome (Figure S2). These findings indicate that unlike basal dendrites, oblique branches of L5b PNs in adult V1 lack the capability for synaptic potentiation under a physiologically relevant plasticity induction protocol.

Figure 2. Synapses in the oblique dendritic domain lack long-term potentiation.

(A) Example two-photon z stack of whole-cell recording with focal stimulation (green) near a basal dendrite. Scale bar: 25 μm.

(B) As in (A), but for an oblique dendrite (purple).

(C) Plasticity induction protocol. Five pairings of pre- and postsynaptic stimulation at 100 Hz, repeated 30 times at 0.1 Hz. Right: a 10-ms interval between the peak of the evoked EPSP and the peak of the first action potential. Scale bar: 20 mV, 50 ms.

(D) Example recording with basal dendrite stimulation, from the cell shown in (A). After 10 min of baseline recording, the induction protocol is delivered (arrow). Inset shows average EPSP before (black) and after (color) pairing. Scale bar: 1 mV, 25 ms. Responses are 1-min binned averages and normalized to baseline.

(E) As in (D), but for oblique dendritic stimulation for the cell shown in (B).

(F) Normalized EPSP amplitude for neurons stimulated at basal (green) or oblique (purple) branches (2-min bins; n = 10 basal branches, 10 animals, and 13 oblique branches, 12 animals. All animals were P56+.). Error bars are SEM. Basal vs. oblique postinduction EPSP, Mann-Whitney U test: p = 4.95E–63.

(G) Change in EPSP after induction for all oblique branches, basal branches, and basal branches with DAP5 or with only presynaptic or postsynaptic activation during induction (see Figure S2 for n). Basal vs. oblique average change in EPSP, Mann-Whitney U test: p = 6.33E–5. All other comparisons were not significant.

Boxplot: median, 25th, and 75th percentiles with ±2.7σ whiskers.

See also Figure S2.

Synaptic plasticity in oblique dendrites is limited to a postnatal critical period

NMDAR-mediated responses and long-term potentiation decline over the course of early postnatal development at thalamocortical synapses.^106,107 These changes coincide with the end of the critical period, after which thalamocortical and intracortical inputs become less susceptible to plasticity driven by sensory experience.^{61,67,77,108,109} Therefore, we hypothesized that the thalamorecipient oblique dendrites of L5b PNs may exhibit more NMDAR-mediated supralinear integration and plasticity during postnatal development. To investigate this, we conducted the same glutamate uncaging and synaptic plasticity tests at basal and oblique dendrites of L5b PNs during postnatal development. We found that at eye opening (postnatal days [P] 12–14), oblique and basal dendrites displayed similar NMDAR-dependent supralinearities and large local Ca²⁺ signals (Figures 3A–3C and S3). These properties persisted into the canonical critical period of postnatal development in V1 (P18–22)^67,110 and are consistent with previous reports of NMDA spikes in basal dendrites of L5 PNs in developing cortex.^12,85,89,90 However, by 4 weeks of age (P28–32), oblique dendrites had developed adult-like properties, with linear synaptic integration and minimal local Ca²⁺ signals (Figures 3C and S3). Synaptic plasticity at oblique dendrites followed the same developmental trajectory: synapses on oblique and basal dendrites exhibited NMDAR-dependent burst-timing synaptic potentiation at juvenile ages (Figures 3D and S3), but by P28, burst-timing synaptic potentiation on oblique (but not basal) dendrites was not evident. The loss of synaptic potentiation at thalamorecipient oblique dendrites aligns with a previous finding that S1 thalamocortical synapses do not sustain long-term potentiation or large NMDAR synaptic currents after a postnatal critical period.¹⁰⁷ Because the period in which oblique dendrites lose NMDAR-dependent properties overlaps with the critical period in V1, our findings suggested that visual experience mediates the maturation of oblique dendrite properties. To test this, we conducted these same experiments in mice deprived of light from birth, which substantially alters thalamic drive¹¹¹ and critical period plasticity.^112,113 In P28 dark-reared animals, oblique dendrites retained immature NMDAR-dependent supralinear integration with large local Ca²⁺ signals and robust synaptic potentiation (Figures 3E and S3). Furthermore, dark-reared mice re-exposed to a normal light cycle for 2–4 weeks did not exhibit adult-like integrative properties (Figure S3), indicating that the oblique compartment matures only during a specific window in development. Collectively, these results reveal a restricted developmental window in which visually driven activity shifts the L5b PN oblique dendritic compartment to a linearly integrating, reduced plasticity mode.

Figure 3. Oblique dendrites lose NMDAR-mediated properties after an experience-dependent critical period in postnatal development.

(A) Expected (left), measured (center), and ΔF/F (right) for oblique (top) and basal (bottom) dendrites from the same neuron from a P14 mouse. Note supralinear integration and large Ca²⁺ signals in both branches. Both are driven to action potential initiation (AP, arrow).

(B) P14 population gain (measured/expected, top) and branch ΔF/F (bottom) (n = 9 basal branches and 9 oblique branches, both from 5 animals). Error bars are SEM. Basal vs. oblique gain*, p = 0.33; ΔF/F*, p = 0.70.

(C) Maximum gain (top) and local branch ΔF/F (prior to AP initiation, bottom) across developmental time points (adult, P56+: n as in Figure 1, basal vs. oblique gain*, p = 1.07E–4; ΔF/F*, p = 1.07E–4. P14: n as in [B]; basal vs. oblique gain*, p = 0.54; ΔF/F*, p = 0.76. P21: n = 9 basal branches, 4 animals; 6 oblique branches, 2 animals, basal vs. oblique gain*, p = 0.52; ΔF/F*, p = 0.68. P28: n = 10 basal branches, 5 animals; 9 oblique branches, 5 animals; basal vs. oblique gain*, p = 2.17E5; ΔF/F*, p = 4.33E–5). Boxplot parameters as in Figure 2.

(D) Change in EPSP after plasticity induction at basal versus oblique dendrites at P14 (n = 5 basal branches, 5 animals; 5 oblique branches, 5 animals; basal vs. oblique postinduction*, p = 0.79), P21 (n = 5 basal branches, 4 animals; 7 oblique branches, 6 animals; basal vs. oblique postinduction*, p = 0.87), and P28 (n = 7 basal branches, 5 animals; 7 oblique branches, 4 animals; basal vs. oblique postinduction*, p = 1.25E–13). Error bars are SEM.

(E) P28 population gain (measured/expected, left) and branch ΔF/F (center) in dark-reared (DR) mice (n = 9 basal branches and 10 oblique branches from 4 animals; basal vs. oblique gain*, p = 0.78; ΔF/F*, p = 0.96). Synaptic plasticity for DR animals (right) (n = 5 basal branches, 4 animals; 6 oblique branches, 6 animals; basal vs. oblique postinduction*, p = 0.81). Error bars are SEM.

*Mann-Whitney U test was used for all comparisons.

See also Figure S3.

Oblique dendrites develop distinct postsynaptic receptor composition

One possible explanation of differential NMDAR properties between adult oblique and basal dendrites may be that synapses on adult oblique dendrites have a higher ratio of AMPA receptors (AMPARs) relative to NMDARs (AMPA/NMDA). Changes in AMPA/NMDA are known to occur over cortical development,^107,114,115 and adult cortical AMPARs are significantly less Ca²⁺-permeable than NMDARs.^116,117 To test this, we compared AMPAR- and NMDAR-mediated responses at single spines from basal and oblique dendrites of L5b PNs at P21 and in adults. We used glutamate uncaging to evoke AMPAR-mediated EPSPs at single spines with amplitudes and kinetics that mimic synaptic release events,^84,118–121 below the threshold for spine Ca²⁺ influx and NMDAR activation^15,71,122 (Figures 4A, 4B, and S4). Mg²⁺-free artificial cerebrospinal fluid (ACSF) containing the AMPAR antagonist 6,7-dinitroquinoxaline-2,3-dione (DNQX) was subsequently washed into the perfusate, and the uncaging protocol was repeated to assess synaptic NMDARs (Figure 4B). The average maximum amplitudes of EPSPs under each condition were compared as AMPA/NMDA. At postnatal day 21, when oblique dendrites behave similarly to basal dendrites, spines from the two branch types had comparable AMPA/NMDA. By adulthood, this ratio increased for oblique dendrites (Figures 4B and 4C), indicating relatively less NMDAR-mediated synaptic conductance and consistent with linear integration, smaller local Ca²⁺ signals and a decreased capacity for synaptic potentiation.

Figure 4. Changes in synaptic AMPA/NMDA at oblique versus basal dendrites underlie differences in integration and plasticity.

(A) Example two-photon image of a V1 L5b PN from a P21 mouse (scale bar: 50 μm). An oblique dendrite (circled) is shown at higher magnification (right, scale bar: 5 μm). Uncaging site indicated by star.

(B) Representative EPSPs from uncaging at single spines in P21 (left) or adult (right) mice from oblique (top) and basal (bottom) dendrites. Black traces: baseline (AMPAR-mediated response in regular ACSF). Colored traces: DNQX+Mg²⁺-free ACSF (NMDAR-mediated response).

(C) Functional single spine AMPA/NMDA from basal and oblique dendrites in P21 and adult mice. P21: n = 72 basal spines from 4 mice and 89 oblique spines from 4 mice. Adult, P56+: n = 66 basal spines from 4 mice and 67 oblique spines from 5 mice. Outliers are not shown. P21 basal vs. oblique, Mann-Whitney U test: p = 0.23; adult basal vs. oblique, Mann-Whitney U test: p = 1.00E–4. Violin plot lines: median, 25th, and 75th percentiles with ±2.7σ whiskers (dotted).

(D) Schematic of the eMAP experimental procedure.

(E) Confocal image of L5 PNs from a Thy1-GFP M P56+ mouse after ~4× eMAP expansion. Scale bar: 300 μm.

(F) Example confocal images of expanded oblique (top) and basal (bottom) dendrites from the same neuron. For each branch, immunostaining for bassoon, NR1, and GluA1+2 is shown for 1 spine (indicated by star). Scale bar: 10 μm, inset: 2 μm.

(G) AMPA/NMDA intensities for all spines (n = 1,305 spines in 20 basal dendrites and 1,719 spines in 20 oblique dendrites from 13 cells and 3 animals). Mann-Whitney U test: p = 8.30E−11. Violin plot parameters as in (C).

(H) AMPA/NMDA intensities for basal and oblique dendrites, averaged per cell (n = 13 cells, 3 animals). Wilcoxon signed rank test: p = 0.01 (**). Violin plot parameters as in (C).

See also Figure S4.

To provide complementary proteomic evidence for our observed changes in functional AMPA/NMDA, we acquired super-resolution images of oblique and basal dendrites from putative L5b PNs and their synaptic receptor content using epitope-preserving magnified analysis of the proteome (eMAP; Figure 4D).^123,124 Fixed slices from V1 of adult Thy1-GFP-M mice were expanded to approximately 4× and stained with antibodies against the presynaptic marker bassoon, the obligate NMDAR subunit GluN1, and GluA1 and GluA2, the predominant AMPAR subunits in cortex.¹²⁵ Oblique and basal dendrites from L5 PNs were imaged with a confocal microscope in ~40-μm-thick segments (~180 μm expanded), and dendritic protrusions were annotated (Figures 4E and 4F). Immature spines and protrusions without bassoon were excluded post hoc (although basal and oblique branches had similar numbers of immature and mature spines; Figure S4). For each spine head, the intensity of each fluorophore was spatially integrated, and the ratio of GluA1+GluA2/NR1 was taken. Consistent with our functional data, oblique dendrite synapses exhibited higher protein AMPA/NMDA than basal dendrites (Figures 4G, 4H, and S4). Protein AMPA and NMDA increased as a function of spine head area, as expected,^35,126–128 but oblique dendritic spines had consistently larger AMPA/NMDA for all spine head areas (Figure S4). Taken together, these results indicate that synapses on oblique dendrites in adults have relatively less NMDAR protein content and conductance, and this underlies the differences in integration and plasticity rules between basal and oblique dendritic compartments.

In vivo evidence of compartmentalized synaptic stability and flexibility in single neurons

Variable levels of synaptic Ca²⁺ from NMDARs can engage structural plasticity at dendritic spines. The formation, enlargement, shrinkage, and/or elimination of spiny synapses has been shown to occur as a result of NMDAR activation across a range of synaptic plasticity paradigms.^{36,127,129–135} The tight correlation between functional plasticity and structural plasticity supports dendritic spine imaging as a readout of experience-dependent plasticity in vivo.^{35,37,38,42,136,137} Structural plasticity occurs in control conditions as a reflection of the ongoing synaptic plasticity necessary for normal brain function, where the amount of baseline structural plasticity varies by age, brain region, and cell type, and also correlates with functional plasticity.^{38,39,138–141} Dendritic spine formation and elimination (i.e., the gain or loss of synaptic connections) are thought to represent the acquisition and refinement of new information.^136,142,143 In contrast, some dendritic spines remain stable for significant periods of time, leading to the idea that stable spines are critical for information storage.^36,40,144 Given that we observed relatively low NMDAR-mediated properties in adult oblique dendrites, we hypothesized that these dendrites would have largely stable spines in vivo, whereas spines on basal dendrites should exhibit more dynamic spine elimination and formation, reflecting the substantial capacity for functional plasticity we observed in brain slices. To investigate this, we longitudinally imaged dendritic spines in putative L5b PNs expressing EGFP in adult mice. With extremely sparse cellular labeling and an optimized two-photon microscopy setup (Figure 5A), we were able to resolve spines in oblique and basal branches, 300–620 μm below the pia (Figures 5B and S5). We tracked spine dynamics for both basal and oblique dendritic compartments in individual cells across 5–10 consecutive days (n = 6 cells, 775 basal spines and 794 oblique spines, 4 animals). Spine density did not significantly differ between basal and oblique compartments, consistent with our observations with eMAP (Figures 5C and S5). The overall survival fraction of basal and oblique spines (or persistent, stable spines) was similar to what was previously reported for V1 apical tuft spines over a similar period of time,^39,41 but spines on oblique branches had significantly higher survival rates (Figure S5). When considering both the formation and elimination of all spines between days (daily turnover ratio), oblique spines were remarkably stable, showing less than half the average turnover found on basal dendrites in the same cell (Figure 5D). When individual spines were categorized by lifetime, where less than 4 days generally represents a “transient” synapse,^{38,41,143,145,146} a larger fraction of basal dendritic spines was more transient than oblique spines (Figures 5B and 5E). The relative stability we observed for spines on thalamorecipient oblique dendrites aligns with previous findings indicating that thalamocortical boutons are highly stable compared to intracortical boutons in adult neocortex.^44,147,148 Our results demonstrate that L5 PNs in adult mice possess a dendritic domain with highly stable synapses, in contrast to other dendritic domains with higher rates of synaptic formation and elimination.

Figure 5. In vivo synaptic stability and flexibility at distinct dendritic compartments.

(A) Three-dimensional projection of an EGFP-expressing L5b PN, imaged in vivo with two-photon microscopy. Scale bar: 50 μm.

(B) Representative oblique (top) and basal (bottom) dendrites from the same L5 PN imaged across 7 days. All transient spines (with lifetimes of ≤4 days) are labeled (arrowheads) on the day prior to disappearance. Scale bars: 5 μm.

(C) Average spine density per cell for basal and oblique dendrites*. Wilcoxon signed rank, p = 0.56.

(D) Average daily turnover ratio, per cell*. Wilcoxon signed rank, p = 0.03 (*).

(E) Fraction of transient (≤4-day lifetime) and persistent (≥5-day lifetime) spines on oblique (purple) and basal (green) dendrites, per cell*. Friedman’s test, p = 1.07E–6.

For (B)–(E), n = 6 cells with 775 basal spines and 794 oblique spines from 4 animals.

*Boxplot parameters as in Figure 2.

See also Figure S5.

DISCUSSION

The plasticity of spiny synapses is widely considered to be the biological substrate of learning and memory. The current model posits that transient dendritic protrusions are involved in the acquisition and refinement of information, whereas stable dendritic spines store acquired information.^{36,38,39,128,140,143,146} The long-term stability of specific spiny synapses is not well understood and likely involves a combination of mechanisms, from network-level properties to molecular regulation. Here, we show that entire dendritic domains can be highly stable and resistant to at least one type of long-term potentiation through a relative downregulation of synaptic NMDARs. These dendrites have comparable spine densities and morphologies to adjacent, plastic dendritic compartments but lack NMDAR-mediated supralinear integration and potentiation. Thus, dendrites provide compartmentalization to allow synaptic plasticity and stability to operate within the same cell.

Our results do not rule out all possible plasticity mechanisms in adult oblique dendrites. However, limited NMDAR-mediated Ca²⁺ influx constrains their potential to engage conventional molecular cascades.⁹⁶ We identified dLGN as a distinct input to the oblique dendritic domain, and by manipulating thalamic activity during the visual critical period, we could prevent the decline of NMDAR-mediated properties, including synaptic plasticity. Thus, critical periods—the developmental window of experience required for normal connectivity and function^110,113,149—occur at the subcellular level, an unprecedented degree of specificity, due to dendritic compartmentalization. Although we have focused on one input to oblique dendrites, our uncaging and synaptic plasticity experiments were agnostic to the type of synaptic input, indicating that other inputs to oblique dendrites would also be unlikely to engage NMDAR-dependent properties after maturation. Notably, NMDARs are still expressed at synapses on oblique dendrites in adults, just at relatively lower density, and other synaptic mechanisms may be involved in the stability of these synapses. Our findings establish a tractable model for future studies on the formation and regulation of stable dendritic domains, which likely includes activity-dependent transcription, protein expression, and subcellular localization.^150–154

Protecting existing knowledge while continually acquiring new information is a long-standing challenge for models of plasticity and biologically inspired neural networks, which are prone to instability and catastrophic forgetting.^1,5 To resolve this plasticity-stability dilemma, some models decrease the plasticity of important weights.^155–157 Here, we report biological evidence for these theoretical limitations at the subcellular level. After an initial strengthening in early development, synapses on oblique dendritic spines are effectively stabilized through a domain-wide relative decline in NMDAR expression and function, while synapses on basal dendrites remain plastic. Our results are distinct from, but may interact with, other models and observations that work to stabilize either plastic networks or individual synapses.^21,158–160 In primary visual cortex, the stable dendritic domain receives input from the visual thalamus, the definitive input to primary visual cortex. PNs in all layers of primary sensory cortices receive both first-order thalamic and intracortical inputs,^{73,75,161,162} and this mechanism may help sensory neurons maintain fundamental sensory representations throughout a lifetime of learning and experience-dependent plasticity. Our results may also apply to other non-sensory cortical areas; the oblique dendritic domain in mouse retrosplenial cortex similarly lacks NMDAR-mediated supralinear integration.⁷¹ Using this mechanism, cortical neurons may limit plasticity within specific dendrites to maintain essential, stable representations without compromising higher-order input plasticity. The restriction of plasticity within select dendritic compartments is a powerful mechanism by which single neurons can solve the trade-off between flexibility and stability inherent to all learning systems.

Limitations of the study

The diffraction limit of two-photon microscopy prevents the accurate detection of small dendritic protrusions, including filopodia, in vivo. Although our eMAP data show that small dendritic spines on oblique dendrites have elevated AMPA/NMDA compared to those on basal dendrites (Figure S4), the in vivo structural dynamics of small spines and filopodia are unknown. However, super-resolution imaging shows that filopodia are present on basal and oblique dendrites at approximately equal densities (Figure S4), and these silent synapses are capable of rapidly forming new connections.¹²⁸ Whether and how oblique dendrites leverage silent synapses to integrate new inputs remain open questions.

Relatedly, while baseline structural plasticity is correlated with functional plasticity,^{38,39,138–141} the stability of oblique dendrite synapses should be tested across a broader range of experiences, such as during visual learning or after loss of vision. Although here we have focused on inputs to oblique dendrites from dLGN, other synaptic inputs should be identified to better interpret which inputs are linearly integrated and stabilized. Future research focused on overcoming the limited plasticity in adult oblique dendrites and specifically promoting the plasticity of dLGN inputs may be crucial for the recovery of vision in cases of injury or neural degeneration.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dr. Mark T. Harnett (harnett@mit.edu).

Materials availability

This study did not generate any new materials or unique reagents.

Data and code availability

• All data are available upon request from the lead contact.

• All original analysis code has been deposited at Zenodo (DOI listed in the key resources table) and is publicly available as of the date of publication.

• Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-GFP	Life Technologies	Cat# A10262; RRID: AB_2534023
Anti-NMDAR1	Synaptic Systems	Cat# 114011; RRID: AB_887750
Anti-AMPAR1	Synaptic Systems	Cat# 182003; RRID: AB_2113441
Anti-AMPAR2	Synaptic Systems	Cat# 182103; RRID: AB_2113732
Anti-Bassoon	Synaptic Systems	Cat# 141004; RRID: AB_2290619
anti-Guinea pig-405	AbCam	Cat # ab175678; RRID: AB_2827755
anti-Chicken-488	Invitrogen	Cat# A11039; RRID: AB_142924
anti-Mouse-AF+555	Invitrogen	Cat # A32727; RRID: AB_2633276
anti-Rabbit-AF+647	Invitrogen	Cat# A32733; RRID: AB_2633282
Bacterial and virus strains
AAV2-hSyn-hChR2(H134R)-mCherry	Gift from Karl Deisseroth	Addgene plasmid # 26976; RRID: Addgene_26976
pENN.AAV.CamKII 0.4.Cre.SV40	Gift from James M. Wilson	Addgene viral prep # 105558-AAV1; RRID: Addgene_105558
pAAV-hSyn-DIO-EGFP	Gift from Bryan Roth	Addgene viral prep # 50457-AAV1; RRID: Addgene_50457
Chemicals, peptides, and recombinant proteins
TTX	Tocris	1078
4-AP	Sigma-Aldrich	A78403
MNI-caged-L-glutamate	Tocris	1490
Alexa 594	Invitrogen	A10438
Alexa 488	Invitrogen	A10436
OGB-1	Thermo Fisher	O6806
D-APV	Tocris	0106
DNQX disodium salt	Tocris	2312
DAPI	Thermo Fisher	62248
Picrotoxin	Tocris	1128
Acrylamide	MilliporeSigma	A9099
Sodium acrylate	MilliporeSigma	408220
Bis-acrylamide	Bio-Rad Laboratories	161–0142
VA-044	Wako Chemicals	NC0632395
Sodium dodecyl sulfate	Sigma-Aldrich	L3771
Sodium sulfite	Sigma-Aldrich	S0505
Experimental models: Organisms/strains
Mouse: C57BL/6	Charles River	Cat# 027; RRID:IMSR_CRL:027
Mouse: Thy-1-GFP-M	Jackson Labs	Cat# 007788; RRID:IMSR_JAX:007788
Software and algorithms
MATLAB (R2019a)	MATLAB	RRID:SCR_001622;
ImageJ (v1.54f)	National Institutes of Health	RRID:SCR_003070; https://imagej.nih.gov/ij/index.html
Custom code for data analysis	This paper	Zenodo: https://zenodo.org/doi/10.5281/zenodo.12791375
Other
Ultima In Vitro Multiphoton Microscope System	Bruker	RRID: SCR_017142
MaiTai DeepSee	Spectra-Physics	MAI TAI HP DS
Electro-optical modulator	Conoptics	M350–50
Photosensor module	Hamamatsu	H7422A-40
Collimated LED for Olympus BX	ThorLabs	M625L4-C1
473 nm laser	OptoEngine LLC	MSL-FN-473/1~100mW
Dagan BVC-700A	Dagan Corporation	N/A
Fully automated vibrating blade microtome	Leica	Cat #VT1200S; RRID: SCR_018453
TCS SP8 upright confocal microscope	Leica	DM6000
Intracellular Microinjection Dispense System	Parker Hannifin	Picospritzer III; RRID: SCR_018152
Flexible Stimulus Isolator Unit	A.M.P.I.	ISO-Flex; RRID: SCR_018945
Thorlabs Bergamo® II Multiphoton	Thorlabs	N/A
Ti:Sapphire laser	Coherent	Chameleon Ultra II

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

All animal procedures were carried out in accordance with NIH guidelines for animal research and were approved by the Massachusetts Institute of Technology Committee on Animal Care and the Animal Care and Use Committee at the University of California, Berkeley. C57BL/6 male and female mice (Charles River Laboratories) were used in approximately equal numbers. All animals were kept in conventional social housing with unlimited food and water on a 12-h light/dark cycle. Mice 8 weeks and older (P56+) were used for adult mouse experiments. For developmental timepoints, mice ages P10-P14, P18–22, and P28-P32 were used. For dark-rearing experiments, pregnant dams were moved to a dark room at E18 and pups were born and raised in complete darkness, with red light exposure during cage changes. Dark-reared mice were ages P28-P32 at the time of experimentation. Dark-reared animals that underwent subsequent light exposure were returned to normal housing conditions at P28, and experiments were conducted at either P42 or P56. 8-week-old Thy-1-GFP-M mice (Jackson Laboratory) were used for eMAP experiments. For chronic imaging experiments, 2 male and 2 female C57BL/6 mice were approximately 8 weeks of age at the time of injection and cranial window surgery.

METHOD DETAILS

Slice preparation

Sucrose-containing artificial cerebrospinal fluid (sACSF) was used during the slicing procedure, containing (in mM): 90 sucrose, 61 sodium chloride, 26.2 sodium bicarbonate, 2.75 potassium chloride, 1.25 sodium phosphate, 9 glucose, 1.1 calcium chloride, and 4.1 magnesium chloride, with an osmolality of 295–302. The sucrose solution was partially frozen to create a small amount of slush and was kept ice-cold. Artificial cerebrospinal fluid (ACSF) was used for recovery and recording, containing (in mM): 122 sodium chloride, 24.5 sodium bicarbonate, 3 potassium chloride, 1.25 sodium phosphate, 11.25 glucose, 1.2 calcium chloride, 1.2 magnesium chloride, 1 ascorbate, and 3 sodium pyruvate, with an osmolality of 302–307. All solutions were saturated with carbogen, 5% CO₂ and 95% O₂. Acute slice preparation was consistent for all ages and experiments. Mice were put under isoflurane-induced anesthesia and decapitated. The brain was extracted in ice-cold sucrose solution in less than 1 min. The brain was blocked at a moderate angle (approximately 20° from coronal) to favor the preservation of apical dendrites in the occipital cortex. After the brain was mounted and submerged in ice-cold sACSF, a vibratome (Lieca VT1200S) was used to cut 300 μm-thick slices, which were transferred to ACSF for 30–50 min at 36°C. Following recovery, slices were kept at room temperature.

Patch-clamp recording

Recordings of V1 L5b PNs were performed in ACSF (concentrations noted above) at 34–36°C. Intracellular recording solution contained (in mM): 134 potassium gluconate, 6 KCl, 10 HEPES, 4 NaCl, 4 Mg₂ATP, 0.3 NaGTP, and 14 phosphocreatine di(tris). Depending on the experiment, 0.05 mM Alexa 594, 0.1 mM Alexa 488, and/or 0.1 mM OGB-1 (Invitrogen) were added to the internal solution. An Olympus BX-61 microscope with Dodt optics and a water-immersion lens (60X, 0.9 NA; Olympus) was used to visualize cells. Whole-cell current-clamp recordings were obtained with a Dagan BVC-700A amplifier. Patch pipettes with thin-wall glass (1.5/1.0mm OD/ID, WPI) and resistances of 3–7 MΩ were used. Pipette capacitance was fully neutralized prior to break-in, and series resistance was kept fully balanced throughout, ranging from 5 to 30 MΩ. The liquid junction potential was not corrected. Current signals were digitized at 20 kHz and filtered at 10 kHz L5b PNs were characterized by their large somas in layer 5, thick apical dendrites, broad arborization in layer 1, low input resistance, and prominent voltage sag.

Stereotaxic surgery and virus expression for sCRACM

For sCRACM experiments, AAV2-hSyn-hChR2(H134R)-mCherry (UNC Vector Core) was expressed in neurons in the dorsal lateral geniculate nucleus. Mice 6 weeks or older were injected 4 weeks or more before slice preparation. Under isoflurane anesthesia, mice were secured on a stereotaxic apparatus with a feedback-controlled heating pad (DC Temperature Control System, FHC). Slow-release buprenorphine (1mg/kg) was administered subcutaneously. After scalp incision, a small burr hole was drilled over the injection site. Bilateral injections were made for dLGN at the following coordinates relative to bregma: anterior-posterior 2.7–2.8 mm; medio-lateral 2.3–2.4 mm. A beveled microinjection pipette containing was lowered to a depth of 2.7–2.9 mm, and approximately 100 nL of virus was injected at a slow rate of 50 nL/min using a Nanoject. Following virus injection and a 5-min rest, the pipette was removed slowly and the incision was sutured. Accuracy of the injection was assessed in acute slices using two-photon microscopy, and animals with expression outside of dLGN were not used. For image clarity, histology in Figure 1 was taken from a mouse that received a stereotaxic injection at identical coordinates but was not used for slice physiology: instead the brain was perfused, stored at 4°C overnight in 4% paraformaldehyde, transferred to PBS, and sectioned in 100 μm-thick slices. Sections containing the injection siteand V1 were immunolabeled with 1:1000 DAPI solution for 5 min. Sections were mounted and coverslipped with clear-mount with tris buffer (17985–12; Electron Microscopy Sciences) and were imaged under a confocal microscope (Zeiss LSM 710 with a 10x objective, NA 0.45).

sCRACM

After 4 weeks or more of ChR2 expression in dLGN, acute slices of V1 were prepared. A two-photon laser scanning system (Bruker) with dual galvanometers and a Mai Tai DeepSee laser were used to confirm the presence of mCherry-expressing cell bodies and axons in dLGN under low magnification (4x, 0.16 NA air objective, Olympus). Animals with expression outside dLGN were not used. For all recordings, TTX (1 μM) and 4-AP (100 μM) were added to ACSF to isolate monosynaptic connections. Following whole-cell configuration, a 5 ms 473 nm full-field LED was used to determine if the neuron received inputs from axons containing ChR2. If the LED drove EPSPs or spikes, a 9 by 17 (400 × 800 μm) stimulation grid with 50 μm spacing was positioned over the area containing the entire neuron, aligned at the pia using two-photon imaging under low magnification (Prairie View software). The stimulation grid controlled the location of a 473 nm laser beam (OptoEngine, LLC), and the duration of the light pulses was kept to 1–3 ms under <2 mW power. Each point on the grid was stimulated in order, progressing forward by column, with an interpoint delay of 1 s. 3–6 rounds of stimulation were averaged for each neuron. Baseline voltage was defined as the 50 ms before stimulation, and EPSP amplitudes were calculated as the baseline-subtracted maximum voltage within 50 ms after photostimulation. Population data were obtained by rotating the maps to align to the pia, overlaying the maps to align at the soma, normalizing peak voltages, and averaging across experiments.

Glutamate uncaging: Input-output and single spine AMPA/NMDA

Following whole-cell dialysis with a structural dye and a Ca²⁺ indicator via the patch pipette, basal or apical oblique branches were localized using the same two-photon microscopy set up described above, with the addition of a second Mai-Tai DeepSee laser used for photolyzing 4-methoxy-7-nitroindolinyl-caged-L-glutamate (MNI-glutamate, Tocris) at 720 nm. This laser path included a passive 8x pulse splitter, which drastically reduces photodamage.¹⁶³ A pipette containing MNI-glutamate diluted in ACSF (10 mM) was positioned just above the slice and over the recording site, and a picrospritzer (Picospritzer III, Parker Hannifin) was used to puff a constant flow of MNI-glutamate into the region of interest. Using Prairie View software, uncaging was targeted just adjacent to spine heads, and in the imaging pathway, a linescan was used for simultaneous calcium imaging. Laser intensity for both lasers was independently controlled (Conoptics). Linescan imaging was performed at 1300 Hz, with a dwell time of 8 μs and a total scan time of less than 250 ms, with baseline fluorescence kept minimal and monitored throughout. Basal branches (extending from the soma) and oblique branches (extending from the apical trunk, at least 30 μm from the soma) were identified. Cells with signs of photodamage were excluded, including decreases or loss of Ca²⁺ signal in response to uncaging or current injection, persistent depolarization of the resting membrane potential, changes in voltage response kinetics during uncaging or current injection, or visible bleaching or blebbing of dendritic morphology.

For input-output experiments (integration of groups of spines) with Ca²⁺ imaging, branches with approximately 30 dendritic spines in the focal plane were identified. The uncaging laser was calibrated to either a threshold calcium signal (~100% ΔF/F) or an action potential, where the minimum power needed to drive either event was identified and used for all points. The uncaging dwell time was 0.2 ms, and the uncaging interval between multiple spines was 0.32 ms. Groups of 2–5 spines were stimulated independently to avoid saturating the synapses, followed by combination with other groups up to a total of 25–40 spines. Expected values were calculated by summing the average response of each group of spines. Ca²⁺ signals are expressed as (F − F_baseline)/F_baseline). For acute blockade of NMDA receptors, the competitive antagonist D-AP5 (50 μM) was used.

For single spine estimates of AMPA/NMDA, no Ca²⁺ indicator was included in the intracellular solution, and MNI-glutamate was diluted in Mg²⁺-free ACSF. Approximately 10–20 spines were individually stimulated at each branch. The uncaging stimulus was delivered in each spine separately, with an inter-stimulation interval of 500 ms. Care was taken to calibrate the uncaging laser power to evoke unitary EPSPs that mimic spontaneous miniature EPSPs by uncaging at approximately 0.5 μm from the spine head throughout the experiment: this method evokes physiologically-sized EPSPs with minimal NMDA-mediated Ca²⁺ in the spine head¹⁵ (see Figure S4). Unitary EPSPs were evoked 15–20 times and the responses were averaged. Following these recordings, Mg²⁺-free ACSF containing 20 μM DNQX was washed in for 10 min, and the same uncaging protocol was repeated at the same spines. The ratio of the average maximum EPSP amplitude between both conditions was taken for each spine.

Burst-timing dependent plasticity

All plasticity experiments were performed with an extracellular calcium concentration of 1.2 mM. For local branch stimulation, theta glass (2.0/1.4mm OD/ID) housing a bipolar stimulating electrode (ISO-Flex, AMPI) was positioned within 10 μm of a basal or apical oblique branch. Stimulation intensity was calibrated to generate a small EPSP of 1–3 mV, typically requiring 5–20 μA of current from the stimulating electrode. Ten minutes of baseline stimulation at 0.1 Hz was recorded to ensure stability of the EPSP. A −50 pA hyperpolarizing step was included to estimate input resistance. For plasticity induction, EPSPs were paired with a somatic current injection of 400–700 pA for 20 ms duration to produce a train of 2–4 action potentials. The peak of the first action potential was timed to be within 10 ms of the peak of the EPSP. Pairs were done in sets of 5 at 10 Hz (theta rhythm), and sets of 5 were repeated 30 times at 0.1 Hz. After the induction period, EPSPs were monitored for at least 20 min. Inclusion criteria were as follows: resting membrane potential could fluctuate from baseline by no more than 3 mV, input resistance could not increase more than 20% of baseline; and series resistance had to be compensated fully throughout (<30 MΩ). Recordings were tested for baseline stability post-hoc by calculating Spearman’s correlation coefficient (Figure S2), and cells with significant changes in baseline were not included. The necessity of pre-post pairing was shown using the same recording set up but driving either the synaptic stimulation or the post-synaptic burst (and not both) during the induction period. For acute blockade of NMDARs, the competitive antagonist D-AP5 (50 μM) was used. Throughout plasticity experiments, picrotoxin (100 μM) was present in the bath to block GABAergic transmission in acute slices from adult mice and at P28. 10 μM picrotoxin was used for acute slices at P21, and no picrotoxin was needed at P14.

Epitope-preserving Magnified Analysis of the Proteome

Adult Thy-1-GFP-M were perfused with cold PBS followed by cold 4% PFA while under deep anesthesia (5% isoflurane). Brains were removed and kept in the same fixative overnight at 4°C and then washed with PBS at 4°C for at least 1 day. 1.0 mm coronal slices of primary visual cortex were cut on a vibratome and kept in PBS at 4C until the day of processing. Slices were then incubated eMAP hydrogel monomer solution (30% acrylamide [A9099, MilliporeSigma, St. Louis, MO, USA], 10% sodium acrylate [408220, MilliporeSigma], 0.1% bis-acrylamide [161–0142, Bio-Rad Laboratories, Hercules, CA, USA], and 0.03% VA-044 (w/v) [Wako Chemicals, Richmond, VA, USA] in PBS]), protected from light, at 4°C overnight. For gelation, slices were mounted between glass slides in eMAP solution and sealed in a 50 mL conical tube with nitrogen gas at positive pressure of 10–12 psi at 37°C for 3 h. The excess gel around the slices was then removed. To reach a first expansion stage of 1.7X, the slices were incubated overnight in a solution of 0.02% sodium azide (w/v) in PBS at 37°C. Slices were trimmed to contain only parts of primary visual cortex and further sectioned with a vibratome to 75 μm thickness (corresponding to ~40 μm thickness of the pre-expanded tissue). Slices containing good candidate cells – L5 PNs whose apical trunk could be reconstructed at its full length in a single slice or at most two consecutive slices – were selected during live low-resolution confocal imaging sessions. These slices were trimmed to smallest possible samples of approximately 1.0 mm in both width and length. Slices were incubated in tissue clearing solution (6% SDS (w/v), 0.1 M phosphate buffer, 50 mM sodium sulfite, 0.02% sodium azide (w/v), pH 7.4) at 37°C for 6 h, followed by incubation in preheated clearing solution at 95°C for 10 min. Cleared samples were thoroughly washed with PBS +0.1% Triton X- at 37°C. Primary antibody staining was performed at 37°C overnight with the following antibodies: Anti-GFP (Life Technologies A10262), Anti-NMDAR1 (SYSY 114011), Anti-AMPAR1 (SYSY 182003), Anti-AMPAR2 (SYSY 182103), and Anti-Bassoon (SYSY 141004). For secondary staining, the following fluorescent antibodies were used: Bassoon: anti-Guinea pig-405 (AbCam ab175678); GFP: anti-Chicken-488 (Invitrogen A11039); NMDAR1: anti-Mouse-AF+555 (Invitrogen A32727); and AMPAR1 and AMPAR2: anti-Rabbit-AF+647 (Invitrogen A32733). Final expansion was performed just before imaging by putting the trimmed slices in 0.1 mM tris in distilled water, and approximately 4X total linear expansion was achieved. Slices were imaged using Leica TCS SP8 upright confocal DM6000 microscope equipped with a 63x HC PL APO CS2/1.2 W objective, hybrid detectors, and a white light laser. In order to be as consistent as possible with our physiological data, we took measurements from putative L5b PNs, identified by their thick trunks and broad apical tufts. Within single neurons, both basal and oblique dendritic branches were imaged. To avoid photobleaching effects, no slice was imaged more than once, and basal and oblique dendrites were imaged in alternating order for each cell. 13 cells from 4 animals were imaged, for a total of 20 basal dendrites and 20 oblique dendrites. Each dendritic branch was imaged in 76 × 38 μm segments, totaling 64 basal dendrite segments and 69 oblique dendrite segments. Dendritic protrusions were analyzed using Fiji software. To draw regions of interest within spine heads in the GFP channel, a custom-written macro code was used to apply a median blur (2 pixels), threshold the image, and draw an ROI on the annotated protrusion. Using custom-written code in MATLAB, all color channels were thresholded to include only signals 2 S.D. above the mean fluorescence intensity, and signals from each channel were extracted from the ROI. The plane with peak antibody fluorescence was identified within the spine head ROI, and each channel’s fluorescence signals within the plane were summed. Long, thin dendritic protrusions without enlarged heads were classified as filopodia (head diameter/neck diameter < 1.3¹²⁸) and excluded from the final analysis, as were spines with no detectable bassoon.

Ultrasparse eGFP expression in L5 PNs for deep layer structural imaging

Prior to surgery, mice were subcutaneously injected with dexamethasone (4 mg kg⁻¹) and buprenorphine (slow-release, 1 mg kg⁻¹) and were anesthetized with isoflurane (2% induction, 0.75–1.25% for maintenance). Mice were secured in a stereotaxic apparatus with a feedback-controlled heating pad. Eye ointment was applied to prevent dryness (Bepanthen, Bayer), and the scalp was shaved using hair removal cream and cleaned with iodine and ethanol. The skull was exposed, etched, and sealed with a light-curing glue (Adhese Universal, Ivoclar). A stainless steel headplate was secured to the skull with a light-curing flowable composite (Tetric EvoFlow Translucent, Ivoclar). Then a 3 mm-wide craniotomy was performed over V1 (2.5 mm lateral, 1 mm anterior to lambda). A 1:2 viral mixture of pENN.AAV.CamKII 0.4.Cre.SV40 (Addgene, diluted 1:100,000) and pAAV-hSyn-DIO-EGFP (Addgene) was injected at a volume of 150 nL at 3 sites, approximately 200–500 μm apart, at a depth of 550 μm. Following virus injection, the pipette was withdrawn after 5 min to avoid infecting superficial layers. Major blood vessels were avoided. Cranial windows consisted of a single 3 mm coverslip attached to a 4.5 mm × 2.5 mm coverslip (170 μm thickness, custom-made, Potomac). The outer ring of the cranial window was fixed to the skull with light-curing flowable composite (Tetric EvoFlow Translucent, Ivoclar). Following 1–2 weeks of recovery from surgery, animals were transferred to UC Berkeley and had 2 additional weeks of recovery.

Longitudinal spine tracking in deep cortical layers with two-photon microscopy

Ultrasparse eGFP expression (1–2 cells per injection site) allowed us to visualize L5 PN dendritic spines 300–620 μm below the cortical surface with an optimized commercial two-photon microscope (Thorlabs Bergamo II Multiphoton). A femtosecond Ti:Sapphire laser (Coherent, Chameleon Ultra II) tuned to 920 nm was used as the excitation source for all experiments. A 25× water-immersion objective lens (Olympus, 1.05 NA, 2 mm WD) was used to focus the excitation light into the brain and collect the emitted fluorescence. Data acquisition and hardware were controlled using ThorImage software. Imaging started approximately 4 weeks after virus injection. Immediately prior to imaging, the mouse was anesthetized (2% isoflurane) and then head-fixed under the microscope with continuous anesthesia (0.5–1% isoflurane). Body temperature was maintained using hand warmers. The headmount of the mouse was aligned so that its cranial window was perpendicular to the excitation beam. The correction collar of the objective lens was adjusted to minimize spherical aberration induced by the cranial window. Putative L5b PNs were identified by cortical depth, with somas between 500 and 600 μm below the pia, and by their thick apical dendrites and broad arborization in layer 1. For each cell, z stacks were taken from the top-most oblique dendrite to the bottom-most basal dendrite, ranging in a total imaging volume of 140–280 μm (1 μm z steps, 203 μm × 203 μm field of view, 0.1 μm x/y pixel⁻¹, 4 fps, 20 frames per step). The same cells and branches were imaged daily for 5–10 days. Post-objective power ranged from 38 to 65 mW for different cells, and care was taken to ensure that fluorescence levels were comparable throughout the tracking period.

Image processing and spine tracking analysis

Two-photon z stacks were analyzed in ImageJ (v1.54f). Smaller stacks of specific basal and oblique branches were sectioned from the original z stack. Branches with steep changes in z were avoided. Each substack was rigidly aligned, gamma adjusted (1.5), and Gaussian filtered (σ = 1). Branches from consecutive days were manually registered with a transparency tool (Vitrite), and similar lookup tables were used for each timepoint. Branches were only included if the dimmest protrusion was at least 5x higher than the standard deviation of the background around the dendrite. Dendritic protrusions were scored on maximum intensity z-projections and cross-checked with the original z stack. The following criteria was used for scoring: protrusions must extend laterally, more than 0.5 μm from the branch. Protrusions that did not reach criteria at any point during the experiment were excluded from analysis altogether. Daily tracking facilitated consistent identification of structures, even with small movements of the tissue. With this dataset, we calculated the fraction of protrusions that were present across the entire imaging period (survival fraction, SF) and the turnover ratio (TOR), defined as the fraction of structures that appear and/or disappear between days, as in.¹⁶⁴ SF(t) = N(t)/N₀, where N₀ is the number of structures at t = 0, and N(t) is the number of structures of the original set surviving after time t. TOR (t₁, t₂) = (N_gained+ N_lost)/(N(t₁) + N(t₂)), where N(t₁) and N(t₂) are the total number of structures at the first and second time point. The lifetime of a given structure is the number of days present within the recording window.

QUANTIFICATION AND STATISTICAL ANALYSIS

We performed statistical analyses using MATLAB, and non-parametric tests were used for all datasets. The Mann-Whitney U test (also known as the Wilcoxon rank-sum test) was used to test two independent groups, and paired comparisons between two groups were tested with the Wilcoxon signed-rank test. Friedman’s test was used for paired comparisons with more than two groups. Outliers were included in all statistical analyses. Boxplots and violin plot lines indicate the median, 25^th, and 75^th percentiles with +/−2.7σ whiskers. Statistical significance noted in figures corresponds to: p ≥ 0.05 (nonsignificant, N.S), p < 0.05 (*), and p < 0.01 (**).

Highlights.

• Highly stable synapses are found in oblique dendrites of L5 PNs in adult mouse V1

• Stable synapses receive specific inputs, integrate linearly, and do not strengthen

• Oblique dendrite synapses have higher AMPA/NMDA across all spine sizes

• Stable synapse properties develop after a critical period of visual experience