Cell-Autonomous Regulation of Dendritic Spine Density by PirB

Visual Abstract


Introduction
Cortical circuits are altered by experience throughout life and undergo extensive restructuring during early developmental critical periods. Underlying these experience-dependent circuit changes are cellular and molecular mechanisms of synaptic plasticity. Different learning and plasticity paradigms involving specific cortical regions result in a persistent increase in the density of dendritic spines, which are postsynaptic anatomical structures at excitatory synapses and represent sites of plasticity. This increase in dendritic spine density is thought to represent a structural trace of new learning. For example, mice trained on a forepaw reaching task show an increase in dendritic spine density on apical dendrites of L5 pyramidal cells in motor cortex (Xu et al., 2009b;Fu et al., 2012); changes are also seen in spine density on the dendrites of L2/3 pyramidal cells (Ma et al., 2016). When these newly formed spines are selectively disassembled, motor memories are erased (Hayashi-Takagi et al., 2015). In the binocular zone of mouse visual cortex, closure of one eye (monocular deprivation) generates an experiencedependent form of plasticity known as ocular dominance (OD) plasticity (Gordon and Stryker, 1996). This plasticity is accompanied by an enduring increase in spine density along the apical tufts of L5 pyramidal neurons (Hofer et al., 2009;Djurisic et al., 2013). As in motor cortex, this net increase in density is thought to provide a structural substrate that mediates a lower threshold for OD plasticity when the same eye is closed again later in life (Hofer et al., 2006(Hofer et al., , 2009. In recent years, a list of molecules that appear to function normally as negative regulators of visual cortical plasticity has emerged, in the sense that gene knockout enhances OD plasticity following brief monocular deprivation. Blockade, removal, or genetic deletion of each of these molecules, including Nogo Receptor 1 (NgR1; Mc-Gee et al., 2005;Frantz et al., 2016), Lynx1 (Morishita et al., 2010;Bukhari et al., 2015), Death Receptor 6 (DR6; Marik et al., 2013), chondroitin sulfate proteoglycans (Pizzorusso et al., 2002), or paired Ig-like receptor B (PirB; Syken et al., 2006;Djurisic et al., 2013;Bochner et al., 2014), generates enhanced OD plasticity even in adult mice, well after normal closure of the critical period, and when significant OD plasticity cannot be elicited in wildtype mice. In addition, cell type-specific deletion of PirB from excitatory pyramidal neurons is sufficient to generate enhanced OD plasticity in adult visual cortex (Bochner et al. 2014). In mice with germline deletion (PirB Ϫ/Ϫ ), dendritic spine density is elevated not only during the critical period but also in adults, implying a developmental pruning defect . This elevation in spine density is thought to mediate the juvenile-like OD plasticity observed in adult visual cortex of these mice.
PirB is expressed in cortical pyramidal neurons. It was discovered in an in situ hybridization screen designed to identify receptors expressed in brain that bind MHC class I molecules, which are involved in activity-dependent plasticity (Syken et al. 2006;Adelson et al., 2012;Djurisic et al., 2013;Bochner et al., 2014;Adelson et al., 2016). The elevated spine density and enhanced OD plasticity in visual cortex of PirB Ϫ/Ϫ mice could arise from a requirement for PirB function exclusively in neurons. Microglia are also intimately involved in synapse pruning (Schafer et al., 2012;Parkhurst et al., 2013), but PirB expression has not been detected in this cell type in vivo. Nevertheless, studies of germline PirB knock-out mice, in which both brain and immune systems are affected, cannot distinguish between these alternatives. Here we use timed in utero electroporation of Cre recombinase into developing PirB fl/fl mouse ventricular zone (Saito and Nakatsuji, 2001) to investigate whether neuron-specific deletion of PirB is sufficient to explain changes in dendritic spine density seen in PirB Ϫ/Ϫ mice  or following pharmacological blockade of PirB in adult wildtype visual cortex (Bochner et al. 2014). In addition, in utero electroporation permits sparse deletion of PirB in single L2/3 neurons embedded in a wild-type environment. This type of mosaic approach has been used in vivo to determine whether a particular gene function is cell autonomous (Hippenmeyer et al., 2010;Akbik et al., 2013;Lu et al., 2013). Here, sparse deletion of PirB in L2/3 pyramidal neurons demonstrates that neuronal PirB is required for the regulation of synaptic density, leading us to conclude that this function of PirB is cell intrinsic.

Mice
PirB fl/fl , PirB ϩ/ϩ , and PirB Ϫ/Ϫ mice were generated by Syken et al. (2006). Briefly, PirB fl/fl mice were generated by electroporating a construct with loxP sites flanking exons 10 -13 of PirB into the 129 J1 ES line derived from agouti 129S4/SvJae mice. Exons 10 -13 of PirB code for the transmembrane domain of PirB, as well as part of the intracellular domain encompassing the signaling immunoreceptor tyrosine-based inhibitory motifs. Thus, Cremediated excision of exons 10 -13 from the PirB gene in PirB fl/fl mice results in a truncated protein that is unable to signal, as shown previously by anti-phosphotyrosine immunoprecipitation experiments (Syken et al., 2006). To generate mice with germ line deletion of PirB (i.e., PirB Ϫ/ Ϫ), a deleter strain that targets Cre-recombinase expression to early mouse embryo via adenovirus EIIa promoter (B6.FVB-TGN(EIIa-cre)C5379Lmgd; The Jackson Laboratory) was crossed to PirB fl/fl mice. Heterozygote sibling matings were then used to generate both control PirB ϩ/ϩ line and a homozygous PirB Ϫ/Ϫ line (Syken et al., 2006). PirB fl/fl , PirB ϩ/ϩ , and PirB Ϫ/Ϫ mice were maintained as three separate lines on the same mixed genetic background (C57BL/6 ϫ SV/129J). Previous studies have shown that the excision of PirB from PirB fl/fl by Cre recombinase under control of the UbC promoter occurs within 1 week (Bochner et al., 2014), and is accompanied by a complete loss of PirB protein after ϳ3 weeks from the onset of Cre recombinase expression (Fig. 1;Bochner et al., 2014). Cre recombinase expression via the GFP.Cre construct under the phosphoglycerate kinase promoter used here for the electroporation experiments described below should be even more rapid and efficient (Qin et al., 2010).
All experiments were performed in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and approved by the Stanford University Institutional Animal Care and Use Committee. Experimental methods are also in accordance with the policies of the Society for Neuroscience on the Use of Animals and Humans in Neuroscience Research. All mice were maintained in a pathogen-free environment.

In utero electroporation
Female mice were mated within the same line (PirB ϩ/ϩ , PirB fl/fl , or PirB Ϫ/Ϫ ) and checked daily for vaginal plugs. The day that a plug was found was counted as embryonic day 0.5 (E0.5). In utero electroporation was performed at E15.5, when L2/3 cortical neurons are generated (Saito and Nakatsuji, 2001;Tabata and Nakajima, 2001;Saito, 2006;Chen et al., 2008). Pregnant mice were anesthetized using 1-2.5% isoflurane in 100% O 2 . Using a sterile surgical technique, a midline incision was made to the abdominal wall to expose the uterine horns. Lateral ventricles of embryos were injected with either 1.0 or 1.5 l of 1.9 -2.0 g/l GFP.Cre in Tris-EDTA buffer (10 mM Tris-HCl, pH 8.0, and 1 mM ethylenediaminetetraacetic acid); injection of the lower (1.0 l) volume of DNA was critical to achieving sparse electroporations.
GFP.Cre is an expression construct in which GFP expression is driven by the ubiquitin C promoter, and Cre expression is driven separately by the phosphoglycerate kinase promoter (gift from Tyler Jacks, Koch Institute for Integrative Cancer Research at the Massachusetts Institute of Technology, Cambridge, Massachusetts; plasmid #20781, Addgene; Andreu-Agullo et al., 2011;Scotto-Lomassese et al., 2011). Injections were achieved using micropipettes made from glass capillary tubes (TW100F-4, World Precision Instruments) pulled into a fine tip with a micropipette puller (P-97, Sutter Instruments). Tweezer-type circular electrodes (5 mm in diameter) were custom made by coiling 24 AWG platinum wire (PTP201, World Precision Instruments) and were used to deliver five 50 ms electric pulses at 45 V with 950 ms intervals, using a squarewave generator (ECM 830, BTX). The exposed uterus was kept moist with 0.9% saline at 37°C. After each electroporation procedure, dams were given buprenorphine intraperitoneally (0.1 mg/kg; catalog #2808, Tocris Bioscience), the abdominal wall was sutured shut, and the dam was allowed to recover in a cage kept at ϳ37°C. Pregnant dams were monitored postoperatively until after giving birth and able to nurse pups.

Histology
Mice received an overdose of sodium pentobarbital (Ͼ86 mg/kg) and sodium phenytoin cocktail (Beuthanasia-D, Merck; Ͼ11 mg/kg, i.p.), and brain tissue was fixed via transcardial perfusion of ice-cold 0.1 M sodium PBS followed by ice-cold 4% (w/v) paraformaldehyde in 0.1 M PBS. Brains were then postfixed overnight in 4% (w/v) paraformaldehyde in PBS at 4ºC, followed by 12-24 h in PBS at 4ºC. Brains were cut coronally using a vibrating microtome (VT1200S, Leica Microsystems) into 150-mthick sections. Sections were mounted on Superfrost Plus glass slides (VWR) with ProLong Gold antifade reagent (Invitrogen) as a mounting medium and covered with #1.5 thickness cover glass (VWR).

Dendritic spine and axonal bouton imaging and analysis
Slides of histological sections prepared as described above were initially screened for GFP fluorescence using low-powered objectives on an Eclipse E800 microscope (Nikon), without knowledge of genotype. GFP ϩ neurons in L2/3 of visual cortex were then identified by comparing landmarks observed via DAPI fluorescence and brightfield imaging (Eclipse E800 microscope), including the shape of the internal capsule, the hippocampus, thalamic structures, and cytoarchitectonic differences between the cortical layers (Paxinos and Franklin, 2008). Labeled cells with low expression levels of GFP and without complete primary dendritic arbors were excluded from further imaging. High-resolution images of apical and basal dendrites of L2/3 pyramidal neurons in visual cortex, and of continuous 100 -300 m segments of descending axons in L5, were taken on an SP2 or SP8 confocal microscope (Leica Microsystems), or on a two-photon microscope (Prairie Technologies); 63ϫ/1.40 numerical aperture (NA) oil-immersion (Leica) or 60ϫ/1.1 NA water-immersion (Prairie) objectives were used. Images were acquired at or over the theoretical Nyquist sampling rate for each objective used (ϳ71 nm/pixel). All dendritic spine and axonal bouton analysis was performed manually while blind to genotype using ImageJ software (National Institutes of Health); manual tracing with ImageJ Simple Neurite Tracer plugin was used for length measurement of neurites. Axonal boutons were identified as in De Paola et al. (2006).

New Research
L2/3 neuron dendritic arbor in a 150-m-thick section were used for Sholl analysis (Sholl, 1953). For each cell, concentric circles were drawn at 10, 30, 50, 70, 90, 110, and 130 m from the geometric center of the soma (0 m), using the Concentric Circles ImageJ plugin. The number of crossings of dendritic branches belonging to individual cells through each circle were counted, and used as the measure of the complexity of the dendritic arbor.

Assessment of neuronal labeling density resulting from electroporations
Low-magnification images of L2/3 GFP ϩ ;Cre ϩ PirB fl/fl neurons were acquired (Eclipse E800 microscope) that included all GFP ϩ neurons within a 0.5 mm radius of a GFP ϩ neuron of interest. For each cell, concentric circles were drawn at 50 m intervals from the geometric center of the soma (0 m), using the Concentric Circles ImageJ plugin. The number of GFP ϩ cells located within each ring (e.g., 50 -100 m) was then counted to derive a density versus distance measurement of distribution of the electroporated neighboring cells surrounding a neuron of interest ( Fig. 2F).
The whole-cell patch-clamp technique was used to record miniature EPSCs (mEPSCs) from individual pyramidal cells in slices containing primary visual cortex and binocular zone. Single, GFP-labeled or unlabeled L2/3 pyramidal neurons in visual cortex were visualized using infrared differential interference contrast illumination combined with fluorescent illumination in an Olympus BX51WI Microscope via a 60ϫ water-immersion objective coupled with an additional 2ϫ zoom lens (120ϫ final magnification). For GFP Ϫ cells, only those with pyramidal-shaped somata and a visible apical dendrite oriented perpendicular to the pial surface were selected for recordings. The external bath solution (ACSF) was maintained at ϳ30°C, and bubbled with 95% O 2/ 5% CO 2 for the duration of the experiment. Patch-clamp electrodes (1.5-4.2 M⍀) were filled with cesium-based intracellular solution (composition: 135 mM CsCl, 10 mM HEPES, 1 mM EGTA, 4 mM Mg-ATP, 0.4 mM Na-GTP, pH 7.4 with NaOH). To reveal the morphology of neurons from which mEPSCs were recorded, biocytin (catalog #B4261, Sigma-Aldrich) was added to the intracellular solution at a final concentration of 5 mg/ml; after recording, sections were fixed in 4% paraformaldehyde overnight, followed by permeabilization and staining with Texas Red-conjugated avidin (catalog #A-2006, Vector Laboratories; 1:10). Once in wholecell mode, the pipette solution equilibrated with the cell contents for 5-10 min before recording. Miniature excitatory postsynaptic AMPA currents were isolated pharmacologically using 1 M tetrodotoxin (catalog #1078, Tocris Bioscience), 50 M aminophosphonopentanoic acid (catalog #A5282, Sigma-Aldrich), and 10 M gabazine (SR95531; catalog #1262, Tocris Bioscience). Series resistance was 9 -20 M⍀. The low-noise AxoPatch 200B Patch-Clamp Amplifier, DigiData 1322A Digitizer, and Clampex Software (Molecular Devices) were used for data collection. Before analysis, recordings across all conditions were filtered with a 60 Hz, three-harmonic, 100 cycle electrical interference filter (Molecular Devices), as well as a Gaussian, 53 coefficient low-pass filter (setting, Ϫ3 dB to 2 kHz; Molecular Devices). Analysis of mEPSCs was performed using MiniAnalysis version 6.0.7 (Synaptosoft), and blinded to both genotype and cell identity (i.e., GFP ϩ vs GFP -). Cells with cell capacitance of 60 pF and/or leak current more negative than Ϫ150 pA were excluded from the analysis. Root mean square (rms) baseline noise was calculated from three different places during the 10 minute recording, and was Ͻ3 pA. Automatic detection parameters in MiniAnalysis were adapted from the studies of Xu et al. (2009a) and Muhia et al. (2012), and were as follows: threshold was set to 4 times the rms noise value; the period to search for a local maximum was 20 ms; the time before peak to establish baseline was 5 ms; the period to search a decay time was 5 ms; the fraction of the peak to find a decay time was 0.5; the period to average baseline was 2 ms; the area threshold was five times the rms noise value; and the number of points to average to establish the peak was 3. Continuous automatic analysis was run during 1-2 min of the 10 min recording in which the leak current was most stable; detected events were manually inspected to eliminate false positives.

Statistics
All statistical analyses and graphs were performed with GraphPad Prism software, and power analyses were performed using G‫ء‬Power software (Heinrich-Heine-Universität Düsseldorf, Düsseldorf, Germany; http://www.gpower.hhu.de/; Faul et al., 2007Faul et al., , 2009). Data are presented as the mean Ϯ SEM; sample size (n) is the number of cells analyzed, followed by the number of mice used for the entire sample, unless noted otherwise in the text (Table 1). Analyses were performed blinded to genotype and condition.

Results
To examine whether PirB expression, specifically in pyramidal neurons, is sufficient to regulate spine density, PirB ϩ/ϩ , PirB Ϫ/Ϫ , and PirB fl/fl mice were studied in combination with in utero electroporation of a GFP.Cre expression vector to selectively target a subset of excitatory pyramidal neurons in L2/3 at the time of their genesis. Because excitatory neurons from different cortical layers are generated sequentially in the cortical ventricular zone (McConnell and Kaznowski, 1991;Chen et al., 2008;Greig et al., 2013), neurons of each cortical layer can be targeted by timing in utero electroporations of the ventricular zone; at E15.5, L2/3 pyramidal neurons are targeted exclusively (Saito and Nakatsuji, 2001). Glia and interneurons are not transfected since the production of glial cells peaks around birth, while interneurons are generated in the ganglionic eminence and not in the ventricular zone (Anderson et al., 1997;Nery et al., 2002). Layer 2/3 pyramidal neurons are of particular interest because of the changes previously observed in PirB Ϫ/Ϫ mice in cellular mechanisms of synaptic plasticity at L4 to L2/3 synapses, in mEPSC frequency recorded from L2/3 neurons, and in OD plasticity as assessed in L2/3 using intrinsic signal imaging .

L2/3 pyramidal neurons of PirB ؊/؊ mice have elevated dendritic spine density
This first experiment was designed to examine whether spine density on L2/3 neurons in PirB Ϫ/Ϫ mice is increased at P30. At E15.5, PirB ϩ/ϩ or PirB Ϫ/Ϫ embryos were injected with a DNA construct (GFP.Cre) expressing GFP, as well as Cre recombinase under a separate promoter and subjected to electroporation. Note that this initial experiment was not performed in mice carrying a conditional allele of PirB. Consequently, neurons are labeled with GFP to permit the assessment of morphology, but Cre expression does not drive excision, and instead serves as a control for subsequent experiments.
Electroporation at E15.5 results in laminar-specific GFP labeling of L2/3 pyramidal neurons in primary visual cortex at P30 (Fig. 1A-D). Dendritic spines were extensively GFP ϩ labeled as viewed using confocal microscopy ( Fig.  1E-H), permitting assessment of density. Spine density on L2/3 pyramidal neurons of PirB Ϫ/Ϫ mice is significantly elevated, relative to that in PirB ϩ/ϩ mice; spine density is 64% greater on apical dendrites in PirB Ϫ/Ϫ versus PirB ϩ/ϩ (Fig. 1I), and 73% greater on basal dendrites (Fig. 1J). The excess dendritic spines are likely to be sites of functional synapses, since EM studies from mouse visual cortex show that Ͼ95% of dendritic spines have machinery needed for synaptic transmission: they are positive for PSD-95 and apposed by presynaptic active zones (Arellano et al., 2007). Moreover, the spine density increase observed here in PirB Ϫ/Ϫ mouse visual cortex is consistent with the previously reported increase in mEPSC frequency recorded from L2/3 pyramidal neurons-a sensitive measure of the density of functional excitatory synapses . These significant changes in dendritic spine density in PirB Ϫ/Ϫ neurons suggest that other aspects of neuronal morphology may also be altered. To examine this possibility, the dendritic complexity of GFP ϩ L2/3 pyramidal neurons in visual cortex of PirB ϩ/ϩ and PirB Ϫ/Ϫ mice was assessed using Sholl analysis (Sholl, 1953). Apical and basal dendritic complexity was unaltered in PirB Ϫ/Ϫ mice relative to PirB ϩ/ϩ mice (Fig. 1K,L). Together, these observations demonstrate that the changes in mEPSC frequency observed previously ) are accompanied by changes in spine density but not dendritic complexity, in L2/3 pyramidal neurons.  (Fig. 4G) Normality not assumed Mann-Whitney 0.8509 0.0058 i (Fig. 4I) Normality not assumed Mann-Whitney 0.1634 0.3286 j (Fig. 4K) Normality not assumed Mann-Whitney 0.0961 0.4940 k (Fig. 4M) Normality not assumed Mann-Whitney 0.1855 0.3918 l (Fig. 5E Sparse deletion in PirB fl/fl mice results in elevated spine density on isolated L2/3 pyramidal neurons An increase in both spine density (Fig. 1) and in functional excitatory synapses  is observed in L2/3 pyramidal neurons of PirB Ϫ/Ϫ versus PirB ϩ/ϩ mice. To examine whether this increase is due to the loss of PirB from the very cells where dendritic spines were counted or to the loss of PirB in other presynaptic or non-neuronal cells, a mosaic approach was used to delete PirB from isolated L2/3 pyramidal neurons. PirB fl/fl mice were electroporated with GFP.Cre at E15.5, and then studied at P30 ( Fig. 2A-E). In PirB fl/fl mice, cells expressing GFP coexpress Cre recombinase, which drives the excision of the floxed PirB alleles.
In a subset of PirB fl/fl mice electroporated with the GFP.Cre vector (see Materials and Methods), a very sparse distribution of PirB knock-out neurons embedded in a "sea" of cells containing intact PirB alleles was obtained ( Fig. 2A-E). To quantify the sparseness of PirB deletion, the density of GFP ϩ cells in V1 was measured by counting the number of labeled cells in a series of concentric circles surrounding the cell selected for spine analysis (Fig. 2F). Using this Sholl analysis, we determined that for the majority of neurons analyzed, there were almost no neighboring GFP ϩ neurons located within a radius of 50 m, and then within each 50 m increment outward from the cell under analysis, there was an average of only one additional labeled neuron at most. In contrast, the density of labeled cells was far greater in Figure 1, where a larger volume of DNA was injected. Thus, electroporation could result in a very sparse deletion of PirB from isolated L2/3 pyramidal neurons, permitting us to examine a cell-intrinsic function for PirB in regulating dendritic spine density.

Dendritic spine density on L2/3 pyramidal cells in all genotypes is similar at P23
The elevated spine density observed at P30 in mice lacking PirB could arise from changes in dendritic spine formation and/or from a failure of pruning. To distinguish these possibilities, the sparse GFP.Cre electroporation experiment described above was repeated, but the density of spines was assessed 1 week earlier at P23, during the period of synaptic pruning and when dendritic spine density in WT visual cortex is highest based on previous Golgi cell and EM studies (Ruiz-Marcos and Valverde, 1969;Markus and Petit, 1987).
The dendrites and spines of L2/3 pyramidal neurons at P23 were extensively GFP labeled (Fig. 3A,B) subsequent to electroporation at E15.5. Multiple comparisons with one-way ANOVA and Bonferroni post hoc test did not reveal any significant differences between the density of spines for all three genotypes (Fig. 3C,D). Moreover, dendritic spine density for PirB ϩ/ϩ is almost twice as high at P23 (Fig. 3C: ϳ13 spines/10 m) than at P30 (ϳ7 spines/10 m). In contrast to wild-type, there are no statistically significant differences between PirB fl/fl mice at P23 versus P30, or PirB Ϫ/Ϫ mice at P23 versus P30, strongly suggesting that spine pruning is deficient in L2/3 neurons lacking PirB.

Sparse PirB excision results in increased mEPSC frequency recorded from isolated L2/3 pyramidal neurons at P30
In PirB Ϫ/Ϫ mice, the increase in spine density on L2/3 pyramidal neurons (Fig. 1) is accompanied by an increase in mEPSC frequency, a measure of the density of functional excitatory synapses . Is the increase in dendritic spine density observed following sparse deletion in PirB fl/fl mice similarly accompanied by an increase in mEPSC frequency? To assess the strength and number of functional excitatory synaptic inputs, the frequency and amplitude of mEP-SCs in acute slices from visual cortex of PirB fl/fl mice was assessed using the patch-clamp technique (Fig. 4). In this experiment, both labeled (GFP ϩ ;Cre ϩ ) and unlabeled (GFP Ϫ ;Cre Ϫ ) neurons could be recorded and compared in the same slice (Fig. 4A-D); cells were also labeled via patch pipette with biocytin to confirm the pyramidal identity of GFP Ϫ ;Cre Ϫ cells targeted for recordings (Fig. 4E,F). The frequency of mEPSCs in GFP ϩ ;Cre ϩ neurons was 69% higher than in GFP Ϫ ; Cre Ϫ controls (Fig. 4G,H), with no measurable change in mEPSC amplitude (Fig. 4I,J).
Miniature EPSC frequencies and amplitudes recorded from GFP ϩ ;Cre ϩ neurons in PirB fl/fl mice (Fig. 4G-J) were similar to those of L2/3 pyramidal neurons in germline PirB Ϫ/Ϫ slices . In addition, GFP Ϫ ; Cre Ϫ neurons in PirB fl/fl slices had mEPSC frequency and amplitudes similar to PirB ϩ/ϩ slices (Fig. 4, compare G-J,  K-N). These results indicate that the in utero electropora-  ϭ 8 cells, 7 mice). The graph shows that, on average, there was only one GFP ϩ ;Cre ϩ cell in every 50 m increment analyzed. G, Spine density on apical dendrites is greater in GFP ϩ ;Cre ϩ neurons from PirB fl/fl mice than PirB ϩ/ϩ mice (PirB ϩ/ϩ : 7.0 Ϯ 0.6 dendritic spines/10 m of dendritic length, n ϭ 9 cells, 5 mice; same data as in Fig. 1I; PirB fl/fl , GFP ϩ ;Cre ϩ : 10.6 Ϯ 0.8, n ϭ 9 cells, 7 mice; PirB Ϫ/Ϫ , GFP ϩ ;Cre ϩ : 11.5 Ϯ 1.0, n ϭ 13 cells, 5 mice; same data as in Fig. 1I; PirB ϩ/ϩ vs PirB fl/fl , p ϭ 0.041 a ; PirB fl/fl vs PirB Ϫ/Ϫ , p ϭ 1.000 a , one-way ANOVA with post hoc Bonferroni's multiple comparisons). H, Basal dendritic spine density is greater in GFP ϩ ; Cre ϩ neurons from PirB fl/fl mice than PirB ϩ/ϩ mice (PirB ϩ/ϩ : 6.6 Ϯ 0.6 dendritic spines/10 m of dendritic length, n ϭ 9 cells, 5 mice; data from tion technique by itself does not affect the development of functional excitatory synaptic inputs to L2/3 pyramidal neurons. To control for off-target effects of Cre recombinase and GFP expression on mEPSC recordings, we also examined mEPSC frequency and amplitude in PirB ϩ/ϩ tissue subsequent to electroporation with GFP.Cre. In this case, Cre recombinase does not cause excision, and thus labeled and unlabeled cells differ only in their expression of GFP and Cre. Results show that GFP ϩ ;Cre ϩ neurons in PirB ϩ/ϩ slices do not differ significantly from GFP Ϫ ;Cre Ϫ neurons in PirB ϩ/ϩ slices in either mEPSC frequency (Fig.  4K,L) or amplitude (Fig. 4M,N). This experiment demonstrates that the electroporation and expression of GFP and Cre in and of themselves do not alter detectably these electrophysiological properties of L2/3 neurons. Together, these observations suggest that PirB is required to regulate the functional excitatory synaptic inputs of the neuron, and that the loss of PirB just from the isolated cell examined is sufficient to account for changes in the density of both functional synapses and spines observed in PirB Ϫ/Ϫ mice.
It should be noted that the 59% increase in mEPSC frequency observed in GFP ϩ ;Cre ϩ neurons in PirB fl/fl mice (Fig. 4G,H) is similar to the average 69% increase in spine density on both apical and basal dendrites (Fig. 2G-J), implying that many, if not all, of the supernumerary dendritic spines represent sites of functional excitatory synapses.

mEPSC frequency is not altered in wild-type neighbors located within 100 m of L2/3 pyramidal neurons lacking PirB
To test further for a cell-intrinsic effect of PirB on the density of functional synaptic inputs, patch-clamp recordings were made from GFP ϩ ;Cre ϩ and GFP Ϫ ;Cre Ϫ L2/3 pyramidal neurons that were in close proximity, separated by Յ100 m from each other (Fig. 5). Within this distance, neurons are known to have a much higher connection probability (Perin et al., 2011;Hill et al., 2012;Jouhanneau et al., 2015). In recordings from pairs of neurons located Յ100 m from each other (Fig. 5A-D), we found that mEPSC frequencies in GFP ϩ ;Cre ϩ neurons were almost always greater than those for the unlabeled GFP Ϫ ;Cre Ϫ neighbor (14 of 16 pairs; Fig. 5E,G). Once again, mEPSC amplitudes between GFP ϩ ;Cre ϩ and GFP Ϫ ;Cre Ϫ neighbors did not differ significantly (Fig. 5F,H). The average mEPSC frequency of GFP ϩ ;Cre ϩ neurons (Fig. 5E) was similar to that previously recorded from L2/3 pyramidal neurons in germline PirB Ϫ/Ϫ mice . The mEPSC frequencies of neighboring GFP Ϫ ;Cre Ϫ neurons (Fig. 5E) were close to previously reported levels in PirB ϩ/ϩ slices .

Discussion
Many factors are now known to regulate aspects of dendritic spine shape, size, and stability, including experience, learning, and environmental enrichment (Holtmaat and Svoboda, 2009;Chen et al., 2014), but just how these external activity-dependent signals are read out into lasting changes is still relatively unclear. Here we have shown that PirB, a receptor whose MHC class I ligands are regulated by neural activity (Corriveau et al., 1998;Huh et al., 2000), acts within individual L2/3 pyramidal neurons in a cell-intrinsic manner to regulate spine density and functional excitatory synapses. We have used the powerful technique of in utero electroporation of Cre.GFP to target PirB deletion selectively to L2/3 pyramidal neurons. It is remarkable that simply driving PirB excision in single isolated neurons surrounded by wild-type glia and wild-type neurons results in a major increase in spine density within the targeted neuron. This observation indicates that PirB function in individual L2/3 pyramidal neurons is required to regulate spine density. It is also consistent with the previous finding that spine density on L5 pyramidal cells is also elevated in mice lacking PirB . Together, these observations imply that PirB may regulate spine density more generally in excitatory neurons of the forebrain.
At P30, spine density on L2/3 pyramidal neurons lacking PirB is Ͼ50% greater than in PirB ϩ/ϩ neurons (Figs. 1,  2). However, just 1 week earlier, at P23, we found that spine density across all genotypes exposed to GFP.Cre at E15.5 is similar, ϳ13 spines/10 m (Fig. 3). Between P23 and P30, spine density on L2/3 PirB ϩ/ϩ neurons falls to about seven spines/10 m, which is consistent with the idea that this is a peak period for spine and synapse pruning (Ruiz-Marcos and Valverde, 1969;Markus and Petit, 1987). In contrast, spine density on neurons lacking PirB fails to decline and instead remains close to the level at P23. Overall, results strongly suggest that PirB function is required for spine pruning during this period. Given the strong evidence of a role for microglia in spine and synapse pruning (Schafer et al., 2012;Parkhurst et al., 2013), our results also imply that, without neuronal PirB expression, glia may not be able to function properly to eliminate spines.
Several lines of evidence presented here argue that the striking increases in spine density and functional synapses are not artifacts of Cre recombinase or GFP expression. First, a significant increase in spine density is observed when Cre.GFP is electroporated into PirB Ϫ/Ϫ mice ( Fig. 1E-J). Second, a similar result is obtained when Cre.GFP is electroporated into PirB fl/fl mice ( Fig. 2D-J). Third, the independent technique of mEPSC recordings revealed a parallel increase in functional excitatory inputs (Figs. 4,5), as signaled by the significantly greater mEPSC frequency in PirB Ϫ/Ϫ versus PirB ϩ/ϩ neurons. Moreover, whole-cell recordings of mEPSCs made from isolated neurons electroporated with Cre. GFP (Figs. 4,5) are indistinguishable from those recorded in germline PirB Ϫ/Ϫ neurons . Finally, when Cre.GFP is electroporated into PirB ϩ/ϩ mice, mEPSCs recorded from L2/3 neurons (Fig. 4K-N) are similar to those seen in unmanipulated PirB ϩ/ϩ mice not undergoing electroporation .
Here we have uncovered a cell-intrinsic mechanism for maintaining spine density and excitatory synaptic input below an upper bound. L2/3 neurons lacking PirB have as much as a 73% greater spine density than PirB ϩ/ϩ neurons (Figs. 1, 2). The corresponding increase in mEPSC frequency (Fig. 4) suggests that many of the spines on isolated PirB Ϫ/Ϫ cells are functional, a finding similar to what has been observed on the L5 pyramidal neurons in PirB Ϫ/Ϫ mice . Moreover, we observed that neurons with intact PirB alleles have lower mEPSC frequencies than their isolated neighbors lacking PirB (Fig. 5). This observation is one of the strongest arguments for a cell-autonomous role for PirB in L2/3 pyramidal neurons. The fact that mEPSC frequency in wild-type neighbors is indistinguishable from that recorded in L2/3 PirB ؉/؉ mice, and that mEPSC frequency in isolated PirB Ϫ/Ϫ cells is indistinguishable from that recorded in L2/3 neurons in PirB Ϫ/Ϫ mice, is additional support for this conclusion.
The term "cell autonomous" is used here explicitly in reference to a role for PirB in regulating spine density on L2/3 pyramidal neurons. There are several other examples in which spine density is thought to be regulated by cell-autonomous mechanisms, including NgR1 in the cortex (Akbik et al., 2013) and Sema5A in the hippocampus (Duan et al., 2014). In all of these cases, including PirB, the conclusion regarding cell-autonomous function could not have been achieved without mosaic analysis in vivo or in vitro. PirB may also have non-cell-autonomous functions. For example, the increase in mEPSC frequency recorded in isolated neurons lacking PirB implies that there has been a parallel increase in the number of presynaptic boutons, which likely derive from a vast majority of PirB ϩ/ϩ neurons. Thus, the spine density increase appears to drive a transynaptic increase in functional inputs: a non-cell-autonomous effect. This interpretation is also consistent with the finding here that in the sparse electroporation experiments, the axon bouton density of neurons lacking PirB does not differ from that of PirB ϩ/ϩ neurons (Fig. 6). These mutant axons are embedded in a sea of PirB ϩ/ϩ neurons possessing normal spine density. A non-cell-autonomous role for PirB in regulating presynaptic boutons would predict that the density of axonal inputs belonging to PirB Ϫ/Ϫ neurons should be wild type, exactly what we have observed.
There is a growing list of molecules known to regulate spines and excitatory synaptic inputs to cerebral cortical pyramidal cells. For instance, FMRP regulates spine density and maturation; in the knockout, spine density is increased, but spines remain thin and immature, and can be rescued by reducing mGluR5 expression (Dölen et al., 2007). Another receptor, DR6, regulates the density of axonal boutons and sprouting following activity-dependent deprivation, but there is no known effect on the normal developmental spine-pruning process (Marik et al., 2013). NgR1 is thought to regulate the density of mature dendritic spines (Karlsson et al., 2016); however, there is currently some disagreement about its exact role in dendritic spine and axonal bouton turnover (Akbik et al., 2013;Park et al., 2014;Frantz et al., 2016). Many downstream effectors have been identified and studied, including Rho GTPases (Murakoshi et al., 2011;Colgan and Yasuda, 2014) and the actin cytoskeleton Kellner et al., 2016). In contrast, our studies suggest that PirB acts to keep spine density below a ceiling level , with no apparent effect either on the distribution of spine types (Bochner et al., 2014) or on the overall dendritic morphology (Figs. 1K,L, 2K,L ). Spine motility on L5 pyramidal neurons lacking PirB is also decreased , implying a connection between PirB and downstream signaling to cofilin and the actin cytoskeleton . Clearly, every aspect of the spine is tightly regulated to enable experiencedependent changes to be encoded structurally. Together, the results of our studies suggest that PirB is needed to match spine density and excitatory synaptic function to activity levels within cortical circuits, thereby providing headroom for the cell to encode additional experiences at new synapses.