Pten and EphB4 regulate the establishment of perisomatic inhibition in mouse visual cortex

Perisomatic inhibition of pyramidal neurons is established by fast-spiking, parvalbumin-expressing interneurons (PV cells). Failure to assemble adequate perisomatic inhibition is thought to underlie the aetiology of neurological dysfunction in seizures, autism spectrum disorders and schizophrenia. Here we show that in mouse visual cortex, strong perisomatic inhibition does not develop if PV cells lack a single copy of Pten. PTEN signalling appears to drive the assembly of perisomatic inhibition in an experience-dependent manner by suppressing the expression of EphB4; PV cells hemizygous for Pten show an ∼2-fold increase in expression of EphB4, and over-expression of EphB4 in adult PV cells causes a dismantling of perisomatic inhibition. These findings implicate a molecular disinhibitory mechanism driving the establishment of perisomatic inhibition whereby visual experience enhances Pten signalling, resulting in the suppression of EphB4 expression; this relieves a native synaptic repulsion between PV cells and pyramidal neurons, thereby promoting the assembly of perisomatic inhibition.

P erisomatic inhibition in cortex is established by fast-spiking interneurons called basket cells that express the calcium binding protein parvalbumin (PV). The establishment of this inhibition during postnatal development is centrally important to the opening of critical periods and the establishment of neural gain control, and its failure is linked to a number of disorders of cortical development, including autism and schizophrenia [1][2][3][4] . Remarkably little is known about how PV neurons establish strong synaptic input to local pyramidal neurons during development.
By comparison, a great deal more is known about the postnatal growth of pyramidal neurons. In mice, the growth of axons and dendrites is most pronounced during the first few weeks of postnatal life. Thereafter, growth is slowed as neurons begin to express PTEN 5 . Deleting Pten from adult pyramidal neurons is sufficient to re-start dendritic and axonal growth [5][6][7] .
Embryonic PTEN deletion from inhibitory neurons impacts the establishment of appropriate numbers of parvalbumin and somatostatin-expressing interneurons, demonstrating a central role for Pten in inhibitory cell development 8 . To examine how Pten signalling impacts PV cells specifically, we use loxP-mediated recombination to generate mice in which PV cells lacked one or both copies of Pten. Heterozygous animals (PV-Pten þ / À ) were studied exclusively, as complete knockouts died prematurely from seizures. Our studies focus on the primary visual cortex, where the development of perisomatic inhibition has been well characterized and is known to be strongly regulated by visual experience [9][10][11] . We find that single copy Pten deletion from PV cells impairs the formation of perisomatic inhibition. Pten expression in PV cells increases with visual experience, and Pten signalling appears to suppress the expression of the synaptic repulsive factor, EphB4, which is initially high in PV cells. Diminishing Pten expression either by dark rearing or by single copy gene deletion maintains elevated EphB4 expression, thereby repelling the establishment of PV cell synapses onto pyramidal neurons.

Results
Pten in PV cells regulates perisomatic inhibition. All studies were carried out using PV-Cre knockin mice obtained from The Jackson Laboratory (B6;129P2-Pvalb tm1(cre)Arbr /J). To validate the expression of Cre recombinase in these mice, we crossed them with the Ai9 line of reporter mice expressing a flexed tdTomato knocked in to the Rosa26 locus to generate mice expressing tdTomato in cells expressing Cre; the fraction of PV cells that also expressed tdTomato, and thus Cre recombinase in visual cortex, was then measured at three ages (postnatal days 28, 40 and 60) by immuno-staining against PV in fixed, frozen coronal sections. At P28, 92 of 93 PV cells counted in three sections (99%) were co-labelled; at P40 this ratio was 166/168 (99%), and at P60 it was 166/184 (90%), indicating that the PV-Cre line of mice reliably express Cre recombinase and tdTomato in virtually all PV cells (Fig. 1). These values are similar to those reported in ref. 12.
To examine the impact of Pten haploinsufficiency on PV cell circuitry, we expressed the light-activated cationic channel, channelrhodopsin-2 (ChR2), in PV cells of PV-Pten þ / À and littermate control mice; subsequently laser scanning photostimulation of ChR2-expressing PV cells with concurrent whole-cell recordings of layer 2/3 pyramidal neurons was used to measure the strength of PV cell input to pyramidal neurons in primary visual cortex in acute brain slices (P35-P56). We note that mapping studies based on ChR2 expression may be confounded by differences in the optical recruitment of PV cells. With this approach, we found that PV cells hemizygous for Pten produced only half of the total inhibitory drive to pyramidal cells of wild-type PV cells ( Fig. 2; Supplementary Figs 1 and 2). This reduced inhibitory current could not be attributed to a reduced expression of ChR2 or responsiveness to blue light stimulation ( Supplementary Fig. 3), or to changes in the intrinsic properties of PV cells in the experimental group ( Supplementary Fig. 4), as these measures were comparable between PV cells hemizygous for Pten and controls. Nor was it due to a reduction in the number of PV cells in mutants (Cells per field of view visualized using a Â 4 objective lens with a field number of 26.5; mean and standard deviation: WT ¼ 354 þ / À 40; Pten ¼ 367 þ / À 60; 10 fields of view per group; P ¼ 0.73, Wilcoxon rank sum). This reduction in PV cell-derived inhibitory current could result from pyramidal neurons receiving fewer PV cell-derived inhibitory synapses; these synapses could be present in normal numbers, but weaker synapses; or it could be some combination of these two. If there are fewer synapses, this would be reflected in a decrease in connectivity. If the synapses are weaker, this would be reflected in a decrease in paired-pulse ratio. To directly measure these, we conducted paired whole-cell recordings from adjacent PV cells and pyramidal neurons and, separately, from adjacent pairs of PV cells (Fig. 3). Recordings were done in L2/3 of the primary visual cortex of young adult mice (P35-56). PV cells were identified by their expression of td-Tomato, high frequency firing pattern, narrow action potentials and sharp after-hyperpolarizations. Pyramidal neurons were identified by their lack of td-Tomato expression and the presence of an accommodating firing pattern (cf. Fig. 3a). Connectivity between the two cells was assessed by stimulating the presynaptic neuron in current clamp and recording from the putative postsynaptic neuron in voltage clamp (cf. Fig. 3b). PV cells hemizygous for Pten exhibited a 32% decrease in probability of connecting to adjacent pyramidal neurons relative to controls (Fig. 3c). This 32% reduction is a measure of connectivity and is distinct from the 50% reduction in total PV cell-derived inhibitory current in pyramidal neurons reported in Fig. 1, which is a measure of total input strength. Notably, the probability of finding the reverse connection, from pyramidal cell to PV cell hemizygous for Pten was no different from control; nor was there any change in the probability of connection between mutant PV cells and between normal PV cells (Fig. 3c).
When we examined the intrinsic excitability of pyramidal neurons in these slices, we found a trend towards increased excitability in PV-Pten þ / À mice ( Fig. 3d; only comparisons at 150 pA had a P value o0.05, no other comparisons revealed significant differences). This data trend could be due to a true change in the intrinsic properties of these excitatory neurons, despite having normal Pten copy number, or it could simply be an epiphenomenon of the reduction in inhibition in these slices, one result of which would be an increase in recurrent excitatory response following the firing of the recorded excitatory neuron. Supporting this latter view, the data trend towards increased To determine whether the synapse that are established by the mutant PV cells are of normal strength, we measured response amplitudes and paired pulse ratio of extant connections. We found no significant differences in response amplitudes in any of the types of synapses examined; PV to pyramidal, pyramidal to PV, and PV to PV were all comparable between PV-Pten þ / À mice and controls ( Fig. 3f-h). Measures of paired pulse ratio were also equivalent (Fig. 3i). These measures indicate that the synapses established by mutant PV cells onto local pyramidal neurons are as strong as synapses established by normal PV cells.
Taken together, these findings show that single copy Pten loss in PV cells reduces the probability that any given PV cell will establish a connection with a local pyramidal neuron; this, in turn, causes a significant reduction in the strength of perisomatic inhibition. This effect is specific to the connection between PV cells and pyramidal cells.
Mutant PV cells express high levels of EphB4. We next examined how a reduction in Pten expression could result in decreased, synapse-specific connectivity between PV cells and pyramidal neurons. Since PTEN is a phosphatase that negatively regulates the AKT signalling pathway, which in turn regulates gene transcription and protein translation 13  possible differences in gene expression in Pten þ / À and wild-type PV cells. To do so, live PV cells were harvested from the primary visual cortex of intact adult brains via fluorescence-assisted laser cell sorting (Fig. 4a,b). RNA from live PV cells was isolated and amplified, and complementary DNA libraries synthesized. Transcript copy number was quantified using Illumina mouse gene chip arrays, with a total of 26,500 genes screened (Fig. 4c). Seventeen transcripts met established criteria for further examination, as described in ref. 14, namely a 41.4-fold change in expression (logFC40.5) and a P-value o0.001 (Table 1). Of these, the majority encoded intracellular glycosylation proteins. One standout transcript was EphB4 (logFC ¼ 0.56, P ¼ 0.00039), which encodes the presynaptic partner of ephrinB2 15 . Quantitative real-time PCR (ddCt ¼ 1.84) confirmed that Ephb4 RNA was in fact increased in PV cells hemizygous for Pten.
These data suggest a model in which Pten expression in PV cells suppresses expression of EphB4 in these same cells. In the central nervous system, ephrinB/EphB signalling mediates cell-tocell repulsion during development [16][17][18] . Supporting this view, RNA sequencing data of adult mouse primary visual cortical neurons, published from the Allen Brain Institute, shows high expression of EphrinB2 in pyramidal neurons; Pten, but not EphB4 is expressed in adult PV cells (see http://casestudies.brain-map.org/celltax). Increased expression of EphB4 in PV cells could account for the reduction in PV to pyramidal inhibitory synapses via increased repulsion between the presynaptic terminals of PV cells and cell bodies of pyramidal neurons.
EphB4 overexpression dismantles perisomatic inhibition. To directly test whether increased EphB4 signalling in PV cells is sufficient to prevent the normal formation of perisomatic inhibition of pyramidal neurons, we selectively over-expressed Ephb4 in mature, genotypically normal PV cells in the primary visual cortex of mice expressing ChR2 under the PV promoter (PV-Cre/Ai32) via viral transduction of Ephb4 (AAV2/1-Syn. flex.mCherry.D2A.mEphB4; Fig. 5a-c) and then mapped net PV cell-mediated inhibition across all cortical layers onto individual pyramidal neurons via laser scanning photo-stimulation of ChR2 expressing PV cells with concurrent whole-cell recordings of pyramidal neurons as in Fig. 1. Control recordings were obtained from PV-Cre/Ai32 mice that received the same viral injection but failed to express Ephb4. A 67% reduction in local PV mediated inhibition of pyramidal neurons was observed in mice overexpressing Ephb4 (Fig. 5d-f), supporting the view that EphB4 signalling in PV cells impairs the formation of perisomatic inhibitory synapses.

Early vision increases Pten and suppresses EphB4 in PV cells.
Since visual experience is required for the formation of perisomatic inhibition in visual cortex, we wondered whether vision also informs levels of Pten and EphB4 expression in PV cells. To do so, we reared PV-Pten þ / À mice and littermate controls in darkness from birth until postnatal day 35. PV cells in these mice also expressed the red fluorescent protein tdTomato. Using fluorescent-assisted laser cell sorting to extract live PV cells and subsequent qrtPCR normalized to GAPDH expression, we found a B3-fold increase in EphB4 expression in PV cells from dark reared mice relative to normally reared controls, and an B40% reduction in Pten expression (Fig. 6). These data support a model in which visual experience, at least in part, regulates the expression of EphB4 and Pten in primary visual cortex.

Discussion
The establishment of perisomatic inhibition progresses through two stages. In the first stage, factors intrinsic to developing cortex guide PV axons to pyramidal cell somas and proximal dendrites 19 . In the second stage, increased neural activity incited by sensory experience drives the proliferation of perisomatic inhibition 20 . Our results suggest a model in which sensory experience suppresses expression of EphB4 in PV cells. This downregulation of EphB4 likely relieves a native synaptic repulsion between PV cells and pyramidal neurons, thereby promoting the assembly of PV-mediated perisomatic inhibition. The link between increased Pten expression and a reduction in EphB4 levels has been shown in human tumour lines 21 , but has not been studied in the central nervous system. The signalling pathway linking Pten expression to a reduction in EphB4 expression is not revealed in this study, and the complexity of protein social networks, which are quite large, almost certainly guarantees that the connection will not be linear. However, Pten is known to suppress transcription factors. In mouse cortex, Pten expression increases approximately one week after birth and its expression increases through the third postnatal week 22 . In pyramidal neurons, Pten signalling slows the growth of axons and dendrites, and its deletion induces their growth 5,6 . Disruptions of the Pten signalling pathway in excitatory neurons are linked to autism spectrum disorders 23 , and, targeted, conditional deletion is being exploited as a strategy for promoting neural regeneration in the central nervous system 7,24 . Because of its link to autism and its potential therapeutic use in neural regeneration, PTEN signalling has been extensively studied in pyramidal neurons. However, PTEN's influence on the development of inhibitory neurons has only recently been examined 8 . This despite evidence that autism and schizophrenia may be deeply rooted in deficits in inhibition 4,25,26 . The data we present here show that in PV cells, Pten deletion promotes the disassembly of their synaptic input to local pyramidal neurons. This effect is specific to these synapses, as we found no evidence for changes in excitatory input to mutant PV cells or in inhibitory connections between mutant PV cells. These results, along with those of other investigators 8 , identify an essential link between PTEN signalling and the establishment of somatic inhibition. PV-Cre þ / À /Ai9 þ / À /Pten loxP þ/ À mice. To genetically label PV-positive neurons, PV-IRES-cre knock-in female mice (Jackson Laboratories, stock#008069, generated by S Arber, FMI 27 ) were crossed with male TdTomato reporter knock-in mice (Jackson Laboratories, stock #007905, 'Ai9', generated by H Zeng, Allen Brain Institute 28 ). Offspring were hemizygous for both transgenes. These mice were then back-crossed with PV-Cre mice to generate offspring that were homozygous for PV-Cre and hemizygous for Ai9. Using separate breeders, we also generated mice that were hemizygous for both Ai9 and a floxed Pten 29 . PV-Cre þ / þ /Ai9 þ / À mice were then crossed with Ai9 þ / À /Pten loxP þ / À mice to generate offspring hemizygous for Cre, Ai9 and Pten in PV cells. Littermate controls were hemizygous for Cre and Ai9 in PV cells, but expressed both copies of Pten.
ChR2-assisted circuit mapping. Briefly, whole-cell recordings were obtained from pyramidal neurons in layer 2/3 of the primary visual cortex of PV-Cre þ / À / Ai32 þ / À /Pten loxP þ / À mice and littermate controls. Spatial maps of PV connectivity strength to each patched pyramidal neuron were derived by systematically stimulating ChR2-expressing PV cells at 256 different sites arranged in a 16 Â 16 matrix spanning all cortical layers 31 . Spatially restricted optogenetic activating of ChR2-expressing PV cells was accomplished using a 473 nm blue laser (30 mW, 0.25 ms). Voltage and current clamp controls were performed to normalize light stimulation protocols between slices and across animals. During optogenetic stimulation experiments, the ionotropic glutamate receptor antagonists (10 mM CNQX and 5 mM CCP) were added to the bath solution to block excitatory synaptic input (GABAergic transmission is unaffected) and avoid any potential dis-inhibition effects. Because interneurons can also be connected by electrical synapses, we blocked gap junctions using 100 mM carbenoxolone. Whole-cell voltage-clamp recordings were made from the recorded pyramidal neurons to measure photoactivation-evoked inhibitory postsynaptic currents at the empirically determined holding potential at þ 5 mV in voltage clamp mode with cesium-containing internal solution.
We validated the effectiveness of our approach as follows: following sufficiently high ChR2 expression in the cortex of the cross-bred pups (PV-Cre:Ai32), living brain slices were prepared and whole-cell recordings were obtained from ChR2/YFP-expressing PV cells that exhibited robust photoactivation-evoked spikes to repeated laser flashes (473 nm, 30 mW). To map their local functional outputs, whole-cell recordings from targeted excitatory and inhibitory neurons were combined with focal photoactivation (laser spot diameter, B50 mm) of ChR2 þ inhibitory neurons to assess functional connectivity and synaptic properties. As ChR2-expressing PV cells only fire action potentials to photoactivation (0.25 ms, 30 mW) close to the cell body, we found that ChR2 photoactivation had sufficient spatial precision for laminar circuit mapping.
For ChR2-evoked PV þ synaptic input analysis, we used the 3 'hit' criterion, and only map locations with three detected inhibitory postsynaptic currents (IPSCs) were selected for further map construction (Supplementary Fig. 1). The IPSC peaks were detected within the window of between 3 and 50 ms to each laser flash, with a threshold determined by background spontaneous IPSCs. The average IPSC input amplitude/strength of each stimulation site was measured by normalizing the sum of individual IPSCs from each photostimulation site by the analysis durations; this average integrated value was expressed in pA. We quantified the total PV þ IPSC inputs (total current) for the recorded individual cells by summing the input values across all map locations. The individual maps from different neurons were aligned to generate group averaged maps, using the 4 Â bright field images that were acquired during experiments to guide and register photostimulation sites. We used laminar cytoarchitectonic landmarks as well as the pial surface and white matter. These routine processes have been described in ref. 32.
Paired patch recordings. Brains were dissected from adult (P35-56) male and female mice. Three hundred and fifty micrometre coronal cortical slices were cut containing the visual cortex. All recordings were done at 33-35°C. Recordings were made in L2/3 at depths greater than 50 mm from the surface of each slice. PV cells were identified by their td-Tomato expression as well as their unique firing pattern, characterized by narrow action potentials, sharp after-hyperpolarization and fast firing rate. Pyramidal neurons were identified by their lack of td-Tomato expression, as well as their accommodating and slower firing pattern. Only cells with access resistance o25 MO were included in the subsequent analysis. Initial assessment of firing pattern was done in current clamp (IC) at resting membrane potential with current injections at 50 pA increments. Assessment of connectivity between pairs was done with the presynaptic cell in current clamp (VC) and the putative postsynaptic cell in VC.
Fluorescent-activated cell sorting. Fluorescent-activated cell sorting was performed at the UCLA Flow Cytometry Core using a FacsARIA cell sorter. In brief, 500 mm thick coronal slices containing the visual cortex were cut in standard artificial cerebrospinal fluid (ACSF). A Worthington Biochemical Papain Dissociation kit was used to dissolve the extracellular matrix to obtain individual cells, which were then filtered through a 70 mm diameter mesh and labelled with the dead cell stain 7AAD. Cells were laser sorted for peridin-chlorophyll protein complex (PerCP) signals (7AAD) and phycoerythrin (PE) signals (td-Tomato). Cells from wild-type animals with no fluorescent proteins were used to calibrate PerCP and PE signals each time. Cells with high PE signals and low PerCP signals were collected. We looked to obtain greater than 60,000 live PV cells per sorting session.
RNA amplification. RNA from collected PV cells was extracted with the Qiagen RNeasy Mini kit and then stored at À 80°C. RNA quality and quantity were assessed with the Bioanalyzer (Agilent). Only samples with RNA Integrity Number 48 were subsequently used. cRNA was then generated and amplified with the NuGEN Ultra Low Mass kit (Ovation).
Microarray. Microarray experiments were performed at the UCLA Neuroscience Genomics Core. cRNA levels were measured using Illumina mouse arrays (MouseWG-6 v2.0 BeadChip). Identification of differential gene expression was accomplished by first normalizing the gene expression of each sample and then comparing the gene expression of Pten þ / À PV cells with that of wild-type (PV-Cre: Ai9) brains. Differential expression analysis was determined by calculating the log2 fold change (logFC), defined as the ratio of intensities between the two experimental conditions under comparison. For all comparisons, a gene was considered differentially expressed if it had a corrected P-valuer0.001, and a logFCZ0.5 (Z 1.4-fold increased or decreased expression).
qrt-PCR. Real-time quantitative PCR of the same cRNA libraries was used to validate hits (here, Ephb4). Approximately 300-500 ng of RNA from the amplified control and PTEN þ / À samples was set aside prior to the microarray experiments. Thus, qrtPCR was performed on the same samples used in the microarray experiments. Total RNA from each sample was treated with DNAse 1 and converted into cDNA. Assays were performed in triplicate and analysed using an ABI 7700 instrument. The fold change was calculated using both standard curve analysis and the Pfaffl method.
Immunostaining. P16-P42 mice were perfused with 1 Â PBS followed by 4% PFA. Brains were removed and stored in 4% PFA overnight and transferred to a 30% sucrose solution with 0.05% NaN3 prior to cryosectioning. Brains were cryosectioned in the sagittal or coronal plane through primary visual cortex at 50 mm thickness per section. Free-floating sections were washed three times with 1 Â PBS prior to being placed in a 2 N HCl solution for 17 min for antigen retrieval of EphB4. They were subsequently washed three times in 1 Â PBS and blocked with 15% Normal Goat Serum in 0.5% Triton for 1.5 h at room temperature. The sections were then placed in a solution containing a monoclonal mouse anti-EphB4 primary antibody (1:50, Invitrogen) with 0.5% Triton overnight on a shaker in 4°C. The following day, slices were washed and incubated with an Alexa488conjugated goat anti-mouse secondary antibody (1:1000, Invitrogen) for 2.5 h at room temperature on a shaker. AAV-EphB4 virus injection. Through a 2.5 mm craniotomy centred on primary visual cortex, we made 3-4 injections of AAV2/1-e/Syn.FlexON.mCherry-T2AmEphB4 (1.2 Â 10^13 GC/ml). Injection sites were roughly 200 mm apart (forming a triangle with three injection sites or a square with four injection sites). Thin walled glass pipettes were pulled to have a long tapered tip with an opening that is o1 mm wide. The glass pipette was filled with B2 ml of virus and the tip of the pipette was gently broken using a Kimwipe to allow a small drop of virus to push through. The pipette was lowered to a depth of 300 mm below the surface of the cortex. Using a pressure of 15-20 psi, 1 msec puffs of virus were given 40-50 times with 2 s separating each puff. Then, the pipette was raised 50 mm towards the surface of the cortex and again 30-40 puffs were injected. This was repeated for every 50 mm up to 100 mm below the pial surface. Mice were euthanized 3 weeks post injection to allow sufficient time for EphB4 expression. Cortical slices prepared and ChR2-assisted circuit mapping was conducted as above.
Statistical analyses and graphs. All data are reported as mean±standard error of the mean, except for data in Fig. 3c where the probability of connection is reported between patched pairs of cells. Significance in Fig. 3c was determined using Fisher's exact test. For the remaining data sets, when comparing two independent groups, a Wilcoxon rank-sum test was used. In the case more than two independent groups were compared a Kruskal-Wallis one-way ANOVA was used and followed by post-hoc comparisons when justified (alpha set to 0.05) using Wicoxon rank-sum tests. In the cases where more than two groups were compared and the groups were not independent, a repeated measures ANOVA (Friedman's test) was used. In all cases, sample size is defined as cell number.
Data availability. All analyses were conducted in Matlab (Mathworks, Natick, MA). The code and data are available on request from the authors