Microglia contact induces synapse formation in developing somatosensory cortex

Microglia are the immune cells of the central nervous system that play important roles in brain pathologies. Microglia also help shape neuronal circuits during development, via phagocytosing weak synapses and regulating neurogenesis. Using in vivo multiphoton imaging of layer 2/3 pyramidal neurons in the developing somatosensory cortex, we demonstrate here that microglial contact with dendrites directly induces filopodia formation. This filopodia formation occurs only around postnatal day 8–10, a period of intense synaptogenesis and when microglia have an activated phenotype. Filopodia formation is preceded by contact-induced Ca2+ transients and actin accumulation. Inhibition of microglia by genetic ablation decreases subsequent spine density, functional excitatory synapses and reduces the relative connectivity from layer 4 neurons. Our data provide the direct demonstration of microglial-induced spine formation and provide further insights into immune system regulation of neuronal circuit development, with potential implications for developmental disorders of immune and brain dysfunction.

I ncreasing evidences point to the important influence immune status and immune molecules have on brain development and function 1 . Maternal infection is a considerable risk factor for the neurodevelopmental disorders such as autism and schizophrenia 2 . Microglia are important components of such immune-brain interactions and defects in microglial genes or signalling molecules are associated with disorders of neuronal circuits and brain behaviour 1, [3][4][5] . As the immune cells of the central nervous system, microglia respond to various brain pathologies, adopt an activated phenotype and secrete a range of cytokines and neuro-trophic factors to modify disease progression [6][7][8][9] . However, microglia also play important roles for normal brain physiology, both in development and in the mature nervous system 10 . In the developing brain, they actively participate in sculpting neuronal circuits 11 . This includes promoting neuronal apoptosis 12 , regulating neurogenesis [13][14][15] and eliminating less active synapse structures identified for phagocytosis via expression of traditional immune complement molecules 10,[16][17][18][19][20] . More recent studies have suggested microglia have bidirectional effects on synapses and neuronal circuits by promoting synapse formation 21 . Parkhurst et al. 22 demonstrated that selective ablation of microglia reduced spine formation as observed by time-lapse imaging in vivo, both during later development and in response to different learning tasks. Selective depletion of brain-derived neurotrophic factor (BDNF) from microglia could largely replicate this observation. A role for microglia in enhancing spine density has also been suggested from in vitro studies. Addition of microglial cells to hippocampal neuronal cultures could increase the number of dendritic spines via release of interleukin-10 (IL-10) 23 , whereas the effect of estradiol to promote increased spine density in cultured pre-optic neurons depends on the presence and function of microglia 24 . Release of cytokines such as IL-10 by microglia is typically associated with an activated microglia phenotype, such as occurs when microglia begin to populate the cortex during early postnatal development. However, (1) how microglia may contribute to functional synapse formation; (2) whether this is a direct result of microglia-neuron interactions; and (3) the functional significance of putative microglia-induced spine formation in development is all unclear. To address these questions, we undertook in this study in vivo two-photon time-lapse imaging of neurons and microglia in the developing somatosensory cortex. We observed that microglia directly contacted dendrites and initiated filopodia formation. This effect of microglia was restricted to the period of intense spinogenesis that occurs in the first two postnatal weeks, a time when microglia have an activated morphological phenotype. We propose that filopodia formation by microglia promotes the maturation of specific neuronal circuit connections through the formation of functional mature synapses.

Results
Microglia promote filopodia formation during development.
Signalling events underlying microglial-induced filopodia. Filopodia formation on dendrites is known to be triggered by synaptic activity, subsequent Ca 2 þ elevation and activation of Ca 2 þ -dependent enzymes 27 . We therefore hypothesized that microglia contacts may trigger Ca 2 þ transients, as have been previously seen in dendrites surrounded by microglia 28 . To test this possibility, we in utero electroporated the genetically encoded green calcium indicator (GCaMP6m), along with tdTomato, into L2/3 neurons in Iba1-EGFP mice. Subsequent in vivo imaging in awake P8-P10 mice detected local neuronal Ca 2 þ responses in dendrites following microglial contacts that were frequently followed by filopodia formation (Fig. 2a-c). Such local Ca 2 þ responses were not detected at these contact sites before the microglia contact and typically disappeared following the contact. Local Ca 2 þ responses were also not observed at dendritic locations 20 mm adjacent to the contact point during the same imaging periods ( Fig. 2d; 20 mm n ¼ 3, before contact n ¼ 4, during contact n ¼ 11 and after contact n ¼ 7 dendrites in 8 mice). The fliopodia formation rate was significantly higher in dendrites in which a contact-induced Ca 2 þ response was observed ( Fig. 2e; 86.1 ± 8.4% versus 16.0±7.1% in dendrites without a Ca 2 þ response; unpaired t-test, exact P-value is P ¼ 0.00006; exact t-value is t(7) ¼ 4.03; with Ca 2 þ : n ¼ 7 animals, without Ca 2 þ : n ¼ 16 animals). The lifetime of filopodia formed on dendrite in which a microglia-induced Ca 2 þ response ( Fig. 2f; 20.7±2.7 min, awake mice) was longer than filopodia formed on dendrites without a Ca 2 þ response (7.1±3.6 min, awake mice) or in dendrites in anaesthetized mice (7.9 ± 1.2 min; one-way ANOVA, post-hoc Bonfferoni test, exact P-values are P ¼ 0.002 (awake Ca 2 þ , awake no Ca 2 þ ), P ¼ 0.0003 (awake Ca 2 þ , anaesthesia), P ¼ 1 (awake no Ca 2 þ , anaesthesia), exact F-value is F(2, 25) ¼ 12; awake Ca 2 þ : n ¼ 7 dendrites, awake no Ca 2 þ : n ¼ 5 dendrites, anaesthesia: n ¼ 16 dendrites). To further probe the signalling mechanisms linking microglia contact and dendritic filopodia formation, we transfected cortical slice cultures from Iba1-EGFP mice with lifeact-mCherry. Microglia contact was followed by accumulation of F-actin in dendrites and subsequent filopodia formation, consistent with the known role of actin in filopodia formation 29,30 (Fig. 2g,h and Supplementary Movie 2). The relative fluorescent intensity at microglia contact sites (1.4±0.04) was significantly greater than the basal intensity at dendritic regions adjacent (by 10 mm) to the contact site ( Fig. 2i; 1.1±0.01; paired t-test, exact P-value is P ¼ 0.0003; exact t-value is t(5) ¼ 8.87; n ¼ 6 sites). Together, this data suggests that microglia initiates filopodia formation via Ca 2 þ -induced actin accumulation, with microglial contact-induced filopodia being more stable.
Microglia ablation selectively reduces synapses from L4. In a final set of experiments, we probed the possible functional consequences of microglia contact-induced filopodia formation, hypothesizing that filopodia formed by microglia may predominantly contribute to the formation of specific connections during this developmental period. To test this hypothesis, we recorded monosynaptic EPSCs from L2/3 pyramidal cells above a cortical barrel using whole-cell patch clamp recordings in brain slices from control (Iba1-tTA) and microglia ablated (Iba1-tTA::tetO-DTA) P23-P25 mice, by which time barrel cortical circuits have matured. EPSCs were evoked by glutamate uncaging with laser scanning photostimulation across L2-L5 (Fig. 5a). Such photostimulation-evoked EPSCs are largely due to single action potentials in individual presynaptic neurons 32 and thus can be used to quantify synaptic inputs originating from the different cortical layer neurons. In both control and microglia ablated mouse slices, L2/3 neurons received inputs from L2-L5. In control mice, L2/3 neurons received stronger inputs from L4 compared with other layers, consistent with previous data 33 (Fig. 5b,c; unpaired t-test between corresponding locations in control and DTA mice; exact P-value is P ¼ 0.031 or P ¼ 0.044; control n ¼ 9 from 4 animals, DTA n ¼ 5 from 3 animals). However, in slices from microglia-ablated DTA mice, the largest synaptic response did not arise from L4, resulting in an averaged spatial distribution of synaptic inputs that was relatively uniform across the cortical layers (Fig. 5b,c). The averaged strength of synaptic inputs originating from L4 was significantly lower in DTA mice, compared with control mice, whereas inputs from L2/3 or L5 were not significantly different ( Fig. 5d; versus control rats; unpaired t-test; exact P-value is P ¼ 0.032). The averaged amplitude of spontaneous synaptic currents was also not different in DTA and control mice. These results indicate that microglia ablation selectively reduced excitatory synaptic connections from L4 to L2/3, suggesting that microglia-induced filopodia formation specifically promotes the establishment of feedforward synaptic connections from L4 to L2/3.

Discussion
Establishing appropriate synaptic connections during development is generally achieved by the balance between synaptogenesis and experience-dependent synapse pruning and maturation.
In the mouse somatosensory cortex, an intense period of synaptogenesis (with a rapid E7-fold increase in synapses) begins around P8, with synapse numbers peaking around P32, before declining (byE20%) to adult levels 26 . Synapses develop from filopodia, labile, postsynaptic protrusions that may stabilize and develop into mature spines 22,25,34,35 when they connect onto presynaptic terminals. Microglia contribute to a number of aspects of neural circuit development, including to help define the neuronal population via regulation of apoptosis and neurogenesis and by phagocytosing excess synapses during synaptic pruning 11,15,17 . Our studies demonstrate a new physiological role for microglia in promoting synapses during a transient Upper left panel shows the same dendrite (red) and microglia (green) contact as in a and indicates the ROIs used to quantify the spatial and temporal aspects of Ca 2 þ elevation shown in the corresponding five traces in the right panel. The blue shading indicates the duration of the microglia-dendrite contact, which induced brief Ca 2 þ transients (indicated by asterisk) in ROIs 2 and 3, and a smaller sustained elevation in the microglial process (ROI 1). (c) Z-stacked images of the same dendrite before the Ca 2 þ imaging and transients were observed and following Ca 2 þ imaging. Arrowheads indicate microglia contacted point. Scale bar, 5 mm. (d) Local dendritic Ca 2 þ responses were exclusively associated with microglial contact. At dendritic contact sites, Ca 2 þ transients were absent before the contact, increased in frequency during the contact and typically decreased back towards zero after the contact. At dendritic locations 20 mm adjacent to the contact points, no localized Ca 2 þ responses were observed throughout the same imaging periods (20 mm n ¼ 3, before contact n ¼ 4, during contact n ¼ 11, after contact n ¼ 7 dendrites in 8 mice). (e) Averaged filopodia formation rates in dendrites in which microglial contact was associated with a Ca 2 þ transient compared with those dendrites in which there was no Ca 2 þ response-associated microglia contact (unpaired t-test, error bars are mean ±s.e.m., exact P-value is P ¼ 0.00006; with Ca 2 þ : n ¼ 7 animals, without Ca 2 þ : n ¼ 16 animals). (f) Lifetime of filopodia formed after microglial contact to dendrites in which a Ca 2 þ response was observed and in those in which microglial contact was not associated with a dendritic Ca 2 þ transient. These experiments were conducted in awake mice and are compared with lifetimes of filopodia arising after microglial contact in dendrites of anaesthetized mice (one-way ANOVA, post-hoc Bonfferoni test, error bars are mean ± s.e.m., exact P-values are P ¼ 0.002 (awake Ca 2 þ , awake no Ca 2 þ ), P ¼ 0.0003 (awake Ca 2 þ , anaesthesia), P ¼ 1 (awake no Ca 2 þ , anaesthesia); awake Ca 2 þ : n ¼ 7, awake no Ca 2 þ : n ¼ 5, anaesthesia: n ¼ Black arrow indicates onset of filopodia formation. Right: averaged fluorescent intensity values from 5 to 10 min after microglia contact at contact sites was significantly increased compared with that at adjacent sites 10 mm from the microglia contact points during the same time period (right) (paired t-test, error bars are mean ± s.e., exact P-value is P ¼ 0.0003; n ¼ 6 sites). period of brain development, as contacts between microglial processes and dendritic shafts could trigger filopodia formation (Fig. 6a). Consistent with known mechanisms of filopodia formation 34,36 , microglial contacts could also trigger localized Ca 2 þ transients and recruitment of actin. Although filopodia formation also occurred in the absence of microglia contact, the filopodia formation rate at microglia contact sites on a dendrite was greater than that observed at non-contact sites. At least some of these microglia-induced filopodia appeared to develop into functional excitatory synapses, as ablating microglia reduced the subsequent functional spine density and frequency of excitatory synaptic transmission. Microglia may potentially be attracted to dendritic locations where the early stages of filopodia formation are already in place, or may initiate de novo filopodia formation. The sequence of events where Ca 2 þ transients only occur during contacts and are directly followed by actin accumulation and filopodia formation seem to favour the latter hypothesis. Some of the filopodia induced by microglia exhibited an increased stability, which may also contribute to the microglia-dependent increase in mature spines and functional synapses, and this was particularly evident in awake mice when the microglia-dendrite contact was followed by a localized Ca 2 þ transient. Interestingly, the lifetime of microglia-induced filopodia in anaesthetized mice was similar to the lifetime of non-contact filopodia. Anaesthetics are known to affect filopodia formation and stability 36 , and may have occluded potential differences in filopodia lifetime in this set of experiments.
A striking observation in our study was the transient nature of the microglia-induced filopodia formation. During early development, microglia populate the brain in an activated, amoeboid phenotype before adopting the more mature ramified surveying morphology 31,37 . Microglia proliferate into the barrel cortex from P5 in the amoeboid state and gradually change their phenotype to a more ramified state 38 . We similarly observed a morphological change between P8 and adults, as well as clear transitions in both cell soma size and Iba1 mRNA expression patterns between P8-P10 and P12-P14 microglia. The microglia contact-induced filopodia formation was only observed in P8-P10 mice, corresponding to the amoeboid, activated state. Consistently, inhibition of microglial activation with Mino abolished microglia-induced filopodia formation. The effects of Mino may be mediated through a change in activation phenotype, which could also include a decrease in the release of various soluble factors such as IL-10, tumour necrosis factor-a or BDNF. Activated microglia directly opposed to, or in contact with, dendrites may physically or via surface receptors promote functional spine formation. Regardless, this novel facet of microglia-induced neuronal plasticity we observe here is restricted to the activated, amoeboid phenotype. Phagocytosis of surplus synapses in developing lateral geniculate circuits also (f) Cortical Iba1 mRNA levels were significantly decreased over age, from P9 through to P14 or P30 mice (one-way ANOVA, post-hoc Bonfferoni test, error bars are mean ± s.e.; exact P-values are P ¼ 0.032(P9, P14), occurs early in development (around the first postnatal week) by the amoeboid microglia phenotype 17 . Experience-dependent pruning of synapses in visual cortex is evident at 4 weeks postnatal, when microglia have a mature ramified phenotype 39 .
Microglial BDNF-dependent spine formation following motor learning occurs in the healthy adult rodent motor cortex when microglia are also of the quiescent, ramified morphology.
Clearly, the precise functional consequences of microglia-   neuron contacts on circuit plasticity-and whether this tends towards synapse elimination, synapse surveillance or spine formation-are likely to depend on the particular brain region and the specific activity or developmental pattern of synaptic plasticity occurring at that time. Our data emphasize the repertoire of actions microglia can participate in across the developmental time scales. The different synaptic inputs to the cortex arrive during different developmental time windows and display corresponding different critical periods of activity or experience-dependent synaptic plasticity 40 . Thalamic inputs to L4/L5 neurons arise initially into the somatosensory barrel cortex (around P0) and establish the columnar or barrel organization. Local cortical inputs to L2/3 neurons subsequently arise from recurrent connections from other L2/3 neurons and from feedforward projections from L4 neurons. Although there is some overlap, the L4 connections and the corresponding critical periods of plasticity occur earlier (around P10-P14) than those from L2/3 projections (P13-P16) 40 . By extension, filopodia that appear around P8-P10 may preferentially mature into L4-derived synapses, whereas filopodia at later stages will preferentially mature into L2/3derived synapses. Consistent with this proposal, genetic and transient ablation of approximately half the microglial population around P8 to inhibit microglia-induced filopodia formation selectively weakened synaptic inputs to L2/3 neurons from L4 stimulation. It appears that the cortical circuits use the capacity of amoeboid microglia at this specific developmental stage to increase particular synapses being formed at that time (Fig. 6b).
Additional cellular mechanisms may exist to facilitate synapse formation from other inputs at other specific developmental time points.
It is unclear at this stage which signalling mechanisms may trigger the spatially restricted increase in filopodia formation rate induced by microglial contact. Numerous mechanisms and molecules have been identified to promote spine formation 41 . For example, microglia may increase neuronal activity and/or NMDA (N-methyl-D-aspartate)-mediated Ca 2 þ transients by directly releasing glutamate or via enhanced release of glutamate from approaching presynaptic neurons 24 . Receptor-mediated direct physical coupling between microglia and dendrites may be involved, similar to that proposed for synapse elimination. In the developing lateral geniculate nucleus, microglia phagocytosis of synapses involves interactions between immune complement receptor molecules C3 and C3R 17 , whose expression depends on neuronal activity. In the hippocampus, fractalkine receptorligand interactions contributed to synapse phagocytosis 16 . Other complementary molecules such as telencephalin, nectin, IL-1 receptor accessory protein-like 1 and C1q family have been reported to be involved in synapse formation [42][43][44][45][46] . Neural cell adhesion molecule (NCAM2) is expressed in both neurons and microglia, and could trigger intracellular Ca 2 þ elevation that may subsequently recruit actin and elicit filopodia formation 47,48 . Microglia secrete a variety of soluble factors, including cytokines, neurotrophic factors and neurotransmitters, with this pattern changing dependent on different phenotypes from quiescent to challenged 7 . For motor learning-induced spine formation, release of BDNF by microglia is critical 22 , whereas microglial release of IL-10 enhanced spine formation in cultured neurons 23 . Microglia can also release tumour necrosis factor-a, which can signal via various receptors and pathways to increase expression of N-cadherin to increase spine density 49 . Our characterization of the precise time course and transient nature of microglia-induced filopodia formation should assist experiments aimed at identifying the underlying signalling pathways.
A number of psychiatric brain disorders, such as schizophrenia and autism, involve disruptions in synapse number, morphology or function, with a pathogenesis that is believed to initiate with synapse development 50 . In autism spectrum disorders in particular, there is a surplus of spines in the cortex during early development in both human and mouse models, and dysfunction in the turnover rates of cortical spines is a common feature in a number of mouse models of autism 51 . Given the association between maternal infection and autism and schizophrenia incidence, microglia are possible pathogenic candidates. Our studies here provide an additional aspect of microglia-neuron interactions-developmental spine formation and specific neuronal circuit connections-and further investigations into the underlying mechanisms of microglia-neuron interactions may shed light on the pathophysiology of brain diseases and provide potential strategies for restoring synapse function.

Methods
All animal experiments were approved by the Animal Research Committee of the National Institutes of Natural Sciences.
Animals and microglia inhibition. To visualize microglia, we used the Iba1-EGFP transgenic mouse, which expresses EGFP under the control of the Iba1 promoter, which is specific for microglia and macrophages 52 . For microglia ablation experiments, double transgenic mice were generated by crossing Iba1-tTA mice 53 and tetO-DTA mice 54 . Withdrawal of Dox in the feed of these mice leads to selective expression of the DTA in microglia. All transgenic mice were derived from the C57BL/6J strain. Transgene induction was inhibited by rearing mice with standard chow containing Dox 0.1 g kg À 1 . For transgene induction, Dox-laced chow was replaced with Dox-free standard chow from P5 through to P11 (for analysis of spine density). Both male and female mice were used for all experiments. Mice (single mothers and their litter) were housed on a 12-h light/dark cycle. Mino hydrochloride (M9511-1G, Sigma-Aldrich, Tokyo, Japan) was used to pharmacologically inhibit microglia and injected daily (i.p., 75 mg kg À 1 ) from P5 through to the experimental day (P8-P10, for in vivo imaging of microglia-dendrite contacts; P11 for analysis of spine density). Control mice received i.p. saline injections with the same dosing schedule.
In utero electroporation to visualize neurons. To visualize L2/3 pyramidal neurons, we performed in utero electroporation of embryos at embryonic days (E) 14 or E15 (ref. 55). At this age, intraventricular injection and electroporation only transfects neurons migrating into the cortex. Under anaesthesia (1.7% isoflurane), the uterus was exposed and B1.5 ml of plasmid solution was injected into the lateral ventricle of each embryo using a glass pipette (tip diameter: 50-100 mm). The head of a single embryo was then placed between tweezer-type electrodes with 5 mm tip diameters (CUY650-P5; NEPA Gene, Chiba, Japan) and square electric pulses (35 V; 50 ms) were applied to the electrodes 5 times, at 950 ms intervals, using an electroporator (CUY21E; NEPA Gene). Each embryo was quickly returned to the abdomen following the electrical stimulus and once all embryos were electroporated the peritoneal membrane was sutured back together. The mother's skin was bound by clips (12032-07, Muromachi Kikai, Tokyo, Japan) to avoid reopening the surgical wound.
In vivo two-photon imaging. Electroporated Iba1-EGFP mice (P8-P10) were anaesthetized with urethane (1.7 g per kg body weight, i.p. and atropine, 0.4 mg kg À 1 , i.p.). Surgery and imaging were performed on a warming plate. After removal of the scalp, a cranial window (1.6 mm in diameter) was made over the primary somatosensory barrel cortex (1 mm posterior from bregma and 2.5 mm lateral from the midline). A cover glass was placed over the cranial window and fixed with adhesive glue (Aron Alpha, Konishi, Osaka, Japan) and dental cement (Quick Resin, SHOFU, Kyoto, Japan). A custom-made imaging chamber surrounded the cranial window and was perfused with warm water (32-34°C) during imaging.
Two-photon imaging was performed with a Ti:sapphire laser (Mai Tai HP, Spectra-Physics, Tokyo, Japan) operating at 960 nm wavelength. A laser scanning system (Olympus FLUOVIEW, Olympus, Tokyo, Japan) and an upright microscope (BX61WI, Olympus) with a water-immersion objective ( Â 25, 1.05 numerical aperture (NA); Olympus) was used for image acquisition. Fluorescence was separated by a 570 nm dichroic mirror with 495-550 nm (green channel: for EGFP fluorescence detection) and 570-630 nm (red channel: for DsRed-express or tdTomato fluorescence detection) emission filters, and detected by photomultipliers. For time-lapse imaging, Z-stack images (512 Â 512 pixels, 0.099 mm per pixel, 0.5 mm Z-steps) were taken every 5 min for between 30 min and 2 h at a depth of 45-250 mm from the cortical surface. For short interval imaging, XYt images were taken once every 1.6 s, or once per minute, for 27 min.
For dendritic Ca 2 þ imaging in awake mice, cranial window (1.5 mm) surgery was performed under isoflurane anaesthesia (1.5% isoflurane in pure O 2 ). After an hour recovery, calcium imaging was performed in the awake state. XYt images were taken at 8 Hz for 30 min (256 Â 128 pixels, 0.25 mm per pixel). Two-photon imaging was performed with a Ti:sapphire laser (Mai Tai HP, Spectra-Physics) operating at 940 nm wavelength. To detect calcium transients, we defined a small region of interest (ROI) (2 Â 2 mm) along the dendritic shaft and calculated an average basal Ca 2 þ fluorescence. A significant Ca 2 þ elevation response was defined if the peak amplitude of DF/F was greater than three times the s.d. of the basal Ca 2 þ level. A localized Ca 2 þ response was defined as one that was restricted to just a small portion of the dendrite (o12 mm), in contrast to transients associated with back-propagating action potentials.
Analysis of in vivo microglia and neuron interactions. Image stacks were visually inspected in ImageJ (National Institutes of Health, USA), to determine regions of co-localization of GFP-labelled microglia and RFP-labelled dendrites, and then subsequently further examined in greater detail across all image time points and Z-sections using ImageJ, to confirm real contacts. To reduce noise, images were filtered with a 3 Â 3 pixel median filter after background subtraction. Contacts were determined by overlap between the red (dendrite) and the green (microglia) channel, after independently determining the baseline thresholds for each channel independently. Identified overlaps were defined as contacts only if the red and the green channel had overlapping pixels in at least two Z-sections and if the overlap extended beyond twice the spatial resolution. Dendrites that retracted during time-lapse imaging were excluded from analysis. Dendritic spines or filopodia were identified as protrusions that were 40.4 mm from the dendrite border. A spine was defined as a protrusion with a head structure, whereas filopodia lacked head structures. This categorization included stubby structures as filopodia. This is appropriate as these types of 'spines' are unstable during development and frequently disappear 56 . Contact-induced filopodia formation rates were defined as the incidence in which a filopodia was formed at a microglia contact region. The contact region on a dendrite was defined as a 2 mm zone (±1 mm from the estimated centre of the microglia-dendrite contact) and a protrusions had to appear within 10 min of the onset of the microglia-dendrite contact to be counted as being formed by microglia (it is noteworthy that Fig. 1c includes all protrusions that appeared within 20 min after microglia contact for comparison). Between 1 and 27 microglia-dendrite contacts were observed in different dendrites in each animal and the incidence of a contact-induced filopodia were averaged to give a single formation rate for each animal. The formation rate of protrusions at dendritic regions 10 mm ( ± 1 mm) and 20 mm (±1 mm) adjacent to the microglia contact region were calculated as a measure of the control (or putative non-contact) filopodia formation rates. This analysis counted the incidence of a protrusion forming at these non-contact regions along the same dendrites at any time during the same imaging period over which contact-induced formation rates were measured. To quantify the formation rate across a single dendrite, the number of any protrusions appearing along the whole region of a dendrite during the imaging was counted (excluding those at the contact region) and normalized by the dendritic length. Dendrite length was expressed in multiples of 2 mm blocks (for example, 10 mm ¼ 5), to compare with the 2 mm length unit used to define contact and non-contact regions. It should be noted that formation 'rate' as used in the study is related to the incidence of filopodia formation per unit length, rather than over a given time period.
Image analysis for microglia morphology. Iba1-EGFP mice were used to visualize microglia, using thick (100 mm) brain slices (as described above) to enable measurement of the complete extent of microglial processes. Confocal imaging was performed with a multi argon laser operating at a wavelength of 488 nm. A laser scanning system (Nikon A1, Nikon, Tokyo, Japan) and an inverted microscope with water-immersion objective ( Â 40, 0.95 NA, Nikon) was used for image acquisition. Fluorescence was separated by a dichroic mirror with 495-550 nm (green channel: for EGFP fluorescence detection) emission filters and detected by photomultipliers. Z-stack images (512 Â 512 pixels, 0.198 mm per pixel, 0.5 mm Z-step) were used for morphological analysis.
Image processing was performed using ImageJ software. Maximum intensity projections of Z-series stacks were created. The extent of microglia process ramification was quantified by measuring the area circumscribed by the distal ends of microglial processes using the segmented line tool in ImageJ. Microglia size was quantified by circumscribing the cell body area using an intensity threshold tool in MetaMorph (Molecular Devices, Tokyo, Japan). The number of microglia primary processes was manually counted, with primary processes defined as those starting from the microglia soma.
Image analysis for spine density. In utero electroporated mice were used for visualization of L2/3 pyramidal neuron dendritic spines. Brains were fixed and sectioned (100 mm) as described above. Confocal imaging was performed with 10 mW 561 nm solid laser. A laser scanning system (Nikon A1, Nikon) and an inverted microscope with a water-immersion objective ( Â 60, 1.2 NA, Nikon) was used for image acquisition. Fluorescence was again separated by a dichroic mirror with 572-700 nm emission filter (red channel: for tdTomato fluorescence detection) and detected by photomultipliers. Z-stack images (512 Â 512 pixels, 0.08 mm per pixel, 0.5 mm Z-step) were used to calculate spine density.
Dendritic spines were identified in a series of Z-stack images and counted using ImageJ software. Serial stack images were used to assist in the delineation of individual spines in dendritic regions where spine density was high. Serial Z-stacks of different optical sections were used to identify individual spines with greater certainty. For this analysis, all dendritic protrusions with a clearly recognizable stalk were counted as spines. Spine number was divided by the length of the dendritic segment to generate dendritic spine density, expressed as number per micrometre.
Slice culture and biolistic transfection. Cortical slice cultures were derived from 350 mm thick slices from P5 Iba1-EGFP mice and used to examine actin dynamics following microglia contact. After 1 day in vitro, cortical pyramidal neurons were transfected with a ballistic gene transfer gun using gold beads (4-6 mg) coated with 10 mg of the pCMV-lifeact-mCherry plasmid. At 4 days in vitro, slice cultures were mounted on a microscope stage for imaging and perfused with warmed (32-34°C) artificial cerebrospinal fluid (ACSF), containing 126 mM NaCl, 2 mM KCl, 2 mM CaCl 2 , 24 mM NaHCO 3 , 1.2 mM NaH 2 PO 4 , 1.3 mM MgSO 4 and 10 mM glucose. Two-photon imaging was used, with a 960 nm excitation wavelength, as described above for in vivo imaging. A water-immersion objective ( Â 25, 1.05 NA, Olympus) was used for image acquisition. Fluorescence was separated by a 570 nm dichroic mirror with 495-550 nm (green channel: for EGFP/microglia fluorescence detection) and 570-630 nm (red channel: for mCherry/actin in neuronal dendrites fluorescence detection) emission filters, and detected by photomultipliers. Actin accumulation images were acquired using real-time imaging at a single XY image plane acquired every 1.6 or 5.0 s for 27 or 30 min, respectively. Images were analysed using ImageJ. Microglia contact was defined as overlap of green fluorescence intensity within a (red) dendritic region, whereas segregation of lifeact-mCherry was defined as an increase of red fluorescence intensity within the microglia contact region in the dendrite. Filopodia formation was defined as the red fluorescence intensity protruding from pre-contact border of the dendrite. The background intensity of the image regions of interest in the absence of neuronal or microglia structures was subtracted from the intensity of the microglia and filopodia regions. The intensity of an adjacent region of the same dendrite, but without microglia contact or the presence of a filopodia or spines, was used as background for the dendritic aggregation of the lifeact-mCherry fluorescence signal. Fluorescence values were normalized to the maximum intensity of each signal. Fluorescent intensity was normalized using a value of 25% of the maximum intensity as baseline (F 0 ). To quantify contact induced actin aggregation, the relative intensity (F/F 0 ) was averaged between 5-10 min after the onset of microglia-dendrite contact and the F/F 0 at the contact region compared with that at a 10 mm adjacent region Quantitative reverse transcriptase-PCR for mRNA analysis. Mice were deeply anaesthetized with ketamine (0.13 mg g À 1 , i.p.) and xylazine (0.01 mg g À 1 , i.p.), transcardially perfused with 0.1 M PBS, which contained 5 mM EDTA and brains were then dissected. Total RNA was extracted from neocortical tissue samples using RNeasy Plus Mini Kit (74134, Qiagen), then reverse transcription was performed by Transcriptor First Strand cDNA Synthesis Kit (04896866001, Roche). Real-time PCR was performed using FastStart Essential DNA Green Master (06402712001, Roche Applied Science) on LightCycler 96 (05815916001, Roche Applied Science) with Iba1 exon primers (5 0 -TCCCAAATACAGCAATGATG AG-3 0 , 5 0 -GCATTCGCTTCAAGGACATAAT-3 0 ) and GAPDH primers (5 0 -AA TGCATCCTGCACCACCAAC-3 0 , 5 0 -TGGATGCAGGGATGATGTTCTG-3 0 ). Fold change was calculated using the DDCt method and normalized by the values obtained with saline-injected mice or P9 mice.
Measurements of spontaneous synaptic currents. Acute brain slices were prepared from Iba1-tTA::tetO-DTA or Iba1-tTA mice at P12 following anaesthesia with ketamine (0.13 mg g À 1 , i.p.) and xylazine (0.01 mg g À 1 , i.p.), and transcardial perfusion with oxygenated (95%O 2 /5%CO 2 ) slicing solution containing (in mM): 230 sucrose, 26 NaHCO 3 , 2 KCl, 1 MgCl 2 , 1 KH 2 PO 4 , 0.5 CaCl 2 and 10 glucose. Mice were decapitated and the brains were rapidly removed, and 350 mm thick coronal cortical slices were cut in cold slicing solution. Slices were stored in oxygenated ACSF (as described above for slice culture imaging) at 34°C for at least 45 min before being transferred to the recording chamber on the stage of an upright microscope and viewed with a Â 40 water-immersion objective. Slices were continuously perfused with oxygenated recording ACSF and recordings were obtained at room temperature (around 25°C). Whole-cell voltage-clamp recordings (at a holding potential of À 70 mV) were made from the somata of visually identified barrel cortex L2/3 pyramidal neurons. Patch pipettes (5)(6)(7)(8) were constructed from borosilicate glass capillaries and filled with an internal solution containing (mM): 9 CsCl, 130 CH 3 SO 3 Cs, 2 EGTA, 10 HEPES, 4 Mg-ATP and 0.4 Na-GTP, pH adjusted to 7.3 with Tris. To isolate mEPSCs, 0.3 mM tetrodotoxin and 10 mM SR95531 were continuously perfused once a stable recording was obtained. Only cells with R series %25 MO and R input^2 00 MO were included for analysis. No corrections for liquid junction potentials were applied and no series resistance compensation was used.
Laser uncaging photostimulation evoked excitatory postsynaptic currents. Coronal slices of barrel cortex (300 mm thick) were prepared from P23-P25 mice under deep anaesthesia with isoflurane and kept in normal artificial cerebrospinal fluid (ACSF) containing (in mM): 126 NaCl, 3 KCl, 1.3 MgSO 4 , 2.4 CaCl 2 , 1.2 NaH 2 PO 4 , 26 NaHCO 3 and 10 glucose at 33°C, as described previously 32 . For whole-cell recording, patch pipettes (4-6 MO) were filled with a solution containing (in mM): 130 K-gluconate, 8 KCl, 1 MgCl 2 , 0.6 EGTA, 10 HEPES, 3 MgATP, 0.5 Na 2 GTP and 10 Na-phosphocreatine (pH 7.3 with KOH). For analysis, we selected cells with a high seal resistance (41 GO) and a series resistance o25 MO. The recording of photostimulation-evoked EPSCs and analysis of the EPSCs was conducted as described previously 57 . Photostimulation was achieved by focal photolysis of Rubi-caged glutamate with 5 ms flashes of blue light (440 nm) from a diode laser. The light was focused on the slices through a Â 4, 0.16 NA microscope objective. Laser power was set to 5 mW at the specimen plane. This resulted in the generation of action potentials in neurons with cell bodies mostly within B100 mm of the centre of the illuminated spot. Photostimulation-evoked EPSCs were recorded from L2/3 pyramidal neurons within a single L4 barrel. Usually, the photostimulations were applied once every 5 s to each of 16 Â 20 sites surrounding the recorded cells in a quasi-random sequence. The maps of photostimulation sites were aligned to laminar borders in fixed and stained tissue (Fig. 5a), and each site was assigned a laminar identity. We measured the peak amplitude of all of the EPSCs occurring within 150 ms after the stimulation and constructed colour-coded, linearly interpolated maps using the total amplitude of the EPSCs at each stimulation site. To quantify laminar input strength, we also calculated the normalized strength of the excitatory inputs from each layer by the summed values from all the layers examined. The data were analysed using custom software written in Matlab.