Postnatal development of the microstructure of cortical GABAergic synapses and perineuronal nets requires sensory input

The brain synaptic circuitry is formed as a result of pre-defined genetic programs and sensory experience during postnatal development. Perineuronal nets ensheath synaptic boutons and control several crucial features of the synapse physiology. Formation of the perineuronal net microstructure during the brain development remains largely unstudied. Here we provide a detailed quantitative description of the 3-dimensional geometry of the synapse and the surrounding perineuronal net in the mouse somatosensory cortex layer IV. We compare the morphology of the synapse + perineuronal net complex in the adult brain formed under normal conditions or in the whisker shaving model of somatosensory deprivation. We demonstrate that the sensory deprivation causes flattening of the 3D PNN mesh geometry and reduction of the VGAT-positive cluster volume in presynaptic boutons. These results reveal a mechanism of the sensory input-dependent synapse morphogenesis during the brain development.


Introduction
Sculpturing of synaptic networks by sensory experience is currently recognized as a fundamental mechanism of the brain plasticity (Fawcett et al., 2019;Ferrer-Ferrer and Dityatev, 2018). This phenomenon of the experience-dependent synaptic rearrangements is comprised by a number of molecular processes both in the synapses and in the surrounding extracellular matrix (Dzyubenko et al., 2016).
Perineuronal net (PNN) is a highly structured subtype of extracellular matrix surrounding synaptic boutons on neuronal cell bodies and proximal dendrites in large neuronal populations in the brain and spinal cord (Sorg et al., 2016;Miyata and Kitagawa, 2017). Starting from the pioneering work of Pizzorusso and co-authors a large body of experimental evidence accumulated showing that PNN restrict neuronal plasticity, i.e., rewiring of synaptic networks in the adult brain (Pizzorusso et al., 2002).
The PNN development takes place during the critical period of the brain maturation and is controlled by synaptic activity within the network (Carulli et al., 2010;McRae et al., 2007;Nowicka et al., 2009;Lander et al., 1997;Bru¨ckner et al., 2004). Synaptic activity-dependent changes in PNN include changes in the expression of three different CSPG epitopes (Lander et al., 1997) and aggrecan mRNA (McRae et al., 2007).
High resolution PNN microstructure remained unstudied until recently (Arnst et al., 2016). Quantitative changes have been now demonstrated at the single mesh level in experimental models of focal cerebral ischemia (Dzyubenko et al., 2018), Rett syndrome (Sigal et al., 2019) and schizophrenia (Kaushik et al., 2020).
In the present study we analyzed microstructure of the complex of GABAergic synapses with PNN in the mouse somatosensory cortex and compared its formation in the presence versus absence of sensory input during the critical period of early postnatal development. We demonstrate that the sensory input is required for the proper morphology formation of the presynaptic terminal and the surrounding PNN.

Animals
For sensory deprivation experiments mice were subjected to whisker shaving once a day during P0-P30. Unshaved littermates were used as control.

Tissue preparation
Mouse brain samples were collected according to regulations of the ethics committee of Kazan Federal University. For immunohistochemistry animals were terminally anaesthetized with an intraperitoneal overdose of Urethane (Sigma-Aldrich) and were immediately perfused through the heart with 40 ml of cold phosphate-buffered saline (PBS, pH 7.4), followed by the same volume of cold 4% paraformaldehyde. Brains were removed and postfixed over-night in 4% paraformaldehyde at + 4 C. After that brains were cryoprotected with 30% sucrose in PBS, pH 7.4 for 48 h and then frozen in embedding medium (Tissue-Tek, Sakura, Japan) at − 80 • C. 18-20 micrometer-thick coronal brain sections were cut on a cryostat. The barrel cortex regions were determined using Comparative Cytoarchitectonic Atlas of Mouse Brain (Patrick R. Hof, Elsevier).

Staining procedure
PNN was stained with the biotinylated Wisteria floribunda Lectin (VectorLab, USA) that is generally thought to bind N-acetyl-galactosamine (GalNAc) of CS chains (Ha¨rtig et al., 1992). The staining was performed on free-floating sections. All incubations were carried out in 24-well plates with 500 microliter per well. After sectioning samples were washed three times using phosphate-buffered saline (PBS; pH 7.4) and then treated for 1 h with a blocking solution containing 5% bovine serum albumin (Sigma) in 0.1 M PBS with 0.5% Triton X-100 (PBST). Streptavidin/Biotin Blocking Kit (VectorLab, USA) was used according to the manufacturers protocol to block endogenous biotin. After that, those sections were quickly rinsed in PBS and incubated overnight at 4 • C using biotinylated Wisteria floribunda agglutinin with final concentration 2 microgram/ml (dilution 1:1000) in 10 mM HEPES and 0.15 M sodium chloride, pH 7.4. Samples were washed three times for 10 min with PBS and incubated for 30 min with AlexaFluor633-conjugated streptavidin (Invitrogen) (dilution 1:500). After that the sections were washed 3 times with PB And pre-treated with 0,3% Triton X-100 for 1 h. Then sections were incubated at blocking solution (5% Normal serum, 1% BSA, 0,3% Triton X-100 in PB) for 1 h and incubated with anti-VGAT IgG (Synaptic Systems) (dilution 1:1000) in PB+ 0,5% Normal serum+ 0,3% Triton X-100 for 48 h at + 4 • C. Samples were washed 3 times for 10 min with PB. Alexa488-conjugated goat anti-rabbit (Thermo-Fisher Scientific) was used as a secondary antibody (dilution 1:200) in phosphate buffer, incubation time 2 h. The sections were mounted on slides, air-dried and coverslipped with ImmunoMount (ThermoScientific).

Image analysis
FIJI (Schindelin et al., 2012) and Imaris (Bitplane) software packages were used for the image analysis. All macros were written in FIJI. The 2D mesh tracing procedure was used as described previously (Arnst et al., 2016). The autothresholding procedure was developed based on the approach described previously (Lipachev et al., 2019). A square area containing a single PNN mesh was subjected to autothresholding using all 17 autothresholding algorithms available in FIJI. As a result, 17 threshold intensity values were obtained and then analyzed together. The intensity range was divided into bins and the final threshold intensity value was calculated as a mean of values falling in the most populated bin and adjacent ones, if their populations were high enough (50% or more of the most populated bin) (Fig. S1). The procedure was first applied to VGAT signal of those meshes oriented parallel to the focal plane. After that the same procedure was applied to the analysis of both WFA and VGAT confocal images of those synapses oriented perpendicularly to the focal plane (transverse section).
By applying the calculated threshold we get the confocal stack masked. We compare the masks of the neighboring layers for matching objects and annotate those as belonging to the same 3D object. The procedure of searching for new elements of the object is repeated through the confocal stack until no new elements are found. This allows us to gradually build up a 3D object. The same z position of a stack is used repeatedly in multiple iterations to search for separate branches of 3D objects with a complex branched geometry.
To calculate the mean width of CS sheath we analyzed profiles of intensities of each pixel of the mesh in the layers below and above the focal plane. The "getPixel" FIJI command was used to find intensities. The intensity values of the traced mesh perimeter were used to set the intensity threshold (30% of the intensity value of the mesh perimeter) for the pixels below and above the initial confocal plane for the detection of the borders of the mesh in the z direction. Then the distances between the upper and lower borders were averaged for all XY positions within the mesh perimeter to get the mesh z-width.
Local intensity maxima for the Z-to-mesh perimeter maps of the CS intensity were calculated with "Find Maxima" command in FIJI.
For 3D analysis of VGAT-positive objects in the focal plane-aligned meshes as well as both WFA-positive and VGAT-positive objects in transverse confocal sections of meshes the algorithm was extended to adequately describe the size and shape changes and branching of the objects between individual confocal planes within a stack.
For the synapse transverse section analysis we designed the following procedure. First, it performs segmentation of the intracellular and extracellular space adjacent to the CS sheath. To select the central line's joints the FIJI PointPicker plugin was used. That allows us to perform segmentation on the objects with a complex 3D shape that cannot be approximated with a 2D polygon of specific width. After that we calculate coordinates of parallel lines for each segment of the central line that are placed on certain distance from it (Fig. 4). That allows us to segment the object and the space around it into sections.

VGAT distribution in the PNN-synapse complex. 2D analysis
To analyze the morphology of VGAT-positive synaptic boutons we developed an automatic algorithm of the VGAT fluorescence segmentation inside PNN meshes based on the autothresholding function in the FIJI software (Lipachev et al., 2019) (see Methods and Fig. S1). The algorithm allows for unbiased choice of a segmentation threshold for VGAT-positive puncta in a synapse inside a PNN mesh (Fig. S1).
We analyzed 1797 PNN meshes in 38 neurons from 3 mice (Fig. 1A, B) and observed 3 types of the VGAT-positive puncta distribution in PNN meshes on neuronal cell surface ( Fig. 1 C-I). A distinct synaptic population had a characteristic morphology with one large cluster of VGAT fluorescence, mean cluster area 0.87 µm 2 (STD 0.37 µm 2 ) occupying > 60% of the mesh area (Fig. 1C, D). Another synaptic population had 2-4 smaller VGAT clusters inside each single mesh of PNN ( Fig. 1

E, F). Mean
VGAT cluster area was 0.22 µm 2 (STD 0.12 µm 2 ) for single cluster and 0.7 µm 2 (STD 0.17 µm 2 ) for total VGAT cluster area over a single PNN mesh for that population. Finally, a third substantial population had just one small cluster of VGAT per single mesh with a mean cluster area 0.29 µm 2 (STD 0.22 µm 2 ) (Fig. 1G, H). A relative occurrence of the 3 types of VGAT distribution is shown in Fig. 1 I. No correlation was observed between the type of VGAT distribution and the PNN mesh area (Data not shown).

Sensory input is required for the proper VGAT distribution within the PNN-synapse complex. 2D analysis
We then asked whether this distribution of the VGAT fluorescence in the PNN-synapse complex is dependent on the sensory input to the developing synaptic network of the brain cortex.
We subjected newborn mice to vibrissae shaving every day during P0-P30 and compared geometry of the GABAergic synapses surrounded with PNN in layer IV of the barrel cortex in deprived versus control littermates. We analyzed 3446 PNN meshes in 76 neurons from 3 independent experiments.
In the 2D analysis we observed significant reduction of occurrence of the morphology variant with single large VGAT cluster under the sensory deprivation as compared to control (Fig. 1I). Consistently, mean area of VGAT-positive objects was reduced in the mice subjected to sensory deprivation (Fig. 1J). About the same portion on meshes contained VGAT-positive objects under sensory deprivation and in control (Fig. 1K).

Sensory input is required for the proper VGAT distribution within the PNN-synapse complex. 3D analysis
We then extended the 2D analysis presented in Fig. 1 to similar 3D procedure on confocal stacks ( Fig. 2A-J, L-N). This type of analysis allows to measure 3D geometry of VGAT-positive objects inside PNN meshes (Fig. 2G-J). We observed significant reduction of occurrence of the morphology variant with single large 3D cluster of VGAT under the sensory deprivation as compared to control (Fig. 2L) supporting the results of 2D analysis (Fig. 1I). Furthermore, mean volume of VGATpositive 3D objects was reduced in the mice subjected to sensory deprivation (Fig. 2M).

Sensory deprivation affects the relative spatial distribution of VGAT versus PNN
We further expanded the 3D analysis of the PNN+synapse complex geometry by analyzing the choldroitin sulphates (CS) staining intensity distribution along the Z axis and the mesh perimeter (Fig. 2K). That type of analysis allowed us to measure relative distribution of the VGAT and CS signals along the Z coordinate (Fig. 2G, J, K). Fig. 2O demonstrates average values of root mean square deviation (RMSD) for the Z position of the highest mean intensity of the CS signal along the mesh perimeter to the Z position of -1) the area maximum, 2) the mean intensity maximum, 3) the median -of the VGAT signal 3D distribution under somatosensory deprivation versus control.
The intensity maximum of CS within the mesh perimeter and the area maximum of VGAT within the synaptic terminal are shifted relative to each other along the Z coordinate under somatosensory deprivation as compared to control (Fig. 2O).

Sensory input is required for the proper morphogenesis of the PNN meshes
To address the effect of sensory deprivation on the PNN geometry we first analyzed the mesh area in the PNNs from animals with shaved whiskers versus controls. We could see no significant difference in the shape, area and the chondroitin sulfate staining distribution between the two groups (Data not shown). We then analyzed 3-dimensional distribution of the CS staining intensity along the mesh perimeter (Fig. 3). We calculate the mean CS intensity along the mesh perimeter for each confocal plane and use 10% of the highest value as the segmentation threshold along the Z axis to define the upper and lower border of the mesh (shown in green in Fig. 3E, F). We demonstrate that the mean Z-width (or "thickness") of PNN meshes decreases upon somatosensory deprivation (Fig. 3E, F, I).
We further demonstrate that RMSD between the Z position of the highest mean intensity and z positions of local intensity maxima decreased significantly under somatosensory deprivation (Fig. 3J, K), i. e., the meshes become more flattened if the developing brain cortex lacks the sensory input.
As both the VGAT-positive volume and the PNN mesh thickness were affected by sensory deprivation (Fig. 2M, 3E, F, I) we then asked the question whether those two parameters correlate with each other. We observed no pronounced correlation between the VGAT volume and the Z-width or between the VGAT volume and the RMSD of intensity maxima (correlation coefficients r = 0.148 and 0.271, respectively).

Sensory input is required for the proper morphogenesis of the PNNsynapse complex. The synapse transverse section analysis
To validate our findings of the PNN mesh flattening and reduction of the VGAT cluster volume caused by sensory deprivation we took an alternative approach to the PNN 3D geometry quantification by analyzing transverse sections of the PNN-covered neuronal cell surface (Fig. 4). As XY resolution of confocal microscopy is remarkably higher than Z resolution, so one advantage of this approach is that a side projection of the mesh is oriented in the confocal plane (XY plane) rather than in Z dimension resulting in higher spatial resolution of the image data.
We performed 3D analysis analogous to what we did on meshes oriented along the confocal plane (Fig. 2J, M) and confirmed the reduction of the mean VGAT cluster volume under somatosensory deprivation as compared to control (Fig. 4A-E, K).
We then performed segmentation of the intracellular and extracellular space adjacent to the CS sheath ( Fig. 4F-J) and demonstrated that the width ("thickness") of the CS sheath surrounding VGAT-positive synapses is reduced under somatosensory deprivation as compared to control (Fig. 4I).

Discussion
The effect of the sensory input on neuronal plasticity has been widely studied following the classical work of Wiesel and Hubel on monocular deprivation (Wiesel and Hubel, 1965;Fagiolini et al., 1994, Berardi et al., 2000. Developmental increase of the activity of cortical inhibitory circuits has been shown to contribute to the closure of the critical period of cortical synaptic network plasticity (Hensch et al., 1998;Huang et al., 1999;Kirkwood et al., 1994).
The experimental evidence on physiological and pathological roles of PNN has been accumulating rapidly over the last decade (Bitanihirwe et al., 2016;Pantazopoulos and Berretta, 2016). This evidence of functional importance raises the question of the structural mechanism for the PNN function. Nevertheless, quantitative structural studies of the PNN fine structure remained absent until recently (Arnst et al., 2016;Dzyubenko et al., 2018;Sigal et al., 2019;Kaushik et al., 2020).
A functional hallmark of the brain cortex postnatal development from the immature to the mature state is a drastic decrease of neuronal plasticity and perineuronal nets were systematically shown to play a major role in that developmental switch (Pizzorusso et al., 2002). Developmental changes in the CSPG expression were previously studied in particular detail in the brain visual cortex. Distinct CSPG epitopes both in the CS moiety and in the protein core exhibit activity-dependent expression in visual cortex at the closure of critical period in the cat model of dark rearing (Lander et al., 1997). Similar reduction of CSPG epitopes was reported in cat lateral geniculate nucleus upon dark rearing (Sur et al., 1988;Guimaraes et al., 1990;Kind et al., 1995).
In the layer IV of the barrel cortex perineuronal nets are formed at the end of the critical period of plasticity (Köppe et al., 1997). Consistently, a significant portion of the neuronal plasticity processes round up (caption on next page) N. Lipachev et al. by the end of the first postnatal month in the mouse barrel cortex (Fox, 2002;Glazewski and Fox, 1996).
We previously hypothesized that the spatial structure of the perineuronal net surrounding a synaptic bouton should play an important role in the synapse physiology (Arnst et al., 2016) based on the experimental data demonstrating that CSPG bind extracellular Ca 2+ and  a range of signaling ligands including pleiotrophin, VEGF, GDNF Rauvala et al., 2017;Nandini et al., 2004). We previously developed a mesh-tracing method for quantification of the mesh geometry and molecular epitope distribution along the mesh perimeter (Arnst et al., 2016). The initial aim of the present study was to study the synaptic terminal microstructure inside the mesh under the control and somatosensory deprivation conditions. We also wanted to search for correlation between the synaptic terminal and the surrounding PNN mesh geometry parameters in order to understand the integral architecture of the synapse + PNN complex as a functional unit.
For that purpose we developed a "synthetic" autothresholding algorithm to measure parameters of the VGAT GABA transporter spatial distribution within a synaptic terminal surrounded by a PNN mesh (Fig. 1, S1). That approach is an extension of our previous procedure using 16 different autothresholding algorithms provided by FIJI (Lipachev et al., 2019). With this method we are aiming at an automatic and unbiased image segmentation procedure retrieving as accurate structural information from the brain tissue confocal images as possible.
Quantitative distribution of VGAT within the GABAergic synaptic terminals remains poorly studied. In their pioneering study of the cerebellar synapse ultrastructure Chaundhry and co-authors demonstrated that VGAT co-localized with synaptic vesicles and was absent from other compartments of a synaptic terminal (Chaudhry et al., 1998).
These data suggest that the VGAT signal observed in our study corresponds to the distribution of vesicles with the neurotransmitter next to the GABAergic synapse active zones. Our observation on the reduction of the VGAT immunofluorescence cluster size resulting from the somatosensory deprivation (Figs. 1, 2, 4) leads to the conclusion that the sensory input is essential for the proper synapse geometry formation during development, more specificallyduring the critical period of the mouse brain cortex synaptic plasticity within the first postnatal month.
Whisker trimming in neonatal mice causes reduction in the aggrecan mRNA expression and Cat-315 immunoreactivity in the barrel cortex (McRae et al., 2007). Taken together the results by McRae and co-authors (McRae et al., 2007) and our results presented here demonstrate that somatosensory deprivation causes changes both in the molecular composition and 3D structure of the perineuronal net (Fig. 3) and in the spatial structure of the synaptic terminals embedded in the PNN (Fig. 2). Possible causal connections between those effects remain to be investigated.
We previously analyzed the 3D spatial structure of PNN using the autodepth filament tool of the Imaris software (Arnst et al., 2016). Here we expand the analysis repertoire by measuring the CS epitope distribution along the mesh perimeter also in the Z direction (Fig. 2K, 3). That gives a quantitative estimate of the mesh width and curvature along the Z axis. Moreover, that approach allows us calculating a Z position with the highest average intensity of the CS signal to be used as a "0 coordinate" of the mesh geometry for the quantification of relative positions of VGAT-positive objects within the synapse + PNN complex (Fig. 2G, J, K, O).
The 2D area and 3D volume of VGAT-positive objects ( Fig. 1C-J, Fig. 2J, L-N) could be viewed as characteristics of the synaptic vesicle pool size. That is a crucial parameter for the synaptic function and thus our quantitative studies on the VGAT-positive object microstructure and distribution should shed light on the structural and functional connection between the synapse and the surrounding PNN.
The approach presented here for quantitative structural studies and the results on the synaptic terminal + PNN complex 3D structure contribute to understanding coordination between the ECM and synaptic network and the molecular physiological mechanisms connecting these two brain components into an integral structural and functional unit.

Conclusion
To summarize, our data demonstrate that the fine microstructure of the cortical GABAergic synaptic terminals and surrounding PNN meshes develops postnatally under the influence of sensory input so that withdrawal of the input leads to malformation of particular structural features of the synapse + PNN complex.

Sample preparation was funded by Russian Foundation for Basic
Research (project number 20-315-90074). The work was performed according to the Russian Government Program of Competitive Growth of Kazan Federal University. This paper has been supported by the Kazan Federal University Strategic Academic Leadership Program (PRIORITY-2030).

Ethical approval
BalbC mice were used according to regulations of the ethics committee of Kazan Federal University.