The protein tyrosine phosphatase RPTPζ/phosphacan is critical for perineuronal net structure

Perineuronal nets (PNNs) are conspicuous neuron-specific substructures within the extracellular matrix of the central nervous system that have generated an explosion of interest over the last decade. These reticulated structures appear to surround synapses on the cell bodies of a subset of the neurons in the central nervous system and play key roles in both developmental and adult-brain plasticity. Despite the interest in these structures and compelling demonstrations of their importance in regulating plasticity, their precise functional mechanisms remain elusive. The limited mechanistic understanding of PNNs is primarily because of an incomplete knowledge of their molecular composition and structure and a failure to identify PNN-specific targets. Thus, it has been challenging to precisely manipulate PNNs to rigorously investigate their function. Here, using mouse models and neuronal cultures, we demonstrate a role of receptor protein tyrosine phosphatase zeta (RPTPζ) in PNN structure. We found that in the absence of RPTPζ, the reticular structure of PNNs is lost and phenocopies the PNN structural abnormalities observed in tenascin-R knockout brains. Furthermore, we biochemically analyzed the contribution of RPTPζ to PNN formation and structure, which enabled us to generate a more detailed model for PNNs. We provide evidence for two distinct kinds of interactions of PNN components with the neuronal surface, one dependent on RPTPζ and the other requiring the glycosaminoglycan hyaluronan. We propose that these findings offer important insight into PNN structure and lay important groundwork for future strategies to specifically disrupt PNNs to precisely dissect their function.

Studies over the past two decades have identified the important and perhaps fundamental role for the neural extracellular matrix (ECM) 3 in regulating neuronal plasticity in the central nervous system (CNS). For example, manipulation of the neural ECM disrupts developmental ocular dominance plasticity in the primary visual cortex and modulates learning and memory in multiple regions within the brain (1)(2)(3). Indeed, alterations and/or disruptions of the neural ECM are associated with a number of neuropsychiatric disorders and neurodegenerative diseases (4 -8). Importantly, these studies primarily attribute these functions to a unique and enigmatic neural ECM substructure called the perineuronal net (PNN).
PNNs are conspicuous reticular ECM formations that ensheathe limited but specific subsets of neurons in the CNS (9,10). These highly aggregated ECM structures surround the cell body and proximal neurites of ensheathed neurons but appear to be excluded from sites of synaptic contact thereby giving them a lattice-like or net-like appearance. Early work noted that expression of PNNs in primary sensory cortices, such as visual and somatosensory cortex, is activity dependent and that the timing of their appearance is consistent with closure of developmental critical periods such as the critical period for ocular dominance plasticity (11,12). Furthermore, in the cortex, PNNs are particularly enriched surrounding parvalbumin-expressing interneurons, which are known to be key regulators of developmental plasticity. Therefore, it is hypothesized that PNNs may be important regulators of developmental plasticity in the developing cortex.
Previous work showed that PNNs are particularly enriched in the glycosaminoglycan hyaluronan (HA) and hyaluronan-binding chondroitin sulfate proteoglycans (CSPGs). The presence of highly aggregated CSPGs in PNNs has received particular attention because CSPGs are notoriously inhibitory molecules in the CNS and a major barrier to regeneration in the injured nervous system (13)(14)(15). Therefore, the appearance of CSPGenriched structure surrounding synapses coincident with closure of the period of heightened developmental plasticity led to the hypothesis that these structures likely inhibit synaptic plasticity. Consistent with this hypothesis more recent studies demonstrated that, indeed, disruption of the neural ECM and PNNs in the visual cortex restores juvenile ocular dominance plasticity in mature rodent brain (1). Importantly these findings suggest that ECM and PNN disruption alone is sufficient to reopen juvenile plasticity, indicating that these are critical regulators of neural plasticity.
The body of work linking PNNs to developmental plasticity (16 -18) is quite compelling, however; an ever-growing list of studies now links PNNs to a vast array of neural functions and pathologies. Of particular interest are prominent roles for PNNs in other forms of plasticity such as learning and memory. An explosion of recent work demonstrated that manipulating the ECM and PNNs in multiple brain regions including the amygdala, hippocampus, cortex, and striatum significantly alters learning and memory (2, 3, 19 -22). Although these results are quite striking and exciting for the field, a mechanistic understanding of PNN function has been surprisingly elusive. We still do not understand what the precise function of PNNs is nor how they modulate plasticity. In addition, there are conflicting findings from different labs with PNN manipulations that further highlight the limits of our understanding of this structure (23,24).
The limited mechanistic understanding of PNN function is derived primarily from an incomplete understanding of its molecular composition and structure and, in turn, the inability to specifically disrupt PNNs without disrupting the surrounding ECM. For example, much of the work identifying roles for PNNs in various forms of plasticity has relied on enzymatic digestion with chondroitinase ABC (ChABC). This treatment impacts PNNs but does not necessarily eliminate the structure, and even disrupts the surrounding ECM (25). Additionally, genetic models disrupting specific PNN components have provided insight into their function, but also typically impact the surrounding neural ECM. Therefore, the goal of this study is to provide a more complete understanding of PNN structure toward the ultimate goal of developing more precise strategies to specifically disrupt PNNs to better study their function.
In this study we detail a novel role for receptor protein tyrosine phosphatase zeta (RPTP) in the structure of PNNs. Utilizing mouse models and neuronal cultures, we demonstrate RPTP is critical for the proper formation of PNNs. Further analysis using molecular and biochemical techniques shows PNNs are bound to the neuronal surface through two distinct mechanisms, one requiring hyaluronan and the other RPTP. Overall, our data provide novel insights into the structure and formation of PNNs to ultimately develop tools to precisely determine PNN function.

PNNs are disrupted in Ptprz1 KO adult mice
A detailed understanding of PNN function has been elusive due in large part to an incomplete understanding of their composition and structure. In this regard the role of RPTP in PNNs is intriguing because although it has been localized to PNNs, its role in these structures has never been thoroughly studied. Although virtually all other proteins identified in PNNs are secreted proteins, the full-length isoform of RPTP is a large transmembrane protein and thereby could provide a key anchor point for PNNs to the neuronal cell surface. Further-more, RPTP is a phosphatase and could also serve a signaling function in PNNs. Finally, RPTP is known to interact with other key PNN components such as Tnr and could provide a key link to the other components in this structure. Therefore, we investigated PNN structure in Ptprz1 KO mice. Of note, the nomenclature surrounding the protein products of the Ptprz1 gene, RPTP and phosphacan, in the literature is somewhat confusing. This largely stems from various laboratories isolating the proteoglycan using different monoclonal antibodies. As such RPTP is also known by DSD-1, 6B4, 3F8, and RPTP␤. It is now generally accepted that the receptor form be described as RPTP, to match its gene Ptpr1 (Ptprz1). Therefore, in this article, we will designate all protein isoforms of the Ptprz1 gene as RPTP and specifically name the secreted variant phosphacan where relevant.
Before investigating PNN structure, we confirmed the loss of RPTP of our Ptprz1 KO mice through immunostaining using RPTP-specific antibody 3F8 in adult cortical sections (Fig.  S1A). 3F8 staining was essentially eliminated in the brains of the knockout animals confirming the validity of this model. We subsequently immunostained cortical sections of PND 90 brains with the most well-established markers of PNNs, antibodies directed against aggrecan and the lectin WFA. In WT brains, PNNs had the typical highly organized lattice structure on a subset of neurons in the cortex. However, PNNs in the KO brains appeared altered in structure and less organized than their WT counterparts. Upon closer observation, we noted that although typical PNN staining reveals bridgelike strings interconnecting foci along the neuronal surface that create discrete gaps or holes that give a netlike appearance, these "bridges" seemed largely absent in the KO brains ( Fig. 1). Interestingly PNNs in brains from heterozygous mice seemed largely unaffected. In the KO animals, however, instead of the intricate, lattice-like structure as seen in WT, PNN components aggregated on the neuronal surface, creating prominent foci. Despite this disrupted structure, interestingly, PNN areal and cell-specific distribution remained unaffected in Ptprz1 KO mice. In addition, we found this disruption occurred as early as PND 21 (Fig. S1B). From these data, we concluded RPTP is necessary for proper formation of PNNs from early in mammalian development through maturity.
To quantify the disrupted PNN phenotype, we analyzed two aspects of PNN staining: PNN intensity and PNN spatial distribution on the neuronal surface. Cortical brain slices were stained with WFA and individual PNN-bearing neurons were imaged and analyzed across the Ptprz1 brains (WT, n ϭ 4 animals, 55 PNNs; Het, n ϭ 3 animals, 26 PNNs; KO, n ϭ 4 animals, 43 PNNs). From our initial observations, we found that disrupted Ptprz1 KO PNNs had large areas devoid of PNN staining, possibly because of component aggregation and collapse of the stringlike interconnections ( Fig. 2A). To quantify this observation, we first developed a binary gap analysis as described in "Experimental procedures" (Fig. 2). A binary process converts any image of varying pixel intensity to black and white using a unbiased calculated threshold, making it a useful analysis tool to assess the empty space of an image (Fig. 2B). Of particular interest, the binary image of control PNNs distinctly shows the well-defined gaps and regular lattice-like structure of PNNs. In Ptprz1 KO mice, however, the gaps are no longer well-defined and PNN staining appears aggregated and discontinuous leading to gaps or large areas of no PNN staining. After calculating the black/white pixel count ratio, we found no significant difference between Ptprz1 WT and Het PNNs (data not shown), therefore the data sets were combined. In Ptprz1 KO PNNs, however, we found a significantly higher black/white pixel count ratio (p ϭ 1.09 ϫ 10 Ϫ5 , Student's t test), confirming greater areas of no PNN staining (Fig. 2C).
The above method effectively describes the ratio of PNNcontaining regions to regions devoid of PNN staining on the surface of neurons between Ptprz1 WT/Het and Ptprz1 KO animals. However, in addition to this, PNN staining from the Ptprz1 KO neurons appears discontinuous and aggregated in contrast to the more regular, continuous staining seen in Ptprz1 WT/Het animals. To quantify this element of PNN structure we developed a PNN peak or node analysis. This method takes into account the difference in spatial distribution and clustering of PNN staining on the cell surface between the different groups. As before we utilized WFA as a marker for PNNs. We determined the sharpness of PNN peaks seen on the neuronal surface using the local maxima function in ImageJ (Fig. 2, D and  E). The more continuous and regular WFA staining of the Ptprz1 WT/Het group resulted in significantly higher number of nodes or peaks in this genotype. The aggregated staining of the Ptprz1 KO PNNs led to detection of significantly fewer nodes (p ϭ 0.0009, Student's t test), which also appeared more isolated than in Ptprz1 WT/Het animals (Fig. 2F). The average prominence of PNN nodes over their surrounding space in the Ptprz1 KO PNNs was also significantly higher compared with Ptprz1 WT/Het PNNs (p ϭ 0.014, Student's t test) (Fig. 2G). Both these data indicate the regular continuous structure of PNNs in the Ptprz1 WT/Het animals and the broken discontinuous PNN staining seen in Ptprz1 KO cells. The increase in the average prominence levels of WFA peaks in the Ptprz1 KO PNNs further points to the isolated nature of PNN nodes and the loss of connections between them in the Ptprz1 KO animals.

ECM components remain bound to disrupted PNN structures in Ptprz1 KO mice
Although our data indicate that proper PNN structure is disrupted in Ptprz1 KO mice, the impact of RPTP on the other known PNN components remains unclear. To further define how RPTP contributes to the underlying structure of PNNs, we stained cortical sections of adult Ptprz1 mice with PNN markers WFA, aggrecan, HAPLN1, neurocan, brevican, and Tnr, and further quantified for fluorescent intensity (Fig. 3, A). Although staining with all components showed the same aggregate PNN structure found with aggrecan and WFA, all ECM components, including Tnr, remained bound to PNNs in Ptprz1 KO mice. Quantifying the various PNN components, and accounting for sex, we found no significant ECM intensity difference between Ptprz1 WT (n ϭ 9; 3 males, 6 females) and Ptprz1 Het (n ϭ 12; 6 males and 6 females) mice for most of the PNN components. When compared with Ptprz1 KO mice (n ϭ 14; 4 males, 10 females), we only found significant losses in WFA (p ϭ 6.53 ϫ 10 Ϫ5 , Student's t test) and aggrecan (p ϭ 0.01) whereas no significant losses were seen for all other PNN components analyzed.
Expression of aggrecan and hyaluronic and proteoglycan link protein 1 (HAPLN1) in the cortex of Ptprz1 mice was analyzed via Western blot analysis (n ϭ 3 per genotype) (Fig. 3B). Similar to IHC results, aggrecan protein levels in the cortex were significantly reduced in Ptprz1 KO mice (p ϭ 0.03, Student's t test), while HAPLN1 protein levels remained unchanged (p ϭ 0.64). To ensure aggrecan loss did not occur at the transcription level, we performed an RT-PCR analysis and found no significant changes in gene expression across all Ptprz1 (data not shown). From these analyses, we conclude that although important for proper structural formation, the loss of RPTP does not affect the overall localization of ECM components to the PNN surface.

The PNN disruption in Ptprz1 KO mice phenocopies the PNN disruption in tenascin-R KO mice
We next sought to determine a mechanism in which RPTP anchors PNNs to the surface. We noted that the PNN phenotype we observed in the Ptprz1 KO brains looked remarkably similar to the phenotype found previously in Tnr KO brains (26,27). Additionally, Tnr is of particular interest because it is a high-affinity binding partner of RPTP (28 -30). To confirm our observations, we obtained PFA-fixed PND 90 adult brains of Tnr WT and Tnr KO mice and directly compared the PNN phenotype to Ptprz1 KO mice (Fig. 4). Staining with WFA and aggrecan antibody, PNNs in Tnr WT resembled a typical lattice-like structure as observed with Ptprz1 WT/Het PNNs. The lattice-like structure of controls was completely absent and replaced by areas of aggregation. More importantly, these phe- In littermate WT and Het, typical lattice-like PNN structures were observed. In addition, there were no discernable differences in structure between the two genotypes. However, PNNs in sections from Ptprz1 KO mice the lattice-like PNN structure is lost with aggrecan and subsequently WFA staining appearing aggregated on the neuronal surface. Scale bar, 10 m.

RPTP/phosphacan is critical for PNNs
notypes were indistinguishable from PNNs in Ptprz1 KO mice. From these data, we concluded RPTP likely interacts with Tnr to mediate PNN structure.

PNN component aggrecan is immobilized on the cell surface by a distinct HA-and EDTA-sensitive interaction
Previous work has identified and demonstrated the importance of Tnr, HAPLN1, and the lectican CSPGs, aggrecan, neurocan, and brevican, in the proper formation of PNNs (23, 26, 27, 30 -41). We have found PNNs to be disrupted in Ptprz1 KO mice. However, the exact mechanism by which these components bind together to form PNNs is not clear. To better understand the mechanism of binding of PNN components we established a biochemical release assay to measure binding of aggrecan to the neuronal cell surface. Aggrecan is the best PNN component for this analysis because it is the most PNN-specific component and recently has been shown to be perhaps the key CSPG in PNN formation (24). Our finding of the potential interaction between RPTP and Tnr in PNN structure led to our hypothesis that we should be able to disrupt this element of PNN structure by chelating calcium. The interaction between RPTP and Tnr is Ca 2ϩ dependent and, subsequently, the interaction of Tnr with aggrecan also depends on Ca 2ϩ . Therefore, we reasoned that treatment with EDTA would disrupt this entire complex and if this complex were involved in PNN cellsurface binding, would enhance the release of aggrecan. In addition, current models of PNN structure suggest that aggrecan is immobilized to the cell surface by interacting with the PNN backbone made up of the glycosaminoglycan HA (32,40,42). Therefore, we reasoned digestion of HA should also increase the release of aggrecan. If both mechanisms are independently involved in the binding of aggrecan to the neuronal membrane then we hypothesized that in the Ptprz1 and Tnr KO brain, aggrecan release would depend only on HA. To test this PNNs from PND 90 adult WT, Het, and KO mice were stained with WFA and quantified. WT and Het animals showed no statistically significant difference in any of the following analyses and were grouped together as WT/Het. A, PNNs from WT/Het appear brighter and show regular meshlike PNN appearance whereas KO PNNs show decreased WFA intensity and a disrupted staining pattern. B, shows a magnified view of the surface of the cells in A converted to binary images. C, KO animals showed significantly higher black/white pixel ratio, indicating an increased area devoid of PNN staining (p ϭ 1.09 ϫ 10 Ϫ5 , two-tail Student's t test, S.D. for error bars). D, representative image of WT/Het and KO PNNs for PNN node/peak analysis at varying prominence levels. Yellow color dots represent nodes/peaks of highest prominence (more isolated) followed by green, blue, and black. E, representative surface intensity profile of PNNs from WT/Het and KO animals. WT/Het PNNs show greater number of peaks as compared with KO PNNs. KO PNNs also appear more isolated with sharper peaks. F, WT/Het PNNs showed significantly higher number of peaks as compared with Ptprz1 KO PNNs (p ϭ 0.0009, two-tailed Student's t test, S.D. for error bars).G, the average isolation index of PNN peaks (represented here as the mean prominence level of all the peaks) was significantly higher in KO PNNs as compared with WT/Het PNNs (p ϭ 0.014, two-tailed Student's t test, S.D. for error bars) indicating the fewer and more sharp peaks in case of the KO animals. Binary image analysis and PNN peak/node analysis were carried out on WT (n ϭ 4 animals, 55 PNNs), Het (n ϭ 3 animals, 26 PNNs), KO (n ϭ 4 animals, 43 PNNs). Images in (B) are 5 m ϫ 5 m. Scale bar, 5 m.

RPTP/phosphacan is critical for PNNs
hypothesis, we isolated membranes from adult mouse brains and assessed the release of aggrecan by HA digestion in WT, Ptprz1 KO, and Tnr KO tissue (Fig. 5). We found that there is very little release of aggrecan from brains of WT mice (15 Ϯ 9%). In contrast, there was significant difference in release of aggrecan among the groups (one-way ANOVA F(2,7) ϭ 14.94, p ϭ 0.003). Both Ptprz1 KO brains (42 Ϯ 4%, p ϭ 0.0340) and Tnr KO brains (58 Ϯ 13%, p ϭ 0.0024) showed greater release as compared with WT. Interestingly, the vast majority of aggrecan was released from WT brains in our ChABC in conjunction with EDTA treatment assays (82 Ϯ 7%). Ptprz1 KO brains (87 Ϯ 19%) and Tnr KO (80 Ϯ 27%) brains also showed an increase and similar levels of aggrecan release to WT brains when treated with both ChABC and EDTA.
These data indicate that in WT mice aggrecan is immobilized on the cell surface by a dual interaction which is sensitive to HA digestion and EDTA, respectively. Attachment of aggrecan in either Ptprz1 KO or Tnr KO brains is compromised and the EDTA-sensitive mechanism of aggrecan binding is lost. These data along with the fact that aggrecan is similarly solubilized in our biochemical release assay from Ptprz1 KO and Tnr KO brains suggest that these two proteins are involved in aggregating this CSPG to the cell surface via the same mechanism. We next proceeded to further test the role of RPTP and Tnr in immobilizing aggrecan to the cell surface.

Tnr and RPTP are both required to bind aggrecan to the cell surface
The role of Tnr in proper formation of PNNs is well-known and has been demonstrated in a number of systems, including brain slices, organotypic cultures, and dissociated neurons from Tnr KO mice (26,37,38). Here we show RPTP disruption phenocopies the PNN structural deficit found in the Tnr KOs. Because RPTP and Tnr are high-affinity ligands for each other, we hypothesize these proteins interact in mediating PNN structure. To more thoroughly characterize the binding of aggrecan to the cell surface through RPTP and Tnr, we endeavored to recapitulate the effect in HEK293 cells, a cell line that does not

RPTP/phosphacan is critical for PNNs
form PNNs endogenously (Fig. 6). We found that adding aggrecan alone or aggrecan and Tnr does not result in binding the CSPG to the cell (Fig. 6, A-C). This is not completely surprising as Tnr is a secreted glycoprotein and does not contain any transmembrane domains to anchor it to the cell surface. Thus, its primary role is likely involved in the crosslinking PNN components but does not immobilize them on the cell. We then tested whether the binding of Tnr to the cell surface can be mediated by RPTP by expressing a construct containing the Tnr binding domain of RPTP in the HEK293 cells (Fig. 6, D-F). We found that in the presence of Tnr, cells expressing RPTP formed pericellular aggregates of aggrecan (Fig. 6D). Furthermore, HEK293 cells did not bind aggrecan when only RPTP was expressed, but required the presence of Tnr. These results demonstrate that both these two proteins are required to bind aggrecan to the cell surface. Interestingly, adding HA to these cells further enhanced staining and the netlike appearance of the structure highlighting the importance of HA in the aggregating PNN components to the cell surface (Fig. 6F).

Ptprz1 KO neurons show disrupted PNN distribution and distinct biochemical properties in culture as compared with WT neurons
Dissociated neuronal cultures present an attractive model to study binding of PNN components to the neuronal surface. This model system allowed us to perform biochemical analysis in live cells. We therefore utilized dissociated neuronal cultures to more thoroughly assess the role of RPTP in binding PNN components to the cell surface. Cultures were derived from E16 CD1 WT and Ptprz1 KO mice and were positive for PNN marker aggrecan (Fig. 7, A and E). However, staining in the Ptprz1 KO cells appeared disrupted and reduced in intensity as compared with the WT cultures, indicating the role of RPTP

RPTP/phosphacan is critical for PNNs
in binding PNN components to the cell surface. Our previous findings have shown that the interaction of aggrecan with the cell surface is mediated by RPTP and Tnr. Further, this interaction depends on two kinds of mechanisms, being sensitive to HA digestion and Ca 2ϩ ions, respectively. To determine the binding mechanism of aggrecan to cell surface and the role of RPTP in PNN formation we treated neuronal cultures from WT and Ptprz1 KO mice acutely with ChABC and/or EGTA. There are significant effects of genotype (F(1,16) ϭ 69.57, p Ͻ 0.0001) and treatment (F(3,16) ϭ 36.46, p Ͻ 0.0001) as well as interaction (F(3,16) ϭ 5.479, p ϭ 0.0088) on aggrecan staining when analyzed by two-way ANOVA. Tukey's post hoc testing (Fig. 7I) showed that acute treatment with ChABC alone had only a moderate effect on the release of aggrecan from WT cultures (Fig. 7B) (55 Ϯ 2%, p ϭ 0.0042). In contrast, the effect of ChABC treatment was dramatically enhanced in Ptprz1 KO cultures and led to a virtually complete loss of aggrecan staining in these cells (Fig. 7F) (14 Ϯ 3%, p ϭ 0.0202). Further, this effect of ChABC treatment was significantly different between the two genotypes (p ϭ 0.0015). EGTA treatment alone only had a small effect on aggrecan staining in both WT (Fig. 7C) (79 Ϯ 11%, p ϭ 0.3788) and Ptprz1 KO cultures (Fig. 7G) (84 Ϯ 19%, p ϭ 0.9958). Interestingly, the combination of the two treatments, i.e. ChABC in conjunction with EGTA, led to almost complete loss of aggrecan in WT (Fig. 7D) (11 Ϯ 2%, p Ͻ 0.0001) as well as Ptprz1 KO cultures (Fig. 7H) (9 Ϯ 2%, p ϭ 0.0137). These findings confirm our previous result and indicate that PNN component aggrecan is immobilized on the cell surface via two interactions, being sensitive to ChABC and a loss of Ca 2ϩ ions, respectively. Further they indicate that the Ca 2ϩ -dependent interaction is mediated by RPTP and is compromised in the Ptprz1 KO animals. To quantify these findings, the ratio of aggrecan intensity was calculated for WT and Ptprz1 KO cultures and is presented here in a graphical form (I). Analysis by two-way ANOVA showed significant differences by genotype (F(1,16) ϭ 69.57, p Ͻ 0.0001), treatment (F(3,16) ϭ 36.46, p Ͻ 0.0001), and interaction between the two (F(3,16) ϭ 5.479, p ϭ 0.0088). Most importantly although Tukey's post hoc test showed significant loss of aggrecan staining with ChABC alone in both WT and KO cells (p ϭ 0.0042 and p ϭ 0.0202, respectively), there was significantly more loss in the KO cells compared with WT cells (p ϭ 0.0015). A combination of ChABC and EGTA treatment resulted in a significant loss of aggrecan in both genotypes as compared with their respective untreated conditions (WT p Ͻ 0.0001, Ptprz1 KO p ϭ 0.0137) but no significant differences between genotypes. EGTA alone had no significant effect of aggrecan staining. These findings indicate that in WT cells PNN components are attached to the cell surface by two mechanisms, being HA dependent and calcium sensitive, respectively. These interactions are compromised in Ptprz1 KO cultures, and PNN components are immobilized on the neuronal surface only in a HA-dependent manner. Bars in graphs represent relative aggrecan intensity to untreated condition of the corresponding genotype Ϯ S.D. (B). Scale bar, 10 m.

RPTP/phosphacan is critical for PNNs Phosphacan and tenascin-R cooperate to stabilize PNN components to the neuronal surface
Currently, we have shown that Tnr remains bound to PNNs in Ptprz1 KO mice. If RPTP does indeed anchor PNNs through Tnr, we would expect RPTP to remain present on PNNs in Tnr KO mice. To test this hypothesis, we stained for RPTP on cortical sections of Tnr mice (Fig. 8A). Intriguingly RPTP, although prominent in Tnr WT, was greatly diminished on Tnr KO PNNs consistent with previous studies (26). This suggests Tnr expression is needed to maintain RPTP on the neuronal surface of PNNs. This made us investigate which specific isoform of RPTP is required for PNN structure in an effort to better understand the molecular composition of PNNs To clarify which Ptprz1 isoform, RPTP or phosphacan, serves to anchor PNNs to the neuronal cell surface, we turned to our culture model system. To differentiate between isoforms, we purified soluble phosphacan from P4 mouse brains and exogenously added it to Ptprz1 WT and KO neuronal cultures. If only the receptor form RPTP anchored PNN structures, we would not expect the exogenous addition of phosphacan to have any affect in our PNN culture model system. Phosphacan was added to Ptprz1 cultures starting at 3 DIV. Cultures were then fixed at 9 DIV and immunostained for PNN components aggrecan, neurocan, and Tnr (Fig. 8B).
In Ptprz1 WT cultures, aggrecan and neurocan tightly coated a population of neuronal cell bodies and proximal neurites, KO were fixed at DIV 9 and stained with PNN components, aggrecan, neurocan, and Tnr. We observed a distinct disruption of these components in Ptprz1 KO neuronal cultures when compared with WT. In addition, Tnr expression is largely reduced in Ptprz1 KO neuronal cultures. To determine which Ptprz1 isoform contributes to PNN structure, we purified the soluble form, phosphacan from PND 4 mouse brains through anion exchange chromatography. Purified phosphacan (0.25 g) was added to Ptprz1 KO neuronal cultures at 3 DIV. Cells were fixed and stained at 9 DIV for PNN components. Intriguingly, exogenous phosphacan seemingly recovered disrupted ECM to resemble WT morphology and expression, suggesting an unknown phosphacan receptor is important for PNN structure. Combining the above data, we conclude Tnr stabilizes the binding of the soluble isoform phosphacan to an unknown receptor on the neuronal surface. Scale bar, 10 m. We quantified this using PNN node/peak analysis. C, representative images of Ptprz1 KO and Ptprz1 KO ϩ phosphacan nodes or peaks at varying prominence levels. Yellow color dots represent nodes of highest prominence (more isolated) followed by green, blue, and black. D, representative surface intensity profile of PNNs. E, the average isolation index of PNNs represented by the mean prominence of peaks is significantly less in Ptprz1 KO ϩ phosphacan PNNs (n ϭ 12 PNNs, 3 cultures) compared with Ptprz1 KO cells (n ϭ 12 PNNs, 4 cultures) indicating a decrease in the sharpness of PNN nodes or peaks with phosphacan addition in Ptprz1 KO neuronal cultures (p ϭ 0.002, two-tailed Student's t test, S.D. for error bars). Scale bar, 10 m.

RPTP/phosphacan is critical for PNNs
whereas in Ptprz1 KO neuronal cultures, aggrecan and neurocan indiscriminately aggregated on neuronal surfaces, losing most discernable cell surface structure. Although Tnr was also spatially disrupted in Ptprz1 KO cultures, its expression on the neuronal surface was greatly diminished. After phosphacan addition to Ptprz1 KO cultures, the aggregated aggrecan and neurocan phenotype and reduced expression of Tnr in Ptprz1 KO cultures remarkably recovered to control conditions. To quantify the recovery of PNN structures in Ptprz1 KO neurons we utilized aggrecan staining and peak node/analysis as previously described. Addition of phosphacan to Ptprz1 KO cultures was sufficient to restore the spatial distribution of PNN components on the cell surface making them appear more continuous and less isolated (Fig. 8, C and D). Further, the average isolation index or mean prominence of PNN peaks was significantly lower in the Ptprz1 KO cultures with phosphacan addition as compared with Ptprz1 KO cultures alone (p ϭ 0.002, two-tailed Student's t test) (Fig. 8E). These data indicate that phosphacan, not RPTP, serves as an anchor to maintain proper spatial distribution of ECM components in vitro and, therefore, is likely the key Ptprz1 isoform that maintains the PNN lattice-like structure. Of particular interest, because of the soluble nature of phosphacan, it must bind to an unknown membrane-embedded receptor to anchor PNNs. Considering the above data, we conclude Tnr stabilizes the binding of the soluble form phosphacan to this unknown receptor on the neuronal surface and it is this binding and the interaction between both proteins that is critical for the intricate PNN lattice-like structure.

Discussion
Understanding and defining the molecular composition and structure of PNNs are a critical step toward unlocking their function. In the current study, we demonstrate a novel and unique role for the CSPG RPTP in the structure of PNNs. We show that in cooperation with Tnr, RPTP provides a key link to the neuronal surface, thereby generating the netlike or latticelike structure of PNNs. These studies enabled us to create a new model of PNN structural composition that will direct more precise investigation of PNN function.
PNNs were first described over a century ago by Camillo Golgi and even though more than a hundred years of PNN research have passed, their structure and exact molecular composition has remained elusive. These structures form a subcompartment of the neural ECM but are clearly distinct from the surrounding matrix (25,31). They are highly ordered and stable and demonstrate a unique geometry which is absent in the broader ECM (11,43,44). PNNs look like a mesh covering the surface of particular subsets of neurons. This meshlike appearance derives from the fact that PNN components are excluded from points of synaptic contact on the neuronal soma (45). We show here that this ordered distribution of PNNs on the neuronal surface depends on RPTP. In particular, there is a specific loss of the reticular structure of PNNs in Ptprz1 KO animals. Interestingly, most PNN components remain bound to the surface of the appropriate neurons in Ptprz1 KO animals but they no longer have a netlike structure. This is the first direct demonstration that RPTP is critically involved in PNN structure.
We noted that the altered PNNs in Ptprz1 KO animals brains seemed to significantly phenocopy disruptions previously found in Tnr KO mice (26,37). Our results suggest that these two PNN components are a part of the same binding mechanism immobilizing PNNs and a loss of either leads to disruption and loss of staining of their components. RPTP is a known high-affinity ligand for Tnr (29,30,46). We therefore sought to better understand the biochemical nature and order of Tnr and RPTP binding within PNNs. Our studies here and the work of others (26) show that in Tnr-deficient brains, RPTP staining is reduced in PNNs, suggesting Tnr is perhaps responsible for recruiting RPTP to PNNs. In contrast, in primary neuronal cultures from RPTP knockout brains Tnr is dramatically reduced in netlike structures. Furthermore, in studies in HEK293 cells we found that RPTP was necessary to recruit Tnr and other PNN components to the cell surface. Therefore, overall our data suggest a cooperative mechanism by which Tnr and RPTP contribute to cell surface binding of PNNs.
It is of particular interest that, even though PNNs are disrupted in mice lacking RPTP or Tnr, the total number and distribution of PNNs in the cortex is not obviously changed. One possible explanation for this is that a key component for PNN formation, the glycosaminoglycan HA, is still expressed by these cells and this could be another nucleating molecule for the formation on PNNs. The current structural model of PNNs revolves around the lectican family of CSPGs binding to HA, which acts as backbone for PNN formation. The entire structure of HA and CSPGs is then crosslinked by Tnr. Future studies will evaluate if knocking out both Tnr and RPTP leads to a more pronounced phenotype to determine whether they functionally depend on each other. In our biochemical assays, the absence of RPTP or Tnr left PNN components susceptible to release by digesting away the HA backbone, in contrast to WT tissue that required disruption of both the HA backbone and Tnr/RPTP interaction. Further, the interaction between HA and CSPGs is stabilized by the protein HAPLN1 (47,48). In mice carrying null alleles for HAPLN1, the interaction between CSPGs and HA is thought to be destabilized. PNNs in these HAPLN1 KO mice appear attenuated but the disruption is distinct from the structurally deficient PNNs of the Tnr KO mice (26,37,49,50).
Our findings along with existing data on PNN structure suggest that PNN components are immobilized on the neuronal surface by two distinct interactions. One dependent on Tnr and RPTP being sensitive to Ca 2ϩ ions and the other dependent on HAPLN1 and the HA backbone, susceptible to enzymatic digestion of HA. These findings enabled us to create a new and more refined model of PNN structure (Fig. 9). In this model, lectican CSPGs are bound to the HA backbone and stabilized by HAPLN1. These CSPGs are then crosslinked to RPTP by Tnr, which acts as an adapter protein between the two arms. Our data suggest that this interaction of the CSPGs with Tnr and RPTP is responsible for the meshlike appearance of PNNs and an absence of this interaction leads to the discontinuous and aggregated phenotype of PNNs seen in the Tnr KO and Ptprz1 KO mice. It should be noted that RPTP exists in mul-

RPTP/phosphacan is critical for PNNs
tiple forms, splice isoforms and proteolytic cleavage products (51)(52)(53). We have also shown that PNN component-binding can be partially recovered by adding the soluble form of RPTP (phosphacan) to dissociated neurons derived from Ptprz1 KO mice and these "rescued" neurons showed recovered binding of PNN components to the neuronal cell surface. Our data suggest that phosphacan interacts with Tnr and binds PNNs to the neuronal surface. A question therefore remains for a receptor that binds phosphacan to the neuron. RPTP is known to interact with several cell-surface adhesion molecules, including Ng-CAM, N-CAM, and Nr-CAM, as well as other cell surface molecules like contactin (54). It remains to be determined how soluble RPTP remains attached to the cell surface, as a goal of future studies. In addition we cannot formally rule out the possibility that the receptor form of RPTP plays a role in net structure in vivo. In fact humans do not make a secreted form of RPTP and therefore further studies will be needed specifically targeting the receptor form to determine its specific role in PNNs. Our work also suggests that a key nucleating event of PNN formation involves cell-surface HA; defining how HA attaches to the neuronal cell surface may provide a further key to understanding PNN specificity and structure (23,32,40).

Animals
Mice lacking the Ptprz1 gene (Ptprz1 KO) were generated as described previously (55)

Preparation of homogenates and soluble and insoluble fractions
Brain homogenates for the aggrecan release assays were derived from postnatal day 90 (PND 90) Ptprz1 WTs and Ptprz1 KOs. Tissue was homogenized in 150 mM sodium chloride and 50 mM Tris with EDTA-free protease inhibitor tablets (Roche, 1 tablet in 10 ml buffer) in a Potter-Elvehjem homogenizer. Homogenates were centrifuged at 8000 ϫ g for 10 min at 4°C. The supernatant was then removed, the pellet washed once and then resuspended in 1 ml buffer. A Bradford (Bio-Rad) assay was performed and protein concentrations were adjusted to 2.5 mg/ml. Samples were treated with 2 l chondroitinase ABC (Sigma-Aldrich) per 500 l of sample and/or 1 mM EDTA for 8 h. Samples were centrifuged again at 8000 ϫ g for 10 min to separate soluble released fraction from insoluble pellet and prepared for Western blotting. . Proposed novel PNN structural model. We propose a two-arm model for PNNs, in which each arm is uniquely attached to the cell surface. On one arm is HA-mediated cell surface binding to an unknown receptor. The other arm depends on the complex between tenascin-R (Tnr) and the secreted Ptprz1 isoform phosphacan. Our data suggest Tnr and phosphacan stabilize each other to bind to an unknown receptor. In addition, this complex is also dependent on the divalent ion Ca 2ϩ .

RPTP/phosphacan is critical for PNNs
For analysis of overall expression of ECM components by Western blotting, PND 90 brains were homogenized in 5 volumes of 25 mM Tris (pH 7.4) containing protease inhibitor mixture, EDTA-free, and processed for analysis.

Primary cortical cultures
Neuronal primary cultures were prepared as described previously (23). Briefly, cortices of embryonic day (E) 16 CD-1 WT or Ptprz1 KO embryos were removed and digested in 0.25% trypsin-EDTA (Thermo Fisher Scientific). Mixed cells were filtered and suspended in Neurobasal medium with 3% B27, 1ϫ GlutaMAX and 1ϫ penicillin-streptomycin (Thermo Fisher Scientific). Cells were then plated at a density of 2.1 ϫ 10 6 on coverslips (500 l/per well) precoated with poly-D-lysine (50 g/ml) and laminin (5 g/ml) (Sigma-Aldrich) in a 24-well dish. To remove glia, cells were treated with 5 M cyotosine arabinoside (AraC, Sigma-Aldrich) at 1 day in vitro (DIV). The medium was then changed at 3 DIV to remove AraC, and given a half change at 6 DIV. Cells were maintained at 37°C 5% CO 2 until fixation.
Phosphacan was purified by anion exchange chromatography as described previously (57). Briefly, the soluble fraction, from PND 4 CD-1 mouse brain, was filtered using a PVDF 0.22 M filter, brought to a 0.5 M NaCl concentration, and run through a 1 ml HiTrap-Q HP column using a peristaltic pump connected to an Amersham Biosciences Pharmacia RediFrac fraction collector (GE Healthcare Life Sciences). Sample was eluted over a continuous gradient of 0.5 M NaCl to 2.0 M NaCl over 10 column volumes and collected as 250 l fractions. Phosphacan-rich fractions, identified by dot blot analysis, were pooled and concentrated using 100,000 MWCO Concentrators (AmiconUltra, EMD Millipore). Approximately 250 ng of purified phosphacan was added to Ptprz1 KO cultures after the first medium change at 3 DIV and 125 ng was added after the halfmedium change at 6 DIV and analyzed at 9 DIV. When noted, coverslips were treated with 10 l ChABC for 30 min and/or 2.5 mM EGTA for 15 min. Coverslips were fixed and subsequently processed for immunocytochemistry.

Immunocytochemistry and immunohistochemistry
Primary cortical cultures plated on coverslips were fixed at 9 DIV in cold 4% phosphate-buffered paraformaldehyde (PFA) with 0.01% glutaraldehyde, pH 7.4. Cells were then blocked in screening medium (DMEM, 5% FBS, 0.2% sodium azide) for 1 h, before adding primary antibodies overnight at 4°C. The following day, Alexa Fluor-conjugated secondary antibodies (Thermo Fisher Scientific) in screening medium were added to the cells for 2 h before mounting the coverslips with ProLong Antifade Kit (Thermo Fisher Scientific). Cell nuclei were visualized with Hoechst solution (Thermo Fisher Scientific) diluted in 1ϫ PBS. For immunohistochemistry on tissue sections, PND 90 Ptprz1 and Tnr (all genotypes) mice were transcardially perfused with cold PBS (Thermo Fisher Scientific) prior to fixation with 4% PFA. Brains were postfixed overnight in 4% PFA before changing to a 30% sucrose solution diluted in phosphate buffer with 0.2% sodium azide. Using a cryostat, brains were cut as free-floating sections at 40 M and placed in phosphate buffer with 0.2% sodium azide. Sections were blocked 1 h at room temperature and then stained in either 5% milk in TBST with 1% Triton X-100 (mouse anti-tenascin-R 619, sheep anti-neurocan, rabbit anti-brevican) or screening medium with 1% Triton X-100 (WFA, mouse anti-aggrecan, goat anti-HAPLN1, and mouse anti-phosphacan). Tissue sections were additionally stained with Hoechst solution to visualize nuclei before being mounted on glass slides. Both cells and tissue were imaged using an epi-fluorescent Zeiss Imager.A2 with Nikon Elements software package. Final images were gathered and formatted using ImageJ software (58) and assembled into figures using Adobe Illustrator CC 2019.

SDS-PAGE and Western blotting
Protein concentrations were determined by Bradford assay before gel electrophoresis. For detection of CSPGs, brain lysates were treated with 2 l ChABC (Millipore Sigma, C3667, 0.1mU/l) for 8 h at 37°C to remove chondroitinase sulfate side chains and allow proper gel migration. SDS-polyacrylamide gels, used at either 4 -12% or 6 -15% gradient, were transferred to 0.45 M nitrocellulose membranes. Western blotting was conducted as described previously (59). Briefly, blots were placed in blocking buffer composed of 5% milk in low-salt TBST and then incubated in primary antibody overnight. Blots were then incubated in HRP-conjugated secondary antibodies (The Jackson Laboratory, Bar Harbor, ME) and exposed using SuperSignal West Pico Chemiluminescent Substrate or Super-Signal West Femto Chemiluminescent Substrate (Thermo Fisher Scientific). Blots were imaged using ChemiDoc MP system (Bio-Rad) or using Premium X-ray film (Phenix Research Products, Candler, NC).

Quantification and statistical analyses
The binary gap analysis was used to quantify regions devoid of PNN. For analysis, high magnification z-stack images (0.25 M steps, 63ϫ magnification) of PNNs, visualized with WFA, were taken throughout the PND 90 adult somatosensory cortex (Bregma, Ϫ2.46 mm) of Ptprz1 strain mice. To better visualize PNN surface structure, nets were flattened using the Z-project, max intensity function on ImageJ (58). Once flattened, a fixed region excluding the PNN periphery (ϳ25 m 2 ) was cropped from the estimated center of the PNN and thresholded to obtain a binary (black and white) image to analyze the PNN surface. The black/white pixel count ratio for each PNN was determined using the histogram function and averaged for each genotype.
PNN peak or node analysis was used to quantitatively describe the PNN aggregation seen on the surface of neurons in Tnr KO and Ptprz1 KO mice. Z-projected images of the PNN were processed using the local maxima function of ImageJ to identify peaks (nodes) of intense PNN staining. Once the nodes were identified, an ad hoc algorithm was used to measure the average distance between those nodes and the difference in intensity between the nodes and their surrounding space on the cell surface (node prominence) The number of unique nodes and their mean prominence was plotted for each genotype.
The quantification of ECM components in Ptprz1 KO was assessed through immunohistochemistry (IHC) and Western blot analysis. For IHC, adult coronal sections approximately at Bregma Ϫ2.46 mm were stained to detect the PNN-specific RPTP/phosphacan is critical for PNNs components aggrecan, WFA (detects CS chains, dependent on aggrecan expression) and HAPLN1, and the PNN-associated components brevican, neurocan, and tenascin-R (Tnr). For quantification of specific net components, 10ϫ large stitch images, processed by the Nikon Elements software, were taken of the cortex of Ptprz1 KO and WT mice. The pixel intensity of each specific PNN component was determined by taking a region of interest of the cortex and using the measure function of ImageJ. To quantify nonspecific PNN markers (brevican, neurocan, and Tnr), a region of interest and intensity measurement was taken of a sample set of individual PNNs (average of 25 PNNs per animal) throughout the adult cortices (images taken at 20ϫ). PNN pixel intensity was then averaged for each genotype and significance determined using an unpaired Student's t test, p Ͻ 0.05. To produce bar graphs, data were normalized and scaled.
Western blot analysis was used to further quantify the PNNspecific components aggrecan and HAPLN1. Cortices were specifically taken from Ptprz1 KO and WT mice, homogenized, separated on a 4 -12% SDS-polyacrylamide gel and transferred to nitrocellulose for blotting. Western blots for aggrecan and HAPLN1 expression were imaged using ChemiDoc MP system (Bio-Rad) and analyzed in ImageJ. Expression was determined by taking the average ratio of PNN-specific component intensity to ␤-actin intensity for each genotype. Differences were found significant at p Ͻ 0.05 (unpaired Student's t test or analysis of variance (ANOVA) with Tukey's post hoc analyses as appropriate) using GraphPad Prism 7/8 or RStudio statistical software.