1.

Introduction

Recent years have witnessed a rapidly growing number of studies focusing on the intriguing functions of perineuronal nets (PNNs), unique specialized polysaccharide-based extracellular matrix structures.^1–4 PNNs form during the end of the critical period and are important mediators of neuronal plasticity, learning, and memory.⁵ Furthermore, it has been suggested that PNNs also protect neurons from oxidative stress⁶ and damage elicited by activated microglia,⁷ and mediate transmission of pain signals from the spinal cord.⁸ These extracellular nets enwrap mostly inhibitory parvalbumin-positive (${\mathrm{PV}}^{+}$) interneurons and are proposed to regulate their activity.⁹ Accordingly, they affect the synchronization of neuronal oscillations,^10–12 thus modulating consequent information processing and consolidation. The late Nobel laureate Roger Tsien postulated that PNNs are the primary structures for holding life-long memories,¹³ pointing to their stability as a key pillar in the basis of this hypothesis. Related to this, it has been shown that Alzheimer’s patients lose about two-third of their PNNs.¹⁴ The view of PNNs as stable structures has been challenged by recent studies that found rapid changes in the number of PNNs following fear conditioning and extinction of learning.¹⁵ However, to better understand when and how PNNs form and degrade, the ability to capture these dynamic processes is essential.

Currently, the role of PNNs is studied using ex vivo stainings to quantify their distribution¹⁶ and structure¹⁷ or in combination with electrophisiological recordings in acute slices⁹ to understand their role in activity. Unfortunetly, ex vivo recordings from tissue slices can be performed only within a short-time frame after their preparation and are only performed on partial neuronal systems with incomplete microenvironments.¹⁸ Enzymatic degradation of components of the PNNs, such as the chondroitin sulfate proteoglycans (CSPGs)¹¹ or hyaluronic acid,¹⁹ is another prevalent approach for studying their role. However, due to the inability to follow the dynamic process of degradation and assembly of PNNs in vivo, the scope of this effective tool remains limited in its use for correlative assesment of phenotypic outcomes.¹³

Hence, although existing experimental approaches have advanced our understanding regarding the important roles of PNNs in brain function, their assesment of longitudinal and dynamic processes, such as sensory processing, learning, memory, and neuroprotection, is limited.^4,20 We set here to address the unmet need for advanced methods allowing to quantify dynamic processes associated with PNNs and PNN-positive (${\mathrm{PNN}}^{+}$) neurons within their native microenvironments. Nearly three decades ago, a study showed that stereotactic intracranial injections of biotinylated Wisteria floribunda agglutinin (WFA) in rats, followed by ex vivo streptavidin incubation, resulted in specific staining of PNNs.²¹ We build upon this pioneering work, which was never used in vivo, and introduce a straightforward approach for direct labeling of PNNs in mice via intracranial injections of fluorescently labeled WFA. This procedure is particularly well-suited for in vivo imaging, as demonstrated here with two-photon imaging. This approach allows, for the first time, to longitudinally image PNNs, observe the dynamics of PNN breakdown and de novo synthesis, as well as to segregate and monitor neuronal activty of ${\mathrm{PNN}}^{+}$ cells, all with high specificity and with subcelluar resolution in a minimally invasive manner.

WFA is a lectin that binds to the glycosaminoglycan chains of the CSPGs found abundantly in the PNNs. Staining with WFA is among the most used approaches for visualizing PNNs in vitro,^22,23 in histological sections,¹⁶ and in acute slices.⁹ Expanding this experimental strategy for in vivo applications, we injected either fluorescein isothiocyanate (FITC)-, Atto 590-, or Alexa Fluor™ 594 (Alexa594)-conjugated WFA into the cortical parenchyma of mice implanted with a cranial window, followed by intravital two-photon imaging [Fig. 1(a)]. We used a monomeric version of WFA to minimize its interference with physiological processes and show that the in vivo labeling of PNNs is robust (Fig. 1 and Figs. S1–S3 in the Supplementary Material) does not affect intrinsic electrophysiological properties of stained neurons (Fig. S4 in the Supplementary Material) or enzymatic degradation (Fig. S5 in the Supplementary Material), rapid [Fig. 1(e)], specific [Fig. 1(g) and Figs. S1–S3 in the Supplementary Material], and stable [Figs. 2(a) and 2(b)], making it ideal for longitudinal imaging. We start by demonstrating the use of this tool to characterize PNNs in a neurodevelopmental disease mouse model of fragile X syndrome [Figs. 1(h) and 1(i)]. We chose fragile X syndrome as a model disease, as the number of PNNs have been repeatedly shown to decrease in multiple studies and were shown to have an important role in this disease (see Ref. 4 for a review).

Fig. 1

Intracranial injection of fluorescent-WFA rapidly and robustly labels PNNs for longitudinal intravital imaging. (a) A diagram representing the workflow used in the study. Stereotactic injection of fluorescent-WFA was followed by intravital two-photon imaging for measuring activity in awake animals, for observing PNNs, or for monitoring enzymatic degradation and synthesis of PNNs. Our system allowed multiple injections and longitudinal imaging. (b) Injection of FITC-conjugated WFA allowed in vivo imaging of PNNs in high resolution with minimal unspecific staining. Top image represents a 3D-reconstruction of a partial stack (200 to $350\mu \mathrm{m}$ deep into the cortex), and bottom image represent a $z$-projection of a single PNN covering the neuron soma and initial segments of the dendrites. Green, WFA and magenta, blood vessels. (c) The number of PNNs counted in layers 2 to 4 in vivo ($n=4$ mice) was similar to the number of PNNs counted in histological sections ($n=3$ mice; layers 2/3, $p=0.8883$; layer 4, $p=0.9094$), whereas in both in vivo and ex vivo, we found more PNNs in layer 4 compared with layers 2/3 (in vivo, $p=0.0022$; ex vivo, $p=0.0104$; ordinary one-way ANOVA, $F_{(\mathrm{3,16})}=10.82$, $p=0.0004$). Box plots represent median with min–max values. (d) PNN staining (green) in ${\mathrm{PV}}^{+}$ tdTomato (magenta) mice. (e) PNNs were visiable in vivo as early as 1 h following intracranial injection of WFA-FITC, and SNR more than doubled in the first 6 h and remained stable thereafter. SNR was calculated based on maximum intensity $z$-projections as median value of free-form encompasing the PNN over median value of square in the background, as indicated in the bottom right image. Red arrowheads point to specific staining that improved over time and yellow arrow heads point to nonspecific staining that reduced or disappeared over time. (f) Quantification of the volume stained by a single injection of $1.5\mu \mathrm{l}$ WFA-FITC at 50 to $500\mu \mathrm{m}$ from pial surface, 4 days following injection. (g) Histological analysis of brains 60 days following WFA-FITC (green) injection revealed that all PNNs around the injection site were labeled, as indicated by ex vivo staining (magenta) with WFA-Alexa Fluor 561. (h) In vivo PNN imaging allowed visualizing PNNs in Fmr1-knockout mice. (i) Quantification of the PNNs indicated there was significantly less PNNs in Fmr1-knockout mice compared with WT mice (Welch’s two-sided unpaired $t$-test, $t(4)=3.159$, $p=0.0277$). Box plots represent median with min–max values. Images represent a $30\text{-}\mu \mathrm{m}$ maximum intensity $z$ projection. Scale bars are $50\mu \mathrm{m}$ for (b)–(e), and (h), $200\mu \mathrm{m}$ for (f)–(g), and $20\mu \mathrm{m}$ for insets in (g). See also Figs. S1–S3 in the Supplementary Material and Videos 1–4 (Video 1, MOV, 11.4 MB [URL: https://doi.org/10.1117/1.NPh.10.1.015008.s1]; Video 2, MP4, 1.44 MB [URL: https://doi.org/10.1117/1.NPh.10.1.015008.s2]; Video 3, MOV, 10.9 MB [URL: https://doi.org/10.1117/1.NPh.10.1.015008.s3]; and Video 4, MOV, 11.2 MB [URL: https://doi.org/10.1117/1.NPh.10.1.015008.s4].).

Fig. 2

In vivo labeling of PNNs proves to be stable, while enabling monitoring degradation and de novo synthesis of PNNs. (a) Longitudinal imaging of the same PNNs labeled with WFA-FITC over a period of 35 days. No loss of signal was detected. (b) In vivo injection of WFA-Alexa594 stained PNNs specifically and was stable for at least 60 days following injection. (c) Timeline for in vivo monitoring of degradation and de novo synthesis of PNNs. (d) A sketch representing the window and cannula preparation for multiple intracranial injections. (e) Longitudinal imaging with multiple injections of WFA-FITC and chABC allowed monitoring the enzymatic destruction process of the PNNs and their reconstruction. “Day ($X$)” indicates the time relative to chABC injection. Red arrowheads point to PNNs that underwent degradation and yellow arrowheads point to newly formed PNNs. Images represent a $30\text{-}\mu \mathrm{m}$ maximum intensity $z$ projection. Scale bars are $50\mu \mathrm{m}$.

Next, we monitor their stability over time [Figs. 2(a) and 2(b)] as well as their degradation and reconstitution¹⁹ following chondroitinase ABC (chABC) injection [Figs. 2(c)–2(e)]; this also strengthens the notion that the staining agent does not interfere with enzymatic degradation in vivo.

We further demonstrate the feasibility of simultaneously combining calcium imaging with this labeling approach to monitor the activity of PNN-enwrapped neurons [Figs. 3(a)–3(c)], paving the road for a direct study in vivo of ${\mathrm{PNN}}^{+}$ and PNN-negative (${\mathrm{PNN}}^{-}$) PV interneurons [Figs. 3(d) and 3(e)].

Fig. 3

Intravital calcium imaging of mice with labeled PNNs. (a) A representative GCaMP6s-labeled field of view (FOV) with markings of some of the detected active components. The $290\times 290\mu \mathrm{m}$ FOV was captured at 30 Hz, with the depicted image being 1000 frames average for display purposes only. Inset – 30 frames averaged image of ${\mathrm{PNN}}^{+}$ cells (WFA-Atto590; magenta) overlaying the GCaMP image (grayscale). (b) Representative $\mathrm{\Delta }F/F$ traces of the marked cells with a PNN (magenta) and without a PNN (black). (c) Representative $\mathrm{\Delta }F/F$ traces of ${\mathrm{PNN}}^{-}$ (left; black) and ${\mathrm{PNN}}^{+}$ (right; magenta) neurons at different cortical depths (measured from pia) in a mouse injected with WFA-FITC and RCaMP7. Vertical lines in the traces indicate the timing of whisker stimulation and are scaled to 300% $\mathrm{\Delta }F/F$ change. (d) A representative FOV of ${\mathrm{PV}}^{+}$ cells expressing labeled with WFA-Alexa594 (magenta) and GCaMP7f (gray) with marking of some of the detected active components (as detected by CaImAn), with their corresponding $\mathrm{\Delta }F/F$ traces of the marked cells with a PNN (magenta) and without a PNN (black) (see Video 5, MOV, 1.52 MB [URL: https://doi.org/10.1117/1.NPh.10.1.015008.s5]). (e) The calcium transient event rate in the barrel cortex was similar between ${\mathrm{PNN}}^{+}$ and ${\mathrm{PNN}}^{-}$ PV cells, while the mean AUCs of the $\mathrm{\Delta }F/F$ traces were significantly higher in ${\mathrm{PNN}}^{+}$ PV cells compared to ${\mathrm{PNN}}^{-}$ cells in the same mouse ($n=3$ mouse, 147 ${\mathrm{PNN}}^{+}$ cells and 1025 ${\mathrm{PNN}}^{-}$ cells, unpaired two-tailed student $t$-test, $t=2.6066$, $p=0.0092$). Moreover, the variance of the mean AUCs of ${\mathrm{PNN}}^{-}$ cells was significantly larger compared to ${\mathrm{PNN}}^{+}$ PV cells (Levene’s test, $p=0.0092$). Data were averaged to 1 fps for display purposes only. Scale bars are $50\mu \mathrm{m}$.

2.

Results

Injection of a commercially available WFA-FITC lectin to the somatosensory cortex resulted in specific PNN staining [Fig. 1(b) and Video 1]. Following fluorescent-WFA injection, PNNs were visible in vivo in layers 2/3 (75 to $200\mu \mathrm{m}$ below the pial surface) and more abundantly in layer 4 (200 to $400\mu \mathrm{m}$; $n=6$ mice; $1440\pm 398.80$ versus $2630\pm 577.10$ WFA-positive cells, mean ± SD, $p=0.0022$; Figs. S1 and S2 in the Supplementary Material). These values were comparable to the number of PNN-bearing cells observed in the same location following ex vivo staining [$n=4$ mice; $1185\pm 445.90$ in layers 2/3 versus $2391\pm 452.60$ in layer 4, mean ± SD, $p=0.0104$; Fig. 1(c)], matching previously reported values measured ex vivo.^16,24 Similarly, ${\mathrm{PNN}}^{+}$ cells were only sporadically found in layer 1, while some large blood vessels were partially stained (Fig. S2 in the Supplementary Material and Video 1).

Past ex vivo histological studies pointed to the fact that most of the PNNs enwrap ${\mathrm{PV}}^{+}$ inhibitory neurons while not all ${\mathrm{PV}}^{+}$ cells are enwrapped by a PNN.²⁵ To demonstrate the feasibility of using in vivo staining of PNNs to segregate between these ${\mathrm{PV}}^{+}$ populations, we injected WFA-FITC into brains of transgenic mice that express tdTomato in most of these cells (i.e., ${\mathrm{PV}}^{+}$-tdTomato). This approach allowed to clearly differentiate between ${\mathrm{PV}}^{+}$ cells with and without PNNs [Figs. 1(d), 3(d), and 3(e) and Videos 2 and 3].

Remarkably, PNNs were visible within 1 h following injection of fluorescent-WFA, and after 2 h we did not detect staining of new components. Unbound fluorescent WFA lectin was washed away rapidly or diluted in the extracellular fluid, whereas the bound lectin remained intact. Consequently, the signal-to-noise ratio (SNR) of the staining improved by twofolds reaching a maximum within 6 h following the injection of fluorescent WFA [Fig. 1(e)].

Injection of $1.5\mu \mathrm{l}$ of WFA-FITC in a single location at depths ranging between 50 and $500\mu \mathrm{m}$ from pial surface resulted in staining of PNNs in an average volume of $943.10{\mu \mathrm{m}}^3$ ($n=4$; $\mathrm{SD}=198.84{\mu \mathrm{m}}^3$). The fluorescent WFA diffused in the lateral axis up to $\sim 2\mathrm{mm}$ and reached as deep as $\sim 1.75\mathrm{mm}$ [Fig. 1(f)].

Histological sections of brain tissues were analyzed 60 days following in vivo injection of WFA-FITC [Fig. 1(g) and Video 4]. This analysis confirmed that PNNs were robustly, specifically, and stably labeled. Ex vivo staining of the sections with WFA conjugated to Alexa 561 [Fig. 1(g)] or with antiaggrecan antibody (Fig. S3 in the Supplementary Material) proved that the in vivo staining was specific to PNNs and stained all the PNNs in the injection area. To test the use of our approach in a disease model, we monitored PNNs in vivo in a mouse model of fragile X syndrome [i.e., Fmr1-knockout mice; Figs. 1(h) and 1(i)] and found considerably less PNNs in these animals ($p=0.0277$), similar to previously described results from ex vivo experiments⁴ highlighting the use of our method to study in vivo this and other pathologies related to PNNs.

Importantly, labeling of the PNNs with either WFA-FITC, WFA-Atto590, or WFA-Alexa594 was stable, enabling longitudinal imaging of the same PNNs at subcellular resolution over more than 8 weeks [Figs. 2(a) and 2(b)]. Together, these findings indicate that the presented approach allows reliable monitoring PNNs very early after injection of fluorescent WFA as well as for weeks to follow.

In vivo detection of PNNs is critical for monitoring dynamic processes, such as degradation, synthesis, and activity. To this end, we monitored enzymatic degradation and reconstitution of PNNs in vivo [Figs. 2(c)–2(e)], a phenomenon that to date was studied only ex vivo.¹⁹ Although in vivo enzymatic degradation with chABC occurred within a similar time frame as previously reported,^11,26,27 it is still possible that the labeling interferes with intrinsic enzymatic processing, carried out by proteases such as matrix metalloproteinases (MMPs) and a disintegrin and metalloproteinase with thrombospondin motifs (ADAMTSs).^4,28–30 To test this, we assessed whether ADAMTS4, a proteoglycanase found in the brain in physiological and pathological conditions,³¹ can degrade labeled PNNs. We found that PNNs labeled with WFA-FITC were degraded by low concentration (10 nM) of the enzyme, indicating that the labeling itself does not significantly affect the stability of PNNs (Fig. S5 in the Supplementary Material).

Longitudinal monitoring of neuronal activity in behaving mice is another essential aspect for understanding the role of PNNs. To demonstrate the applicability of our method in this context, we conjugated a monomeric WFA lectin to Atto 590 (see Sec. 4) and injected it together with an AAV-hSyn-GCaMP6f virus, or injected WFA-FITC together with an AAV-hSyn-RCaMP7.01, to induce expression of the calcium indicator in all neuronal populations in the injected area. We show the activity of single neurons and correlate it with the presence of PNNs [Figs. 3(a)–3(c)]. Next, we investigated the difference in activity dynamics of ${\mathrm{PV}}^{+}/{\mathrm{PNN}}^{+}$ compared with ${\mathrm{PV}}^{+}/{\mathrm{PNN}}^{-}$ interneurons by injecting WFA conjugated to Alexa Fluor™ 594 (see Sec. 4) together with AAV-floxed-GCaMP7f into ${\mathrm{PV}}^{\mathrm{cre}}$ mice [Fig. 3(d)]. First, to ensure that the labeling of the PNNs does not alter the activity of the enwrapped neurons, we quantified the electrophysiological properties of labeled and nonlabeled PV-interneurons in primary cultures (Fig. S4 in the Supplementary Material). Electrophysiological measurements indicated that WFA labeling did not affect firing rates ($p=0.6456$), action potential threshold ($p=0.5305$), action potential half width ($p=0.4920$), or membrane resistance ($p=0.8377$). Notably, other studies have used WFA for labeling PNNs during electrophysiological recordings in acute slices and cultures and did not report any methodological artifacts;^9,32 further justifying the approach presented here. In vivo, we found that while the detected calcium transient event rate in the barrel cortex was similar in ${\mathrm{PV}}^{+}/{\mathrm{PNN}}^{+}$ and ${\mathrm{PV}}^{+}/{\mathrm{PNN}}^{-}$ interneurons, the mean and median area under the curve (AUC) of ${\mathrm{PV}}^{+}/{\mathrm{PNN}}^{+}$ were larger [$p<0.01$ and $p<0.05$, respectively; Fig. 3(e)]. Furthermore, the variance between the neurons’ means and medians AUC was much larger in cells without PNNs ($p<0.01$ and $p<0.05$, respectively). The results herein are not directly comparable with previous studies that investigated the regulating role of PNNs on PV activity, because they used electrophysiological recordings to measure the effects of removal of PNNs or one of their components.³³ Complete removal of PNNs by enzymatic degradation or removal of specific PNN components by genetic manipulations resulted in most cases in the reduction of firing rate and increased spiking variability. Although these are useful tools, they neither allow to elucidate if PNNs enwrap neurons with intrinsically different activity characteristics nor do they allow to elucidate any mediating role of PNNs on such activity under physiological conditions. Future studies should combine in vivo labeling of PNNs and activity monitoring (using calcium or voltage imaging or electrophysiological reordering) with manipulations on PNNs to tackle these critical gaps in our current knowledge.

3.

Discussion

Overall, we demonstrate a simple and robust method for specifically detecting PNNs in vivo using commercially available reagents. We prove this staining is specific and stable over time, enabling longitudinal in vivo imaging without affecting the underlying physiological properties or the efficacy of enzymatic degradation of the PNNs. We also track longitudinal and dynamic processes, such as degradation and reconstitution of PNNs in normal and pathological conditions. We further demonstrate that this approach is compatible with the most common calcium imaging indicators (i.e., GCaMP and RCaMP) and can be used to monitor activity of ${\mathrm{PNN}}^{+}$ neurons in a comparative manner within their natural environments while maintaining their integrity and physiological interactions. Utilizing this labeling approach, we provide initial evidence that PNNs are found around ${\mathrm{PV}}^{+}$ interneurons with distinct activity characteristics, including apparent increased and more stable firing properties. Future work should investigate whether the PNNs enwrap two, or more, distinct PV populations or that the presence of PNNs themselves changes the activity of the cells through processes, such as stabilization of synapses² or binding of metabolites and growth factors.⁴

The PNN-labeling approach presented herein involves an intracranial injection and therefore has some limitation compared with genetically encoded labeling, such as induction of local and transient inflammation or the need of repeated injection for de novo synthesis studies [as done in Figs. 2(c)–2(e)]. However, transgenic models that allow in vivo PNN visualization are not yet available. In addition, in vivo imaging with high spatial resolution at depths, PNNs are present necessitates implantation of a cranial window, which, by itself, unavoidably induce similar transient inflammation. Furthermore, extrinsic labeling approaches also have advantages over transgenic models, such as reduced costs, time, and animal breeding requirements, as well as providing improved labeling robustness. We foresee that like the historical development of calcium and membrane potential indicators, vast amounts of knowledge can be gained using a synthetic indicator (i.e., WFA labeling) until their genetically encoded counterparts become available and mature while we attain the performance level of the synthetic ones, a process that can take several years. With the growing interest in PNNs, we expect that this approach will become a groundbreaking tool for studying the intricate physiological role of PNNs in vivo in health and disease. We expect the application of the technique presented here to elucidate the role of PNNs in regulating activity of PV cells, the interaction of PNN with glial cells and more, during development and disease progression.

4.

Materials and Methods