Perineuronal nets are specialized ECM that encapsulate the soma and the proximal dendrites of some neurons in the CNS. This mesh-like structure has been described in various mammals [e.g., rodents (Kalb and Hockfield, 1988; Galtrey et al., 2008), cats (Hockfield, 1983), dogs (Atoji et al., 1997), sheep (Härtig et al., 2017), primates (Mueller et al., 2016), and humans (Jäger et al., 2013)] and non-mammal species [e.g., fish (Takeda et al., 2018), birds (Balmer et al., 2009), frogs (Gaál et al., 2014), and chickens (Morawski et al., 2009)]. Among all of them, spinal PNNs have been characterized in fish, rodents, cats, primates, and humans.

Perineuronal nets are irregularly distributed throughout the brain (Brückner et al., 1996) and spinal cord (Vitellaro-Zuccarello et al., 2007). In the brain, they mainly ensheath fast-spiking, gamma-aminobutyric acid (GABA)ergic parvalbumin (PV) interneurons (extensively reviewed in Härtig et al., 1992; van't Spijker and Kwok, 2017). However, PNNs also surround pyramidal neurons in the hippocampus (Carstens et al., 2016), visual, somatosensory and motor cortex (Hausen et al., 1996; Alpár et al., 2006), PV-positive and PV-negative neurons in the striatum (Lee et al., 2012), and excitatory neurons in the amygdala and deep cerebellar nucleus (Carulli et al., 2006; Morikawa et al., 2017). PNNs' heterogeneity increases even more between species. For instance, while PNNs are mainly found in sensory brain areas in the rat, primates present a larger proportion of neurons with PNNs in motor areas (Mueller et al., 2016).

Along the spinal cord, PNNs surround motoneurons (MNs) and spinal interneurons (Figure 1). Remarkably, there are differences in the proportion of PNN-enwrapped neurons throughout the spinal laminae (Figure 1B). In the dorsal horn, 20% of neurons present PNNs and specifically, none in the laminae I and II (Galtrey et al., 2008). The lack of PNNs in those laminae correlates with the grade of synaptic plasticity found in that region after injury, which is related to neuropathic pain development (Woolf et al., 1992). In contrast, in laminae VII and VIII, 50% of neurons have PNNs. This proportion has been associated with spinal interneurons including both PV- and calbindin-positive cells (Renshaw cells) (Vitellaro-Zuccarello et al., 2007). Finally, in contrast to the brain, PNNs located in the ventral horn surround large neural somas: spinal MNs (Vitellaro-Zuccarello et al., 2007; Galtrey et al., 2008). Notably, only α-MNs present PNNs in the spinal cord, unlike γ-MNs (Al'joboori et al., 2020). Possibly, the different PNNs expression around α- and γ-MNs could be explained by the electrophysiological properties of each MN type (Manuel and Zytnicki, 2019). Since PNNs formation during development is activity-dependent, the level of activity in each MN type seems to determine the presence or absence of this net. In fact, we have reported that vestigial PNNs modify the physiological properties of α-MNs (Sánchez-Ventura et al., 2021), suggesting a role in MN maturation. Considering the percentage of α-MNs enwrapped by PNNs, it was initially thought that only 30% of α-MNs had PNNs when they were stained with Wisteria floribunda lectin (WFA) (Galtrey et al., 2008). Nonetheless, selectively staining for the expression of aggrecan offers a different perspective on spinal PNNs, detecting around 80% of α-MNs with dense aggrecan-positive PNNs in rat (Irvine and Kwok, 2018) and 76% in primates (Mueller et al., 2016). These results certainly reinforce the physiological relevance of spinal PNNs.

Perineuronal nets are condensed ECM rich in CSPGs with holes at the site of synaptic contacts (Celio et al., 1998). PNNs are composed of hyaluronan (HA), CSPGs, link proteins, and tenascin-R (tn-R) (Kwok et al., 2011), which are highly organized in a ternary stable structure (Figure 2). PNNs' composition makes them different from the diffuse ECM, widely present throughout the CNS. Although both types of ECM contain HA, tn-R, and some CSPGs, the diffuse matrix lacks link proteins.

Among all PNNs components, link proteins are crucial for their formation. In vitro, the lack of the Crtl1 gene, which encodes the link protein 1, prevents PNNs formation around PNN-bearing cells (Kwok et al., 2010). In vivo, it generates attenuated PNNs in the visual cortex (Carulli et al., 2010) and aberrant PNNs in the spinal cord, with an altered proportion of its components (Sánchez-Ventura et al., 2021). Another important element in PNNs structure is aggrecan (Rowlands et al., 2018). While mice deficient for the lecticans neurocan (Zhou et al., 2001) or brevican (Brakebusch et al., 2002) present organized PNNs, aggrecan-deficient mice show altered PNNs with no WFA staining (Giamanco et al., 2010). The importance of tn-R in PNNs formation is observed in the tn-R knockout (KO) mouse whose PNNs structure is affected in both development and adulthood (Weber et al., 1999; Brückner et al., 2000; Haunso et al., 2000).

Most PNNs components, including aggrecan, hyaluronan synthase (HAS), and link proteins (Matthews et al., 2002), are produced by neurons, although astrocytes and oligodendrocytes also contribute to PNNs formation through synthesizing neurocan (Jones et al., 2003) and tn-R (Galtrey et al., 2008), respectively. PNNs are anchored to neurons by the enzyme HAS, which produces an HA polymer chain on the neuronal surface, constituting the backbone of PNNs. This conclusion was reached in view that HA receptors such as CD-44 are expressed in glial cells but not in neurons (Aruffo et al., 1990), dismissing the idea that HA receptors are the link between PNNs and neurons and accepting that HAS are the ones that attach PNNs to the neural membrane. In the spinal cord, HAS 3 is expressed in PNN-enwrapped neurons located in both the dorsal and ventral horn in adulthood. In contrast, HAS 1 is only expressed in the developing spinal cord (Galtrey et al., 2008).

Hyaluronan provides a scaffold for the binding of CSPGs. CSPGs are the major components of PNNs representing only 2% of all the CSPGs of the nervous system (Fawcett, 2015). Among CSPGs, aggrecan is the major constituent of spinal PNNs, and three other CSPGs (i.e., brevican, neurocan, and versican) are present in PNNs in variable degrees. All types of CSPGs are composed of two parts, namely, a core protein and a variable number of glycosaminoglycan (GAG) chains. The core proteins are a tridomain structure formed by a N- and C-terminal globular domain, necessary for the binding of HA and tn-R, respectively, and a central region (Yamaguchi, 2000). In the N-terminal, the interaction between HA and CSPGs is maintained by link proteins (HAPLN family genes). Although HAPLN1, 2, and 4 are found in the nervous system, in the spinal cord, only HAPLN1/Crtl1 and HAPLN4/Bral2 are described in PNN-bearing neurons (Galtrey et al., 2008). Tn-R binds to the C-terminal of the CSPGs and, thanks to its trimeric structure, facilitates the cross-link between HA and CSPGs (Lundell et al., 2004). In the central region, GAGs are covalently attached to the core protein through serine residues.

Although PNNs disposition is quite preserved, slight changes in CSPGs conformation contribute to PNNs heterogeneity. Their heterogeneity arises from the type of core protein found (Dauth et al., 2016) and/or the number and sulfation pattern of GAG chains (Miyata and Kitagawa, 2017). Antibodies against different CSPGs core protein showed a different distribution compared with the brain and spinal cord sections (Galtrey et al., 2008). The heterogeneity of the aggrecan core protein can increase even more since different glycosylation variants of this lectican also display differential distribution in different neuronal types in the CNS (Matthews et al., 2002). The GAGs can be sulfated at different positions including the 4th (CS-A), 6th (CS-C), 2nd−6th (CS-D), and 4th−6th (CS-E). Changes in this sulfation pattern can modify CSPGs charge and thus, provide binding properties to PNNs, a characteristic that also differs from the diffuse ECM (Carulli et al., 2006). Interestingly, CS-A and CS-C have different characteristics, while CS-A is inhibitory to axon growth and suppresses plasticity, CS-C is permissive to axon growth and increases plasticity (Wang et al., 2009; Miyata et al., 2012). Furthermore, this sulfation pattern may vary depending on the CNS region since WFA, which detects the non-sulfated GalNAc residues (Nadanaka et al., 2020), shows a low percentage of detection around spinal PNNs (Irvine and Kwok, 2018) compared with that in cortical ones.

Although PNNs are widely known for their role in plasticity inhibition and synaptic stabilization, other functions have been attributed to cortical PNNs such as ionic buffering, neuroprotection, and neural maturation (extensively reviewed in Kwok et al., 2011; van't Spijker and Kwok, 2017; Lorenzo Bozzelli et al., 2018; Fawcett et al., 2019). PNNs functions are tightly related to the structural properties of their CSPGs. Hence, manipulation of PNN's CSPGs has a direct impact on the function of PNNs and the neuron surrounded (Gama et al., 2006). The highly negative charge of GAGs is one of the most important factors determining PNNs functions (Brückner et al., 1993), and this can be considered in several different ways:

Importantly, the sulfation pattern can contribute to other neural functions such as maturation. After the binding to the PNNs through CS-E chains, the Otx2 protein can translocate into neurons and facilitate PV positive-neurons maturation (Sugiyama et al., 2008; Beurdeley et al., 2012). PNNs components also interact with ion channels and receptors and thus, regulate synaptic activity and membrane current. Tn-R binds to GABA receptors (GABAr) through the HNK-1 motif (Saghatelyan et al., 2000). This PNN component also interacts with subunits of voltage-gated Na⁺ channels (Xiao et al., 1999), while brevican interacts with K⁺ channels and α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAr) (Favuzzi et al., 2017). Consequently, their removal can alter the excitatory/inhibitory balance. Furthermore, the lateral mobility of some receptors is also restricted by PNNs. ECM digestion with ChABC increases the mobility of AMPAr (Frischknecht et al., 2009).

Overall, while PNNs have been classically defined as a barrier for plasticity, they also regulate numerous different neuronal functions. These functions that appear to be crucial to neural development may play adaptive or maladaptive roles after SCI, and effort is needed to improve our understanding of these functions in the spinal cord.

After an SCI, there is an upregulation of CSPGs, at the injury site, forming the so-called fibro-glial scar. Specifically, there is a significant increase in lecticans such as neurocan, versican, phosphacan, NG2, and brevican but not aggrecan (Lemons et al., 2001; Jones et al., 2003; Buss et al., 2009). These CSPGs persist in the glial scar (Silver and Miller, 2004) and are remodeled by MMP, which expression significantly increases after injury (de Castro Jr et al., 2000; Zhang et al., 2011). Apart from CSPGs upregulation, selective changes in the sulfation pattern of GAGs have been reported. Following SCI, a large increase in the 4-sulfated (CS-A) GAG chains in the injury site have been observed, with no changes in the 6-sulfated (CS-C) GAG ones (Wang et al., 2009; Hussein et al., 2020).

The glial scar cannot be simply defined as beneficial or detrimental to CNS repair since it walls the injured area but also forms a barrier for axonal regeneration (Fawcett and Asher, 1999; Sofroniew, 2010). A lot of effort has been made to find an effective strategy to overcome the inhibitory influence of GAGs and thus, promote regeneration. These strategies are extensively exposed in the last section.

The fate of PNNs located below the level of the injury is controversial since different studies have reported a reduction (Lemons et al., 2001), no changes (Al'joboori et al., 2020) or an increase (Alilain et al., 2011) in PNNs thickness. Alilain et al. (2011) showed an upregulation of CSPGs, labeled by WFA, associated with PNNs around phrenic MNs after a cervical SCI. However, they did not specifically quantify PNNs around the neuronal soma, which is needed to differentiate PNNs from the loose ECM found in the grey matter. In addition, since an upregulation of CSPGs in the brainstem was also observed after a dorsal column section (Massey et al., 2006), it was assumed that SCIs increase PNNs from any type of neuron denervated by the injury, reducing plasticity and limiting axonal regeneration. Later studies tried to address the fate of spinal PNNs far from the injury site. Reduced staining of aggrecan around lumbar MN was reported 35 days after a thoracic SCI, recovering normal levels at later stages (Sánchez-Ventura et al., 2020). Similar findings were described after hemisection in goldfish (Takeda et al., 2018). A study evaluating aggrecan synthesis and degradation found a significant decline in aggrecan levels after SCI too (Lemons et al., 2001). In contrast, a recent study described no change in the amount of WFA+ CSPGs in the lumbar ventral horn after a thoracic SCI at early time points. However, at 67 days post injury, a significant increase was found (Al'joboori et al., 2020). Taken together, these findings point to some controversy regarding the fate of spinal PNNs after SCI not only because of the marker used to assess CSPGs' changes (WFA vs. aggrecan) but also to the type of neuron evaluated. PNNs surrounding phrenic MNs can behave differently than lumbar ones, given their differing functions (breathing vs. locomotion) and innervation [low vs. high proprioceptive Ia afferents (Alvarez et al., 2004; Nair et al., 2017)]. Besides, neurons in the brainstem are not located in the spinal cord, and hence, their PNNs can present the same activity-dependent modulation than encephalic PNNs (Sánchez-Ventura et al., 2020). Finally, the chronicity and type of injury may also impact on PNNs modulation. Long-term changes can be masked by the spontaneous recovery observed in some SCI models. Indeed, the progressive locomotor recovery in injured animals can modulate PNNs, since physical activity increases PNNs around spinal MNs (Wang et al., 2011; Arbat-Plana et al., 2015; Sánchez-Ventura et al., 2020). This increase is linked to the type and intensity of the activity (Arbat-Plana et al., 2015, 2017; Sánchez-Ventura et al., 2020), as well as the ability of the animal to be active besides the physical protocol (Arbat-Plana et al., 2015, 2017; Al'joboori et al., 2020). Thus, the denervated state of neurons below the injury produced by the disruption of descending inputs and proprioceptive inactivity due to muscle paresis might reduce PNNs, reverting neurons to their developmental state before PNNs first appear. In fact, PNNs reduction after injury is accompanied by changes that resembled those seen during the critical period, including increased synapse formation and altered neuronal excitability (Arbat-Plana et al., 2015; Sánchez-Ventura et al., 2020). Hence, developmentally immature neurons and denervated neurons post-SCI share similarities in PNNs integrity and neuronal properties (Figure 4). Lumbar PNNs disintegration after SCI was accompanied by reduced expression of the KCC2 co-transporter (Sánchez-Ventura et al., 2020), probably due to the role of PNNs sustaining receptors in the cell membrane (Frischknecht et al., 2009), modifying Cl⁻ intracellular levels. In parallel, the loss of negative charges provided by the CSPGs of spinal PNNs can shift the GABAergic response of spinal neurons and thus, increase spinal excitability (Sánchez-Ventura et al., 2020). Therefore, the increased collateral sprouting and neuron excitability observed after SCI (review in Oudega and Perez, 2012) are also reported in transgenic mice with aberrant PNNs. Animals lacking the link protein 1 presented an altered ratio of excitatory and inhibitory synapses leading to hyperexcitability of spinal circuits. After SCI, these animals presented increased sprouting of the corticospinal tract (CST) (Sánchez-Ventura et al., 2021), corroborating that spinal PNNs play an important role in stabilizing synapses and limiting plasticity in the mature nervous system. Overall, while the absence of PNNs before the end of the critical period is necessary for developmental synaptogenesis, PNNs reduction after SCI can lead to uncontrolled sprouting and hyperexcitability in the spinal cord that leads to maladaptive symptoms such as neuropathic pain and spasticity (Costigan and Woolf, 2000; Boulenguez et al., 2010; Pitcher and Cervero, 2010; Côté et al., 2014; Mòdol et al., 2014). Interestingly, the blockage of MMP-9 and MMP-2 in the spinal cord inhibits the early and late phases of neuropathic pain, respectively (Kawasaki et al., 2008).

Finally, altered spinal PNNs are also observed in other pathologies such as amyotrophic lateral sclerosis (ALS) and spinal muscular atrophy (SMA). Indeed, disorganized and vestigial PNNs are found in the ventral horn of terminal SOD1 mice (Forostyak et al., 2014), an experimental model of ALS. In vitro and in vivo studies revealed that patients with SMA present a downregulation of the Ctrl1 gene, altering their PNNs (Dayangac-Erden et al., 2018).

The application of the enzyme ChABC in the fibro-glial scar is one of the most successful strategies to overcome the inhibitory influence of GAGs and consequently, promote regeneration (McKeon et al., 1991). The effectiveness of ChABC in degrading CSPGs have been tested in vitro and in vivo (Lee et al., 2010). In vivo, most of its efficacy has been evaluated in rodents, though it has also been studied in larger mammals such as cats and non-human primates (Tester and Howland, 2008; Rosenzweig et al., 2019). In most cases, ChABC has demonstrated huge potential in promoting regeneration of dopaminergic (Moon et al., 2001) and sensory axons (Shields et al., 2008), as well as, inducing sprouting of both intact and injured serotoninergic (Tom et al., 2009), corticospinal (Barritt et al., 2006; García-Alías et al., 2009), and sensory fibers (Massey et al., 2006). However, controversial results do appear in the literature too. When CST regeneration was compared in hemisected and contusion models, ChABC injections only enhanced regeneration of the CST in the hemisection model (Iseda et al., 2008). Accordingly, ChABC's effect may depend on the severity and location of the lesion, in addition to the number of spared axons and the time of the ChABC injection. Indeed, Warren et al. (2018) described that ChABC was more effective when applied chronically than acutely after a cervical hemisection.

Although axonal growth (regeneration or sprouting) is typically necessary for recovery, it is not always sufficient. Despite some studies have demonstrated improved motor, sensory and bladder function after ChABC application (Bradbury et al., 2002; Caggiano et al., 2005; Massey et al., 2006; Cafferty et al., 2008), many others found limited results. The administration of a Sema3A inhibitor enhanced axon regeneration but not recovery due to a lack of functional connections (Zhang et al., 2014). Similarly, some works have reported limited recovery after ChABC application since the plasticity promoted by this enzyme needs an appropriate interaction with their target to make functional networks (García-Alías et al., 2009; Tom et al., 2009; Harris et al., 2010; Alilain et al., 2011; Wang et al., 2011).

Indeed, the combination of ChABC or Sema3A inhibitor with specific rehabilitation further enhanced functional recovery compared with applying them alone. In this regard, it has been proposed that the plasticity achieved by ChABC or Sema3A inhibitor only establishes meaningful connections when is combined with rehabilitative training, which activates spinal circuits, increases spinal PNNs around active neurons (Arbat-Plana et al., 2015; Smith et al., 2015; Sánchez-Ventura et al., 2020), and stabilizes these connections (García-Alías et al., 2009; Wang et al., 2011; Zhang et al., 2014). The ability of rehabilitation to rewire and stabilize synapses in a functionally meaningful manner after SCI has been already reviewed (reviewed in Torres-Espín et al., 2018). Nevertheless, many questions remain unanswered regarding the mechanisms involved. Since this synaptic stabilization is comparable to that found at the end of the critical period, in which PNNs contribute, it is plausible that these nets participate in the recovery promoted by rehabilitation after SCI. As stated before, activity increases spinal PNNs' thickness, in contrast to the brain (Smith et al., 2015). Physical activity and EE following SCI resulted in increasing spinal PNNs around lumbar MN whereas reducing PNNs in the brainstem sensory nuclei (Sánchez-Ventura et al., 2020). At the spinal level, physical activity prevented PNNs decline caused by the injury (Wang et al., 2011; Arbat-Plana et al., 2015; Smith et al., 2015; Sánchez-Ventura et al., 2020), which probably contributes to the synaptic stabilization of the newly formed connections and hence promotes functional recovery (García-Alías et al., 2009; Wang et al., 2011; Zhang et al., 2014; Sánchez-Ventura et al., 2020).

The importance of PNNs integrity after SCI is also observed in the CST. Digestion of cortical PNNs located in the boundary between motor and sensorimotor cortex by ChABC perturbed anatomical and functional CST reorganization after injury, resulting in an aggravation of motor deficits (Orlando and Raineteau, 2015).

Overall, combinatorial therapies appear to be the most effective approach to enhance functional recovery after an SCI. This combination can include therapies to widen the window of plasticity and modify the glial scar, such as the enzyme ChABC, and rehabilitation to shape the newly formed connections. However, the application of this enzyme can indirectly digest spinal PNNs and entail deleterious effects on the physiology of spinal neurons. Therefore, the ChABC approach presents limitations in the context of human therapy not only due to its non-specific nature but also due to the multiple injections or large volumes needed. Hence, more precise and selective manipulations than ChABC have been proposed to enhance axonal regeneration. Transgenic mice expressing ChABC under the GFAP promoter, thus limiting ChABC expression on astrocytes, showed enhanced corticospinal regeneration at the injury site (Cafferty et al., 2007). However, given that astrocytes can also contribute to PNNs' turnover, this type of approach can also indirectly act on spinal PNNs. Similarly, targeting the mRNA of critical enzymes in CSPGs' glycosylation and elongation has been tested in vitro (Grimpe and Silver, 2004; Laabs et al., 2007) and in vivo (Grimpe and Silver, 2004) with encouraging results in regeneration around the lesion site. However, the selectivity of these molecules in the glial scar is unclear. To solve this non-specificity, antibodies against specific structures are found in the literature. On the one hand, antibodies that neutralize the CSPG NG2 have demonstrated effectiveness in increasing regeneration of sensory axons after injecting into the dorsally transected spinal cord (Tan et al., 2006). Considering that no work has previously described the presence of the lectican NG2 in spinal PNNs, the application of this neutralizing antibody could offer an alternative way to specifically target the glial scar (Galtrey et al., 2008; Irvine and Kwok, 2018). In contrast, the application of selective antibodies against 4-sulfated GAG chains improved neurite growth (Yang et al., 2017), suggesting that modulating the sulfation pattern of proteoglycans can be an alternative to ChABC. In fact, while 4-sulfated GAG is distinguished by its inhibitory properties, the 6-sulfated GAG can also be beneficial for regeneration (Gilbert et al., 2005).

The therapeutic modulation of the PNNs must strike an accurate balance. Widely, PNNs digestion can generate excessive plasticity and increase neural vulnerability to neurotoxic stimuli, whereas excessive PNNs deposition can generate increased levels of synaptic stability. Apart from the modulation provided by activity-dependent therapies such as rehabilitation or exposure to an enriched environment, PNNs can be also modulated pharmacologically. In this sense, targeting specific PNNs components, such as Sema3A protein or the sulfation patterns of GAGs, are promising approaches.