The most recent estimate of autism spectrum disorder (ASD) prevalence rose to 1 in 44 children (Walensky et al., 2021), an 18% increase from the previous rate of 1 in 54 announced by the Centers for Disease Control and Prevention (Redfield et al., 2020). With this steady increase in ASD prevalence each year, it is becoming increasingly important to identify the underlying neurodevelopmental mechanisms that contribute to ASD symptoms. Although the brain regions involved and the molecular underpinnings implicated in ASD are diverse, a diagnosis centers on social communication and sensorimotor challenges within the domain of restricted and repetitive behaviors (American Psychiatric Association, 2013). As many complex behaviors, such as language, require the integration of information from visual, auditory, tactile, and motor systems, it has been proposed that people with ASD struggle to unify multisensory information into a single percept (Baum et al., 2015; Chan et al., 2016; Stevenson et al., 2016; Robertson and Baron-Cohen, 2017; Beker et al., 2018; Zhou et al., 2018). As such, brain plasticity mechanisms are implicated in ASD and have driven research into vulnerable aspects of critical period plasticity (Wang et al., 2014; Yousif Ismail et al., 2016; Hansel, 2018; Reh et al., 2020) as well as its relationship to altered network dynamics and excitatory/inhibitory imbalance (Blatt et al., 2001; Hussman, 2001; Rubenstein and Merzenich, 2003; Gogolla et al., 2009; Levin et al., 2017; Gabard-Durnam et al., 2019).

The disruptions in critical period plasticity may lead to altered network dynamics later in life and have been connected to the function of parvalbumin (PV) positive interneurons (Reh et al., 2020). As recently reviewed in Ruden et al. (2021), PV interneurons are particularly vulnerable to a diverse range of stressors and have been linked to ASD. While the exact timing may differ between brain regions and cell-types, the closure of the critical period of plasticity coincides with the formation of perineuronal nets (PNNs) that surround the soma and proximal dendrites of PV interneurons (Cabungcal et al., 2013; Fawcett et al., 2019; Reichelt et al., 2019).

Perineuronal nets (for recent general review refer to: Shen, 2018; Bosiacki et al., 2019; Burket et al., 2021; Carulli and Verhaagen, 2021; Jakovljević et al., 2021) are a specialized and condensed form of extracellular matrix (ECM), which are consistently found to have genetic susceptibility in ASD from genome wide association studies (GWAS) (Wang et al., 2009; Weiss and Arking, 2009; Anney et al., 2010; Hussman et al., 2011). Among these genes, Reelin, a disintegrin and metalloproteinase with thrombospondin motifs (ADAMTS) and semaphorins are some of the strongest evidence linking ASD to disruption of the ECM (for review see: Pantazopoulos and Berretta, 2016; Sorg et al., 2016). Additionally, studies of PNNs have been prominent in animal models of fragile X syndrome, where reduced PNN expression has been demonstrated in Fmr1 knockout mouse auditory cortex and amygdala with relevance to altered fear-associated memory (Reinhard et al., 2019). In another study, pharmacological inhibition or genetic reduction of matrix component metalloproteinase-9 results in increased PNN production surrounding PV neurons, which normalizes auditory deficits in Fmr1 knockout mice (Wen et al., 2018; Pirbhoy et al., 2020). Xia et al. (2021) utilized a valproic acid (VPA) mouse model of ASD and found differences in intensities of PV and PNN subpopulations that possibly contribute to the progression of ASD. While these few studies show potential in exploring PNN function in relationship to ASD, the literature is lacking on specific mechanisms that may link PNNs to relevant behaviors and whether the GWAS mutations reported lead to PNN dysfunction. It is unclear whether a connection of PNNs to ASD definitively exists and, if so, whether or not the PNN changes may be a primary cause or a secondary effect due to other processes in brain development.

In contrast to animal model studies, there is a lack of literature on PNN distribution in postmortem samples from patients with idiopathic autism. Thus, the current investigation centered around regions of the brain with high PNN density and relevance to ASD-related behaviors. Postmortem brain samples were taken from both the dentate nucleus (DN) of the cerebellum that has high PNN expression (Blosa et al., 2016; Carulli et al., 2020; Hirono et al., 2021) as well as the globus pallidus (GP) of the basal ganglia (Adams et al., 2001; Cabungcal et al., 2019). The DN and GP are sources of outgoing projections to the thalamic, motor, premotor, and sensory cortices that can affect the functionality of excitatory cortical neurons and play critical roles in many ASD-related cognitive, sensory and motor behaviors. Therefore, PNN expression in these two critical regions was quantified in ASD compared to neurotypical samples. While the DN did not show differences in neuronal number or PNN expression, the GP had significantly reduced PNN expression in ASD cases that was not dependent on seizure status. Future studies may aim to clarify the role PNN and PV neurons may play in plasticity and, ultimately, an ASD diagnosis.

Given the reported PC reductions in ASD (Bauman and Kemper, 1985; Bauman et al., 1995; Bailey et al., 1998; Whitney et al., 2008; Schumann and Nordahl, 2011; Skefos et al., 2014; Hampson and Blatt, 2015), we aimed to determine whether there are further deficits in neuronal numbers within the cerebellar circuitry, as DN neurons are likely to be impacted by PC deficits. PNNs around the DN neurons were also quantified as a readout for disrupted activity from PCs using hyaluronan and proteoglycan link protein 1 (HPLN1). HPLN1 is a critical component of PNN formation and has been shown to affect critical period plasticity after knockout (Carulli et al., 2010). Following stereological principles for estimating number within a volume, five sections for each case had a contour manually drawn around each region of interest (Figures 1A–C). The counting grid was adjusted to produce an average of 150 counting boxes per case and the total number of neurons with or without a PNN were totaled then divided by the number of counting sites times the dissector volume (0.00095 mm³).

The density of neurons (Figures 1D–F) with PNNs, based on HPLN1 expression, was not different (control mean = 1,682.27 ± 417.51 and autism mean = 1,658.57 ± 228.93 neurons/mm³, p = 0.83). The density of neurons without a PNN also showed no differences (control mean = 1,487.74 ± 386.44 and autism mean = 1,462.14 ± 384.31 neurons/mm³, p = 0.84) and therefore total neuron numbers were similar (control mean = 3,170.01 ± 632.64 and autism mean = 3,120.71 ± 520.12 neurons/mm³, p = 0.79).

The shapes of the GPi and GPe (Figure 2A) allowed for circling the entire region of interest with one contour, with counting boxes filling the center. The same grid box size was utilized as for the DN, except the number of counting sites averaged 370 for the GPi and 442 for the GPe across the five sections in series of every 6th interval. In pilot studies, there were not a significant portion of neurons that were either PV positive or PNN positive only, therefore, only neurons with a clear PNN and PV filled in the center were counted. PNN stain could be seen throughout the tissue as black “lines” following neuronal processes, but only somas with PV staining were counted as separate neurons.

Neurons throughout the GPi and GPe that stained with both HPLN1 and PV were quantified and compared by diagnosis (Figure 2C), resulting in significant decreases in the ASD cases within the GPi (control mean 929.91 ± 312.81 and autism mean 687.96 ± 174.95, p = 0.0043) and the GPe (control mean 570.46 ± 178.85 and autism mean 399.54 ± 146.89, p = 0.0019).

In animal studies, degradation of PNNs occurs after seizures (Mcrae et al., 2012; Rankin-Gee et al., 2015; Yutsudo and Kitagawa, 2015) so the reduction of PNNs in the GP of ASD cases could be a result of seizure activity instead of a common feature across individuals. Within this cohort, 7/20 ASD cases also had a diagnosis of epilepsy, but comparing group means (Figure 3A) showed that the cases with seizures were not driving the decreases in lower PNNs for the ASD group as a whole. Instead, there was a trend toward higher PNN density in the seizure cases compared to the ASD only cases that did not reach significance (GPi p = 0.24, GPe p = 0.07).

Perineuronal net density increases with age in human postmortem samples (Rogers et al., 2018). Therefore, PNN density was compared to age (Figure 3B), which showed a tendency to increase with age in controls, but was slightly decreased in ASD within the GPi. However, the difference between slopes measured by a Student’s t-test was not quite significant (p = 0.093). Cases had relatively the same slope across age in the GPe (p = 0.57). Intercepts of the line, representing PNN density between groups, were significantly decreased in ASD cases for both the GPi (p = 0.0048) and the GPe (p = 0.0022).

Since the numbers of PNNs in ASD tissue were lower and the tissue often appeared to have a generally lighter stain (Figure 2), fresh frozen tissue from the opposite hemisphere from the same cases were assessed with western blotting in an attempt to quantify decreases in PNN component levels. However, these data were unreliable, as individual components of nets that should increase or decrease together, such as brevican and HPLN1, were highly variable even when measured from the same case on the same blot. Therefore, we did not include the results from these analyses, as discussed in the section “Limitations”.

Although our number of cases included for comparisons is significantly higher than typically used in postmortem studies, the variability inherent to tissue processing and individual variation makes interpretation challenging. This is especially true for PNN analyses because, as other researchers have noted, variable fixation practices after brain collection, differences in collection strategies and the difficulties of extracting PNN components from surrounding tissue in a homogeneous manner all culminate in variance within the dataset (Rogers et al., 2018). In particular, even rigorous homogenization of tissue to release PNN components into solution proved difficult, as western blots were highly variable across cases and within a case for the different PNN component levels. As such, other strategies will need to be considered for quantifying amount of expression between cases. We quantified regions (DN, GPi, and GPe) as a whole, so while we did not observe decreases in neuronal number within the DN, the results could be different if the area was subdivided into functional regions based on input from ASD affected areas in the cerebellar hemisphere. Since we did not have access to sections throughout the entire brain region of interest, true stereological assessments were not possible and will be useful for future studies when consistently fixed regions throughout an entire structure are available. While we observed significant density decreases in PV projection neurons with PNNs in the GP, it will be important to consider how this difference intersects with volume differences of the GP in ASD. Again, studies where the entire span of the GP can be dissected and therefore the total area of each region can be quantified and sampled unbiasedly will provide more definitive assessments of neuron density as compared to volume changes.

Despite reports of significant reductions in the number of PCs in the lateral hemisphere of the cerebellum, the targets of PC output within the DN appear to be sustained by the remaining PCs and inferior olive input. Individual neurons within the DN are each contacted by hundreds of PCs (Chan-Palay, 1973; Baumel et al., 2009) so it is reasonable that loss of a percentage of PCs would not impact numbers of DN neurons. The proportion of DN neurons surrounded by PNNs compared to those without appear to be unaltered as well. Future work may reveal differences in expression of PNN components that may be dependent on activity, but the variability inherent to current human postmortem methods did not allow for comparison of expression levels with IHC or western blots (see section “Limitations”).

Within the GP, PNNs mainly surround PV positive projection neurons. This is in contrast to the DN neurons, in which PNNs mainly surround excitatory projection neurons that have input from PV positive PCs. Therefore, differences in ASD could be due to the different functional roles of these neurons subtypes. Both the GPi (with a higher percentage of PNNs) and the GPe had lower density counts in ASD tissue. It is typical for postmortem ASD studies to include seizure status as a covariate, as seizures can lead to many alterations and may be treated with medications that can affect the components being analyzed. One study measured PNN degradation and found that seizures lead to shifts in expression of PNN components, including HPLN1 (Mcrae et al., 2012), which may be due to release of matrix metalloproteinases following status epilepticus (Dzwonek et al., 2004; Dubey et al., 2017). However, a human postmortem study of PNN degradation after epilepsy did not show dysregulation of PNNs (Rogers et al., 2018), which may be dependent on the PNN markers examined and timing following seizures (reviewed in Chaunsali et al., 2021). Consequently, it is not clear whether PNNs degrade similarly in humans or whether the degradation is transient and potentially stabilizes over time.

Since the basal ganglia in general has been reported to have volume differences (for review: Subramanian et al., 2017) and receptor differences (Brandenburg et al., 2020) in ASD, the alterations in PNN density could be due to a general phenomenon of altered activity-dependent mechanisms. Emerging research implicates PV network dynamics in both critical period plasticity and adult learning, where experience related plasticity modulates learning and is dependent on a high or low PV network configuration (Hensch, 2005; Donato et al., 2013). Given that PNNs regulate input onto the soma of PV positive neurons, their digestion with chondroitinase ABC can alter this input and have effects on gamma oscillations, a key component of the role of PV neurons in network dynamics (Carceller et al., 2020). The development of complex skills, such as language and social communication, may be dependent on the function of PV neurons and PNNs in the temporal alignment of critical periods of plasticity across brain regions (Carulli and Verhaagen, 2021). Accumulating evidence implicates vulnerability of PV neurons in neuropsychiatric conditions (Ruden et al., 2021), making them intriguing targets for research and treatment strategies in ASD. The current study suggests alterations in PV neurons within the GP, which warrants further investigations into the function of these specialized neurons, as they are projection neurons that likely have different physiological functions compared to the typically studied PNN positive PV interneurons in cortical regions. As restricted and repetitive behaviors are core symptoms of an ASD diagnosis and the BG is a key regulator of repetitive behaviors, targeting these cell types could be valuable in mitigating undesirable symptoms.

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

CB: conceptualization and investigation. CB and GB: methodology, writing – original draft, writing – review and editing, and funding acquisition. GB: supervision. Both authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.