The striatum is a highly integrative structure. In rodents, each of the ~2.5 million GABAergic medium-sized spiny projection neurons (SPNs) receives ~25,000 glutamate afferents from nearly all areas of the cortex and thalamus (Kincaid et al., 1998; Doig et al., 2010). These are modulated by nigrostriatal dopamine and form the only striatal output pathways (Tritsch and Sabatini, 2012). As the gateway to the basal ganglia, the striatum mediates action selection, motor control, motivation, and learning (Friend and Kravitz, 2014), and its dysfunction is implicated in neurodevelopmental disorders, addiction, and neurodegeneration (Graybiel et al., 2000; Smith et al., 2009; Gerfen and Surmeier, 2011). In contrast to the prenatal topographic development of sensory cortices, striatal circuit organization is slow and requires the functional maturation of cortical and thalamic excitatory inputs throughout the first postnatal month (Tepper et al., 1998). During this time, glutamate release promotes the formation and stabilization of excitatory synapses on SPNs (Kozorovitskiy et al., 2012, 2015). While mechanisms regulating synaptic development, maintenance, and plasticity are quite well described at glutamate synapses in principal excitatory neurons of the neocortex, much less is known of how these mechanisms translate to subcortical areas such as the striatum. Here, we developed assays to investigate excitatory synapse plasticity in a striatal in vitro system, where glutamate activity can be examined in isolation from dopamine and other neuromodulators.

Dendritic spines are specialized excitatory post-synaptic structures, which are thought to compartmentalize signaling processes to regulate glutamate receptor activation, calcium flux, cytoskeletal remodeling, membrane trafficking, and protein synthesis/degradation (Bourne and Harris, 2008; Yoshihara et al., 2009). Activity-dependent morphological changes in dendritic spines and associated presynaptic elements modulate neural function, with growth, pruning, and remodeling likely underlying cognitive processes (Villalba and Smith, 2013; Sala and Segal, 2014), and spine loss being a potential structural correlate of cognitive deficits (Penzes et al., 2011).

Excitatory synapse development, spine formation, and dynamics have been extensively studied in hippocampal/cortical pyramidal neurons, both in vivo and in vitro. Long-term potentiation (LTP) and long-term depression (LTD)-like paradigms have received particular attention, as these lasting activity-dependent modifications are considered the leading cellular model for learning and memory (Bliss and Collingridge, 1993). Spines alter shape and receptor composition in response to plasticity induction paradigms, with studies showing that LTP is typically associated with spine swelling (Matsuzaki et al., 2004; Okamoto et al., 2004; Tanaka et al., 2008) or de novo spine formation (Maletic-Savatic and Malinow, 1998; Engert and Bonhoeffer, 1999; Goldin et al., 2001), whereas LTD is associated with spine shrinkage (Okamoto et al., 2004; Zhou et al., 2004) or spine loss (Nägerl et al., 2004).

In recent genetic models where glutamate transmission is absent, principle excitatory neurons still develop normal dendritic architecture and spine numbers in vivo (Sando et al., 2017), a recent finding that adds weight to the traditional view that neuronal activity serves as a mechanism of refinement after synaptic connections are established (LeVay et al., 1980; Katz and Shatz, 1996; Sanes and Lichtman, 1999; Huberman et al., 2008). Conversely, there is evidence to support that activity is also important for synapse formation per se (Sabo et al., 2006; Andreae and Burrone, 2014; Choi B. J. et al., 2014; Choi S. H. et al., 2014; Okawa et al., 2014), reviewed in Andreae and Burrone (2018). In neuronal cultures, chronic action potential silencing throughout synaptogenesis reduces spine density by ~50% (Kossel et al., 1997), whereas after synapses form, a 3-day (but not 24 h) silencing period reduces spine number by only ~15% (Papa and Segal, 1996), suggesting that activity throughout the first 3 days contributes to synapse (and consequently spine) formation.

Alternatively, the observations that spine formation occurs on pyramidal neurons in the absence of vesicular glutamate release (Sando et al., 2017; Sigler et al., 2017), but is reduced in chronically-silenced cultures (Kossel et al., 1997), raise the possibility that early stages of spine formation may be regulated by GABAergic activity; inhibitory neurons are present in both scenarios, and GABA transmission (which is depolarizing in early development, and thus excitatory), is blocked in silenced cultures. Against this suggestion as a general rule, striatal cultures (almost entirely comprised of GABAergic SPNs) fail to generate appropriate dendritic arbors or dendritic spines in the absence of glutamatergic neurons, but do when co-cultured with cortical or hippocampal neurons (Segal et al., 2003; Kaufman et al., 2012; Fasano et al., 2013; Paraskevopoulou et al., 2019). Chronic silencing prevents spine formation even on co-cultured SPNs, which will develop spines within 2 h of TTX wash-out (Segal et al., 2003); this suggests connections are made despite silencing, and that spinogenesis specifically requires action potential-dependent transmission. Thus, SPNs require glutamatergic input to develop their eponymous morphology, and their dendritic spines appear to be highly plastic.

Alterations to striatal dendritic structures and synaptic plasticity are observed in multiple disorders, with evidence from many studies suggesting a role in the pathophysiology of Parkinson’s disease, Alzheimer’s disease, schizophrenia, and autism (McNeill et al., 1988; Day et al., 2006; Milnerwood and Raymond, 2010; Penzes et al., 2011; Villalba and Smith, 2013; Sala and Segal, 2014; Volta et al., 2017). Most knowledge of striatal plasticity comes from electrophysiological studies in acute brain slices; these have demonstrated a propensity to presynaptic plasticity and long-term depression, as well as the importance of neuromodulation by dopamine (Calabresi et al., 1997; Spencer and Murphy, 2000; Wang et al., 2006; Kreitzer and Malenka, 2007; Sergeeva et al., 2007; Li et al., 2009; Lovinger, 2010; Blackwell et al., 2019). That said, glutamate uncaging has also been shown to be sufficient to trigger de novo spine formation on SPN dendrites in ~50% of trials (Kozorovitskiy et al., 2012, 2015), similar to basal rates in cortical pyramidal neurons (Kwon and Sabatini, 2011). Elsewhere, Shen et al. (2008) demonstrated that dopamine was necessary for determining the directionality, but not necessarily the induction, of spike-time dependant plasticity in cultured striatal slices. Thus, glutamatergic modulation of striatal dendritic spines merits further attention, both during development and in response to activity-dependant plasticity.

Here, we examined how glutamate transmission, in the absence of dopamine, modulates SPN dendritic spine development, plasticity, and associated synaptic markers. Using an in vitro cortico-striatal co-culture system (Segal et al., 2003; Tian et al., 2010; Randall et al., 2011; Kaufman et al., 2012; Milnerwood et al., 2012; Penrod et al., 2015), we investigated the effects of blocking action potential-dependent network activity, manipulating spontaneous AMPA receptor (AMPAR) and NMDA receptor (NMDAR) activity, and the effects of an NMDAR-dependent LTP induction paradigm. Long-term glutamate silencing (>24 h) induced presynaptic alterations, reduced spine density, and had variable effects on filopodia, while short-term silencing (<3 h) did not. The LTP induction paradigm rapidly induced spine and GluA1 cluster changes, consistent with LTP-like modifications. We add to the literature by showing that glutamatergic activity is required for the maturation of striatal neurons, and demonstrate that glutamate receptor activity can induce structural plasticity in the absence of dopamine or other neuromodulators. The experiments here provide a foundation for future studies of activity-dependent striatal plasticity both in development and disease.

The structural responses of SPNs to altered glutamate input were examined within the context of cortico-striatal co-cultures, and findings are summarized in the graphical abstract (Figure 5). A reduction in spine density following chronic or medium-term TTX application suggests that excitatory action potential firing in cortico-striatal networks is a crucial regulator of dendritic spines, and thus of excitatory synapse development and/or maintenance on striatal neurons; however, a contributing role of GABA from interneurons or SPNs themselves cannot be ruled out in the present work. NMDAR-dependent structural LTP-like changes were rapidly induced by glycine stimulation, as evidenced by a ~30% increase in spine density and GluA1 cluster signals within 30 min. Overall, these pharmacological silencing and plasticity experiments indicate that altering glutamatergic activity is sufficient to drive structural plasticity in SPNs, even in the absence of dopamine and other striatal neuromodulators. Furthermore, while there is an ongoing debate about the role of filopodia as intermediates in spine formation (reviewed in Sala and Segal, 2014), our finding that spine and filopodia densities were not always negatively correlated supports the notion that they are, at least in part, regulated by distinct processes.

We replicate a previous finding that the continuous presence of TTX causes a reversible reduction in spines and increase in filopodia in developing co-cultured SPNs (Segal et al., 2003), and additionally demonstrate that 24 h silencing is sufficient to drive an increase in filopodia along with a strong trend to decreased spine density. Based on previous reports in cortico-striatal co-cultures, SPN spine density increases from ~0.07 to 0.3–0.4 spines/μm between DIV7 and 14 (Burguière et al., 2013; Penrod et al., 2015; Thibault et al., 2016) to ~1 spine/μm by DIV21 (Tian et al., 2010; Penrod et al., 2015). Thus, our reported spine density of ~0.36 and ~0.46 spines/μm at DIV21 following 2× TTX and 3× TTX, respectively, suggest a suppression in the maturing spines between DIV7–21; however, it is also possible that newly formed spines are lost or revert to filopodia upon additional TTX applications. In contrast, given the relative stability of spine densities by three weeks in culture (Penrod et al., 2015), our findings following a 24 h TTX application at DIV20 most likely reflect either a conversion of mature spines to filopodia or distinct regulation of each. Further examination of the mechanisms underlying these changes would benefit from examining SPNs at different developmental stages and at multiple time points following glutamatergic silencing.

That said, the silencing experiments here allow direct comparison of SPN silencing with studies in hippocampal and cortical pyramidal neurons. While a 24 h silencing period in DIV19 hippocampal neurons did not affect spine density, an increase in spine length was interpreted as a possible conversion to filopodia (Papa and Segal, 1996), in support of the results here. Chronic glutamate blockade has yielded somewhat contradictory results elsewhere; one study reported that TTX application throughout synaptogenesis reduced spine density ~50% in cultured hippocampal neurons (Kossel et al., 1997), whereas normal synaptogenesis and spine formation was observed in TTX-treated rat hippocampal slice cultures (McKinney et al., 1999; Soares et al., 2013) and in those from transgenic mice entirely lacking presynaptic glutamate release (Sigler et al., 2017). SPNs also respond differently to disrupted glutamatergic transmission in vivo; while hippocampal neurons develop mature spines in the absence of glutamatergic transmission (Sando et al., 2017), reducing glutamatergic release at cortico-striatal synapses in postnatal day (P) eight mice led to a ~40% reduction in spine density measured at P14–15 (Kozorovitskiy et al., 2012). Such changes may be distinct to spiny GABAergic neurons, as GABAergic cerebellar Purkinje cells also lack spines when cultured in the presence of TTX (Schilling et al., 1991). Future comparison of chronic silencing and glutamate receptor blockade in both SPNs and cortical pyramidal neurons in the same co-culture may prove enlightening. This would enable cell-specific spine/filopodia responses to be assessed under the same treatment paradigms and at the same developmental stages.

Intriguingly, we found increased presynaptic Synapsin-1 cluster size and intensity (but not density) following 24 h glutamatergic silencing, in contrast to the decrease observed following the chronic blockade. One possibility for this difference is a transient response to 24 h silencing, related to the immediate pause in the activity-dependent vesicle cycle. Studies of homeostatic plasticity in hippocampal and cortical neurons indicate that ~4–24 h suppression of action potential firing (or AMPAR activity) increases postsynaptic responses (O’Brien et al., 1998; Turrigiano et al., 1998; Stellwagen and Malenka, 2006; Goel and Lee, 2008; Ibata et al., 2008); however, it remains unclear whether this happens in other neuron types (Rutherford et al., 1997; Kim and Tsien, 2009), as does the extent to which homeostatic presynaptic changes in glutamate release and protein expression occur (Erickson et al., 2006; Wierenga et al., 2006; Turrigiano, 2011; Zhao et al., 2011). Although our results indicate that some presynaptic change occurred at cortico-striatal synapses following silencing, which may precede structural changes (as a trend to increased Synapsin-1 cluster size was visible 3 h post-treatment), the interpretation is limited by our choice of presynaptic marker. Given that Synapsin-1 is present at both glutamatergic and GABAergic synapses, we cannot distinguish whether our silencing paradigms differentially altered excitatory or inhibitory input onto SPNs; moreover, opposite regulation of these inputs could cancel one another, or mask additional effects in our readouts. To verify whether the presynaptic input changes observed here were occurring at excitatory synapses, we conducted additional analysis in which we quantified Synapsin-1 puncta on dendritic spines and filopodia only. While this strongly suggests that the observed Synapsin-1 changes were indicative of plasticity at excitatory synapses, future experiments would benefit from staining for synapse-specific pre- and post-synaptic markers to distinguish between inhibitory and excitatory synapses (for example VGLUT1 and PSD95 vs. VGAT and gephyrin, respectively; Rao and Craig, 1997; Levinson and El-Husseini, 2005). This, in combination with electrophysiological recordings of both excitatory and inhibitory post-synaptic currents, would provide more insight as to how different forms of glutamatergic blockade (chronic vs. acute; burst firing vs. receptor blockade) affect both structural and functional plasticity at synapses on SPNs. Future experiments could additionally examine activity-dependant effects on AMPA and NMDA receptor subunit composition and subcellular distribution; given that alterations to these are hallmarks of activity blockade in other glutamatergic neurons (Rao and Craig, 1997; Ehlers, 2003; Soares et al., 2013), it is worth investigating whether similar changes occur in SPNs. Nonetheless, examining the overall change in presynaptic input onto SPNs, in parallel to quantifying dendritic protrusions, highlighted an interesting difference between the response to chronic and 24 h glutamatergic blockade.

Our observations following the additional blockade of AMPARs and NMDARs during 24 h silencing suggest a potential disconnect between the spine and filopodia dynamics. While blocking ionotropic glutamate receptors prevented any suggestion of a change in spine density, a very strong trend to increased filopodia density remained. This is in agreement with results in cortical pyramidal and hippocampal neurons, where NMDAR activity is required for activity-dependent spinogenesis (Fischer et al., 2000; Kwon and Sabatini, 2011) and spine shrinkage/loss (Nägerl et al., 2004; Zhou et al., 2004; Oh et al., 2013), and where AMPAR activity regulates spine motility (Fischer et al., 2000) and maintenance (McKinney et al., 1999). Distinct effects on spine and filopodia dynamics have also been observed elsewhere; blocking AMPA receptors reduced spine density in hippocampal slices 7 days post-treatment, whereas NMDAR blockade had no effect on spines, but instead caused the appearance of filopodia-like protrusions (McKinney et al., 1999).

Beyond methodological variability, different responses to glutamate receptor blockade may arise from the existence of different filopodia sub-types (Portera-Cailliau et al., 2003; Richards et al., 2005), as well as the effects of developmental stage on filopodia dynamics (Sala and Segal, 2014); these distinctions could be tested in future experiments. It is also worth considering that some differences between blocking burst firing vs. all glutamatergic activity may arise from extrasynaptic receptors; while we assume antagonists are most effective in blocking spontaneous glutamate release and activation of receptors close to release sites, we cannot rule out the possibility that silencing extrasynaptic receptor activation through ambient glutamate in the media may contribute. Regardless, our results extend the literature on the effects of glutamatergic receptor blockade to striatal SPNs, demonstrating that, as at other glutamatergic synapses, spine pruning is an active process requiring ongoing low-level glutamate activity, and can be uncoupled from filopodia formation.

The role of NMDARs in striatal activity-dependent plasticity has been revealed primarily by slice electrophysiology (reviewed in Perrin and Venance, 2019). Although LTD was initially considered the dominant form of plasticity at cortico-striatal synapses, many reports have since shown that high-frequency stimulation can result in either NMDAR-dependent LTP, or mGluR-dependent LTD (Calabresi et al., 1996; Spencer and Murphy, 2000; Tang et al., 2001; Reynolds and Wickens, 2002; Wang et al., 2006; Sergeeva et al., 2007; Li et al., 2009; Lovinger and Mathur, 2012; Johnson et al., 2017). However, debate remains as to whether dopamine or other neuromodulators are necessary for the expression of LTP (Spencer and Murphy, 2000; Calabresi et al., 2007; Li et al., 2009; Lovinger, 2010; Burguière et al., 2013; Park et al., 2014; Cerovic et al., 2015). Only a few studies have specifically examined activity-dependent spine alterations in the context of AMPAR trafficking in SPNs (Kozorovitskiy et al., 2012, 2015; Matikainen-Ankney et al., 2018), and, to our knowledge, only one other study has done so in the absence of dopamine (Burguière et al., 2013). Here, we show that LTP-like changes in SPNs can be driven by NMDAR activity alone, as the NMDAR co-agonist glycine (in the absence of Mg²⁺) produced a rapid increase in dendritic spines and associated GluA1 expression. Thus, even in the absence of dopamine, SPNs show a similar response to NMDAR stimulation compared to principle excitatory neurons, in which LTP induction by glutamate uncaging (Matsuzaki et al., 2004; Yang et al., 2008) or chemical paradigms (Lin et al., 2004; Huang et al., 2005; Park et al., 2006; Sharma et al., 2006; Korkotian and Segal, 2007; Fortin et al., 2012; Brigidi et al., 2014) leads to increased spine head volume or de novo spine formation without filopodial intermediates (Kwon and Sabatini, 2011). However, a contributing role of non-neurotransmitter neuromodulators within the culture media, such as BDNF, cannot be discounted in the present findings.

The observed increase to GluA1 cluster intensity and association within spines also suggests a functional change, consistent with results in hippocampal neurons showing trafficking and membrane insertion of GluA1-containing AMPARs in spines following LTP induction (Shi et al., 2001; Malinow and Malenka, 2002; Matsuo et al., 2008; Fortin et al., 2010). However, electrophysiological measures of activity-dependent changes here were less clear. Whole-cell patch-clamp recordings ~30–60 min following treatment revealed variable effects on mEPSC frequency, although a trend to increased frequency in glycine-treated SPNs was observed. Moreover, we found no indication of increased mEPSC amplitude following glycine, despite the increased GluA1 signal. A potential reason for this is that the synaptic effects of glycine stimulation are not fully captured by measuring quantal (miniature) glutamatergic transmission, and that changes in evoked activity would be more apparent, given the growing body of literature suggesting that these are mechanistically distinct (Ramirez and Kavalali, 2011; Kavalali, 2015; Abrahamsson et al., 2017; Andreae and Burrone, 2018; Chanaday and Kavalali, 2018). Alternatively, the observed increase in spine density could precede functional changes requiring associated new presynaptic elements (as Synapsin-1 density did not increase); apropos, spine enlargement before AMPAR insertion has been observed following chemical LTP induction in hippocampal slices (Kopec et al., 2006).

In a separate study in which we used the same chemical LTP protocol on cultured hippocampal neurons, we found increased mEPSC amplitude and frequency, which correlated with increased spine width and density 30–60 min after glycine stimulation (Brigidi et al., 2014). It is thus possible that SPNs, unlike hippocampal neurons, require neuromodulators to fully express synaptic LTP in terms of current flux, while structural plasticity can be induced by NMDAR activation alone. In support of this, glutamate uncaging alone leads to spinogenesis in SPNs ~50% of the time, whereas D1 or A2a receptor agonists significantly increase the probability of novel spines and functional synapses (as evidenced by increased mEPSC frequency) in D1R and D2R SPNs, respectively (Kozorovitskiy et al., 2015). Nevertheless, the significant decrease in mEPSC event-decay constant following glycine treatment indicates that some functional change occurred at cortico-striatal synapses, possibly reflecting altered glutamate receptor subunit composition or phosphorylation (Lambolez et al., 1996; Banke et al., 2000; Chater and Goda, 2014). GluA2-lacking, calcium-permeable AMPARs exhibit faster decay kinetics than those containing GluA2 (reviewed in Diering and Huganir, 2018), and multiple studies have reported their integration at specific synapses, including cortico-striatal ones, during LTP induction (Lamsa et al., 2007; Soares et al., 2013; Ma et al., 2018; Park et al., 2018; Benke and Traynelis, 2019). Thus, the faster decay of mEPSCs in glycine-treated co-cultures, together with the increased GluA1 signal in spines, may reflect the activity-dependent insertion of calcium-permeable AMPARs in SPNs. Future work could extend these findings by recording evoked vs. miniature EPSCs, verifying differences in SPN subtype, and/or additionally examining the response to D1 and A2a receptor agonists. Regardless, our results show that NMDAR activation drives rapid structural, and some electrophysiological changes at cortico-striatal synapses.

This study presents an examination of activity-dependent structural development and plasticity within GABAergic striatal projection neurons. Chronic and short-term glutamatergic manipulations to co-cultured SPNs provides a comparison with similar studies in hippocampal and cortical pyramidal neurons and highlights the distinct but overlapping regulation of spine and filopodial activity-dependent plasticity. In particular, we show that SPN structural plasticity occurs within 24 h of glutamate activity blockade, and within 30 min of a 3-min NMDAR activation by glycine, even in the absence of dopamine; thus, the cortico-striatal co-culture system is useful for examining the specific role of glutamate receptor activity in shaping SPN physiology and cortico-striatal synapses. While our primary aim was to provide a characterization of structural plasticity in SPNs and how these may differ from principal excitatory neurons, we offer several ideas on how these assays can be refined and built upon. These could easily be applied to examining activity-dependent plasticity in disease models, particularly those in which altered glutamatergic transmission and aberrant structural plasticity may play a pathophysiological role, and in which SPNs have shown distinct vulnerability.

The raw data supporting the conclusions of this article will be made available by the authors upon request, without undue reservation.

The animal study was reviewed and approved by Canadian Council on Animal Care.

NK and AM designed the study and co-wrote the manuscript. With supervision from AM, NK conducted all experiments, analyzed data, interpreted results, and made figures. MW and LQ-R helped with image processing and analysis. MF provided scientific input and supervisory support throughout. All 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.