Abstract
Protein palmitoylation is the only reversible post-translational lipid modification. Palmitoylation is held in delicate balance by depalmitoylation to precisely regulate protein turnover. While over 20 palmitoylation enzymes are known, depalmitoylation is conducted by fewer enzymes. Of particular interest is the lack of the depalmitoylating enzyme palmitoyl-protein thioesterase 1 (PPT1) that causes the devastating pediatric neurodegenerative condition infantile neuronal ceroid lipofuscinosis (CLN1). While most of the research on Ppt1 function has centered on its role in the lysosome, recent findings demonstrated that many Ppt1 substrates are synaptic proteins, including the AMPA receptor (AMPAR) subunit GluA1. Still, the impact of Ppt1-mediated depalmitoylation on synaptic transmission and plasticity remains elusive. Thus, the goal of the present study was to use the Ppt1−/− mouse model (both sexes) to determine whether Ppt1 regulates AMPAR-mediated synaptic transmission and plasticity, which are crucial for the maintenance of homeostatic adaptations in cortical circuits. Here, we found that basal excitatory transmission in the Ppt1−/− visual cortex is developmentally regulated and that chemogenetic silencing of the Ppt1−/− visual cortex excessively enhanced the synaptic expression of GluA1. Furthermore, triggering homeostatic plasticity in Ppt1−/− primary neurons caused an exaggerated incorporation of GluA1-containing, calcium-permeable AMPARs, which correlated with increased GluA1 palmitoylation. Finally, Ca2+ imaging in awake Ppt1−/− mice showed visual cortical neurons favor a state of synchronous firing. Collectively, our results elucidate a crucial role for Ppt1 in AMPAR trafficking and show that impeded proteostasis of palmitoylated synaptic proteins drives maladaptive homeostatic plasticity and abnormal recruitment of cortical activity in CLN1.
SIGNIFICANCE STATEMENT Neuronal communication is orchestrated by the movement of receptors to and from the synaptic membrane. Protein palmitoylation is the only reversible post-translational lipid modification, a process that must be balanced precisely by depalmitoylation. The significance of depalmitoylation is evidenced by the discovery that mutation of the depalmitoylating enzyme palmitoyl-protein thioesterase 1 (Ppt1) causes severe pediatric neurodegeneration. In this study, we found that the equilibrium provided by Ppt1-mediated depalmitoylation is critical for AMPA receptor (AMPAR)-mediated plasticity and associated homeostatic adaptations of synaptic transmission in cortical circuits. This finding complements the recent explosion of palmitoylation research by emphasizing the necessity of balanced depalmitoylation.
Introduction
Protein palmitoylation is a post-translational lipid modification that is unique for the fact that it is readily reversible. Palmitoylation diversely affects protein function but generally enhances their association with cellular membranes, particularly the synaptic membrane (Washbourne, 2004; Y. Fukata and M. Fukata, 2010; Levental et al., 2010). Thus, a critical balance between palmitoylation and depalmitoylation of any substrate protein must be maintained for its proper trafficking, turnover, and degradation (Young et al., 2012; Holland and Thomas, 2017; Jin et al., 2021). However, while the addition of palmitate to proteins is conducted by over 20 enzymes called protein acyltransferases, some of which act at the synaptic membrane (M. Fukata et al., 2004; Noritake et al., 2009; Woolfrey et al., 2015; Tabaczar et al., 2017), depalmitoylation is conducted by fewer enzymes (Won et al., 2018). Emphasizing the importance of depalmitoylation in cellular health are mutations in the depalmitoylating enzyme palmitoyl-protein thioesterase 1 (PPT1) that cause a devastating pediatric neurodegenerative condition, infantile neuronal ceroid lipofuscinosis (CLN1), which also belongs to the category of lysosomal storage diseases (Vesa et al., 1995; Nita et al., 2016).
Although PPT1 was originally shown to localize chiefly to the lysosome to aid in the degradation of palmitoyl-modified proteins (Verkruyse and Hofmann, 1996), several lines of evidence also show that PPT1 is localized to the broader endolysosomal compartment in neurons, including synaptic vesicles (Lehtovirta et al., 2001; Ahtiainen et al., 2003; Lyly et al., 2007). In fact, functional studies have identified presynaptic and postsynaptic deficits in neurons lacking Ppt1, indicating that Ppt1-mediated shuttling of molecules is needed for synaptic transmission (Virmani et al., 2005; Kim et al., 2008; Sapir et al., 2019). Of note, a recent stringent proteomic screen for Ppt1 substrates showed that roughly 10% of palmitoylated synaptic proteins (>100 proteins) are depalmitoylated by Ppt1, including the AMPA receptor (AMPAR) subunit GluA1 (Gorenberg et al., 2022). Together, these observations suggest that a proper balance between palmitoylation and depalmitoylation of AMPAR subunits is needed to support normal synaptic transmission and its homeostatic adaptations. However, the impact of Ppt1-mediated depalmitoylation on synaptic transmission and plasticity remains obscure, despite the established fact that palmitoylation of AMPARs regulates their synaptic expression (Hayashi et al., 2005; Lin et al., 2009; Dolah et al., 2011; Spinelli et al., 2017; Itoh et al., 2018).
Considering the role for Ppt1 in regulating the NMDA receptor (NMDAR) function during neurodevelopment (Koster et al., 2019), the present study was designed to test whether Ppt1 also regulates AMPAR plasticity, which is crucial for the formation and maintenance of cortical circuits (Sheng et al., 1994; Erisir and Harris, 2003; Hall et al., 2007; Gambrill and Barria, 2011). To this end, we used the Ppt1−/− mouse model (Gupta et al., 2001) to assess the extent to which the lack of depalmitoylation by Ppt1 dysregulates AMPAR-mediated homeostatic plasticity and its impact on cortical circuit function (G.G. Turrigiano, 1999; G.G. Turrigiano and Nelson, 2004). Ultimately, the goal of the present study is to gain insights into the principles by which palmitoylation and depalmitoylation work in concert to enable proper regulation of synaptic plasticity and homeostatic adaptations in neural circuits.
Materials and Methods
Study approval
All animal procedures were approved with the guidelines of the University of Illinois Chicago Institutional Animal Care and Use Committee and in accordance with the National Institutes of Health Guidelines for Care and Use of Laboratory Animals.
Animals, group allocation, and data handling
Ppt1+/− (heterozygous) mice were obtained from Jackson Laboratory and maintained on 12/12 h light/dark cycle with food and water ad libitum. Breeding of Ppt1+/− animals results in litters containing Ppt1−/−, Ppt1+/−, and Ppt1+/+ [wild type (WT)] animals. Ppt1−/− and WT littermate controls were genotyped in-house (Gupta et al., 2001) and mice of both sexes (in roughly equal proportion) were used for experiments at specified time points (see Results). Although we used the littermate control system, in which WT and Ppt1−/− mice from the same litters were compared, each n was treated independently in statistical testing for in vivo comparisons (pair-wise tests were not used). In contrast, in vitro and imaging data were treated as paired analyses, since each individual culture was often collected, processed, or immunostained in parallel (rather than multiple n being processed simultaneously). Imaging data were acquired randomly for each experiment (no criteria for selecting cells, view fields, etc. except where anatomically necessary). All data were acquired and maintained without descriptive naming/labeling to ease randomization. Data were either analyzed by lab members blinded to condition or randomized before analysis by K.P.K., with the exception that two-photon data were not randomized before analysis; however, these calcium data are extracted in a fully automated manner by the EZ calcium (Cantu et al., 2020) MATLAB plugin (see below).
Electrophysiology
WT and Ppt1−/− animals at postnatal day (P)15 or P28–P32 were deeply anesthetized via intraperitoneal injection of chloral hydrate and decapitated. All recordings were conducted from layer 2/3 pyramidal neurons of monocular visual cortex (350-μm-thick coronal slices) at 33–35°C using a low-chloride, cesium-gluconate based internal solution and an external solution free of glutamate and GABA blockers to enable concurrent acquisition of excitatory and inhibitory synaptic currents at the single-cell level (Flores-Barrera et al., 2017, 2020). Contents of the internal solution are the following (in mm): 10 CsCl, 130 Cs-gluconate, 10 HEPES, 2 MgCl2, 5 NaATP, 0.6 NaGTP, and 3 QX-314, pH 7.23–7.28, 280–282 mOsm. The recording artificial cerebral spinal fluid contained the following (in mm): 122.5 NaCl, 3.5 KCl, 25 NaHCO3, 1 NaH2PO4, 2.5 CaCl2, 1 MgCl2, 20 glucose, and 1 ascorbic acid, pH 7.40–7.43, 295–305 mOsm. As a result, both spontaneous glutamatergic and GABAergic events could be assessed readily by recording the frequency of postsynaptic currents (PSCs) at the −60 mV (PSC−60 mV) and +15 mV (PSC+15 mV) holding potentials, respectively. These holding potentials were estimated from the current–voltage relationship showing a reversal for IPSCs close to the theoretical reversal potential of −62 mV, whereas the excitatory component reversed at around +15 mV following bath application of picrotoxin (see Flores-Barrera et al., 2020). Only neurons with at least 10 min of stable baseline activity were included for analyses. The frequency of PSC−60 mV and PSC+15 mV events from at least two noncontiguous epochs of 60 s each was compared.
Brain fractionation, biochemical assays from tissue samples, and immunoblotting
For collection of brain for biochemistry (immunoblot), Ppt1−/− and WT animals were decapitated following isoflurane anesthesia, then the brain was removed, and washed in ice-cold PBS. The occipital cortex (visual cortex), hippocampus, and remaining cortex were dissected and separately collected on an ice block. Isolated visual cortices from Ppt1−/− and WT animals were homogenized in ice-cold synaptosome buffer [320 mm sucrose, 1 mm EDTA, 4 mm HEPES, pH7.4 containing 1× protease inhibitor cocktail (Roche), 1× phosphatase inhibitor cocktail (Roche) and 1 mm PMSF] using 30 strokes in a Dounce homogenizer. Aliquots of lysates were stored at −80°C and the remaining sample was used for synaptosome preparation, performed as previously with slight modification. In brief, whole lysates were centrifuged at 1000 × g to remove cellular debris, supernatant was then centrifuged at 12,000 × g for 15 min to generate pellet P2. P2 fraction was resuspended in synaptosome buffer and spun at 18,000 × g for 15 min to produce synaptosomal membrane fraction, LP1, which was used for downstream biochemical analyses (synaptosomes). For immunoblot, protein concentration of each sample was determined using BCA protein assay (Pierce). Samples were then measured to 20-µg total protein in 2× Laemmli buffer containing 10% β-mercaptoethanol (Bio-Rad), heated at 70°C for 10 min and loaded into 4–20% precast gels (Bio-Rad) for electrophoresis (130 V, 1.5–2 h). Proteins were wet-transferred to PVDF membranes (Immobilon-P, Millipore), blocked in TBS, pH7.4 containing 5% nonfat milk and 0.1% Tween 20 (TBS-T + 5% milk). Membranes were incubated in primary antibody solutions containing 2% BSA in TBS-T for 2 h at room temperature (RT) or overnight at 4°C. Primary antibodies were used according to the key resources table (Table 1). Membranes were then incubated with appropriate secondary, HRP-conjugated antibodies (Jackson ImmunoResearch) at either 1:1000 or 1:5000 for 1 h at RT before washing three times with TBS-T. Visualization and quantification was performed using Pierce SuperSignal ECL substrate and Odyssey-FC chemiluminescent imaging station (LI-COR). Signal density for each synaptic protein was measured using the LI-COR software, Image Studio Lite (version 5.2) and was normalized to the signal density for β-actin loading control for each lane. The number of visual cortices used for analysis at each age are displayed in Figure 1 (individual points on graph), with two technical replicates for each experiment averaged together.
Key resources table describing all relevant reagents used
APEGS assay from visual cortices
The APEGS assay was performed as described by Kanadome and colleagues (Kanadome et al., 2019), following the guidelines for tissue samples. Visual cortices first underwent the synaptosome preparation protocol as above, except that homogenate buffer used was as directed by the APEGS protocol (20 mm Tris-HCl, 2 mm EDTA, and 0.32 m sucrose, pH 8.0). Lysates and synaptosomes were then brought to 300 µg total protein in a final volume of 0.5-ml buffer B (PBS containing 4% SDS, 5 mm EDTA, and 8.9 m urea, and protease inhibitors). The remaining sample was used for inputs. 300 µg protein was reduced by addition of 25 mm Bond-Breaker TCEP (0.5 m stock solution, ThermoFisher) and incubation at RT for 1 h. Next, to block free thiols, freshly prepared N-ethylmaleimide (NEM) in 100% ethanol was added to lysates (to 50 mm) and the mixture was rotated end-over-end for 3 h at RT. Following 2× chloroform-methanol precipitation (at which point, protein precipitates were often stored overnight at −20°C), lysates were divided into +hydroxylamine (HA) and –HA groups for each sample, which were exposed to three volumes of HA-containing buffer (1 m HA, to expose palmitoylated cysteine residues) or Tris-buffer control (–HA), respectively, for 1 h at 37°C. Following chloroform-methanol precipitation, the samples were solubilized and exposed to 10 mm TCEP and 20 mm mPEG-5k (Laysan Bio Inc; see Table 1) for 1 h at RT with shaking (thereby replacing palmitic acid with mPEG-5K on exposed cysteine residues). Following the final chloroform-methanol precipitation, samples were solubilized in a small volume (70 µl) of PBS containing 1% SDS and protein concentration was measured by BCA assay (Pierce). Samples were then brought to 20-µg protein in Laemmli buffer with 2% β-mercaptoethanol for immunoblot analyses as above. Quantification of palmitoylated versus nonpalmitoylated protein was conducted as for standard immunoblot analysis, with the additional consideration that signal from palmitoylated bands demonstrating the APEGS-dependent molecular weight shift was divided by the signal from the nonpalmitoylated band, the location of which was verified by matching to the –HA control sample. This ratio was divided by β-actin control from the same lane for normalization.
Stereotaxic viral injection
For designer receptor exclusively activated by designer drugs (DREADD)-induced scaling experiments, animals between P16 and P18 were anesthetized via isoflurane inhalation (4% induction, 1–1.5% maintenance) and placed in a stereotaxic frame (RWD Life Science Inc.). A small (0.5–1 cm) incision was made to expose the posterior portion of the skull. Connective tissue beneath the scalp was removed using sterile cotton swabs and the coordinates of bregma were recorded. The injection needle was then moved to the appropriate injection location (2.2 mm posterior to bregma or 0.4 mm anterior to λ and 2.2 mm lateral) for the left visual cortex and the location was marked on the skull using a fine tip permanent pen. A small burr hole was drilled in this location with a dental drill (Marathon) and the needle was lowered so that it rested just above the cortical surface. The needle was then lowered into the cortex (400 µm in depth) and 50 nl of virus was injected. The needle was then slowly retracted at 100-µm increments, and 50 nl of virus was injected at each increment (200 nl total through dorsoventral aspect of the visual cortex). The same procedure was performed for animals in which the control (hDlx-GFP) virus was injected. After removal of the needle (10 min total), the scalp was sutured and animals were monitored until recovery, at which point they were returned to their home cages. Following a recovery period of 10 d to allows for expression of the DREADD in visual cortical neurons, all animals (including GFP controls) were injected with the DREADD agonist, CNO (3 mg/kg) daily for 5 consecutive days and killed for biochemical analysis 4–6 h after the final injection.
Transcardial perfusion and histology
Ppt1−/− and WT mice were anesthetized using isoflurane and transcardially perfused with ice cold PBS (pH 7.4, ∼30 ml/mouse) followed by 4% paraformaldehyde (PFA) in PBS (∼15 ml/mouse). Brains were removed and postfixed overnight at 4°C in 4% PFA and transferred to PBS, pH7.4 containing 0.01% sodium azide for storage if necessary. Brains from Ppt1−/− and WT animals were incubated in 30% sucrose solution for 48 h before sectioning at either 50 or 100 µm in cold PBS using a Vibratome 1000 (Technical Products International). Serial sections were stored free floating in cryoprotectant solution (30% glycerol, 30% ethylene glycol in PBS) at −20°C until analysis mounting and imaging was performed.
Primary cortical neuron culture
For primary cortical neuron cultures, embryos from timed-pregnant, Ppt1-/+ dams were removed, decapitated under anesthesia, and cortices resected at embryonic day (E)15.5. All dissection steps were performed in ice cold HBSS, pH7.4. Following cortical resection, tissue from each embryo was individually collected to a separate microtube, genotyped, and digested in HBSS containing 20U/ml papain and DNase at 37°C (20 min total, tubes flicked at 10 min) before sequential trituration with 1 ml (∼15 strokes) and 200 µl (∼10 strokes) pipettes, generating a single-cell suspension. For live-cell imaging experiments, cells were counted then plated at 150,000–180,000 cells/well in 24-well plates containing poly-D-lysine/laminin-coated coverslips. For biochemical experiments, i.e., immunoblot, APEGS assay in vitro, cells were plated on poly-D-lysine/laminin-coated six-well plates at 1,000,000 cells/well. Cells were plated and incubated at 37°C in plating medium (Neurobasal medium containing B27 supplement, L-glutamine and glutamate) for 3–5 days in vitro (DIV), before replacing half medium every 3 d with feeding medium (plating medium without glutamate). For synaptic scaling experiments, neurons were treated with either bicuclline (20 μm, solubilized in DMSO, Tocris) or TTX-citrate (1 μm, solubilized in sterile water, Tocris) for 48 h where indicated.
Primary cortical neuron harvesting, biotinylation assay, and immunoblotting
Primary cortical neurons from E15.5 WT and Ppt1−/− embryos were cultured for 16–18 DIV before harvest for immunoblot or APEGS assay. To harvest protein extracts for APEGS, cells were placed on ice, washed 2× with ice-cold PBS before addition of 400 µl per well of PBS with 4% SDS and protease inhibitor cocktail (as in Yokoi et al., 2016). Cells were incubated and swirled with lysis buffer for 5 min, scraped from the plate, triturated briefly, and collected in 1.5-ml tubes. Lysates were centrifuged at 20,000 × g for 15 min to remove debris, and the supernatant was collected for biochemical analysis. Immunoblotting analyses were performed as above.
Surface biotinlyation assay was performed as in (Ehlers, 2000) and (Yoshii et al., 2013) with some modifications. Briefly, neuron cultures (six-well plates) were cooled on ice to stop membrane trafficking in cold DPBS plus 1 mm MgCl2 and 2.5 mm CaCl2, then treated with 1.5 mg/ml sulfo-NHS-biotin in DPBS for 30 min to label surface receptors. Unbound biotin was quenched with cold DPBS plus 1 mm MgCl2, 2.5 mm CaCl2, and 50 mm glycine (rinse, 2 × 3 min). Neurons were then lysed for 5 min in SDS-RIPA buffer (Cell Signaling) with protease inhibitor cocktail (Roche), scraped and the lysed solution was centrifuged at 16,000 × g for 15 min at 4°C. Protein concentration of the supernatant was determined and 400 μg of protein was adjusted to a volume of 1 ml with 0.1% SDS-RIPA buffer. Biotinylated GluA1 receptor subunits were precipitated with 75-μl streptavidin coated magnetic bead (Thermo Scientific) slurry either for 3 h at RT or overnight at 4°C with end over end rotation. Beads were then washed with the SDS-RIPA buffer twice before brief centrifugation. Pulled down subunits were eluted into Laemmli sample buffer with 2% β-mercaptoethanol by boiling for 6 min and the magnetic beads were removed from solution. Samples were then loaded onto 4–20% Bio-Rad gradient gels as above for immunoblot analysis.
Immunocytochemistry and surface GluA analyses
All experimental and control groups were immunostained simultaneously to match exact staining conditions. Neurons were first immunostained for surface receptors (N-terminal GluA1/GluA2 antibodies were used) before permeabilization and immunolabeling of intracellular antigens (MAP2) according to the following procedure. Coverslips were washed 3× with TBS and blocked for 1 h at RT in TBS containing 5% BSA. Then, primary antibody (GluA1; see Table 1) at 1:400 dilution was added to coverslips in TBS containing 1% BSA and incubated for 2 h at RT or overnight at 4°C. Following 4× washes with TBS cells were incubated with 1:400 secondary, fluorophore-linked antibody (Alexa Fluor 488, catalog #A-11034) in TBS containing 1% BSA. These steps were repeated to immunolabel MAP2 (1:400; see Table 1 for primary and secondary antibody information), with the addition of 0.1% Triton X-100 in all solutions. Coverslips are then mounted on SuperFrost Plus slides in DAPI Vectamount medium.
For analysis of GluA surface immunofluorescence, confocal z-stacks (12 bit, 193 × 193 µm at 1028 × 1028-pixel density, 4-µm z-plane interval) were acquired at 63× magnification with 0.6 digital zoom, such that one cell with its primary and secondary dendrites were within the field of view; although, some images contained up to four cells that had unobstructed dendrites suitable for analysis. Cells were imaged in blinded fashion and chosen at random, although the only qualification for image acquisition was identification of view fields where dendrites were not strongly intermingled to avoid misinterpretation of GluA puncta as belonging to an underlying dendrite. For analysis of puncta count, size, and area, maximum-intensity projections of the GluA fluorescence channel (488 nm) were generated and a standard region of interest (ROI; 20 µm long, 5 µm wide) was drawn over secondary dendrites that were between 50 and 100 µm from the edge of the cell soma. This approach normalizes for differences in GluA puncta properties because of their relative distribution across the dendritic arbor. The image was thresholded using the automated Fiji algorithm, Yen, and analyze particles tool (Fiji) was used. Between two and six ROIs per cell were analyzed for each cell (depending on the number of secondary dendrites that fell within the 50- to 100-µm range and were unobstructed by overlapping dendrites) and were averaged, constituting the count, and percent area values for that cell.
Transfection, FRAP, and in vitro calcium imaging
For fluorescence recovery after photobleaching (FRAP) analyses, WT and Ppt1−/− cortical neurons were transfected with superecliptic (SEP)-GluA1 (Kopec et al., 2006) at DIV10 using Lipofectamine 2000 (ThermoFisher) according to manufacturer protocol with a slight modification. SEP-GluA1 DNA construct (added at ∼1 µg/well) was mixed with Lipofectamine-containing Neurobasal medium, incubated for 30 min to complex DNA-Lipofectamine, equilibrated to 37°C, and added to the cells 250 µl/well for 1 h. Following incubation, complete medium was returned to the cells. Neurons treated with TTX for 48 h at DIV16 were imaged in Tyrode's solution (in mm: 135 NaCl, 5 KCl, 2 CaCl2, 1 MgCl2, 25 HEPES, and 10 glucose, pH 7.4) using a Zeiss LSM880 in Airyscan mode. Five to six spines were randomly selected for each neuron to bleach during the imaging program. Images were acquired every 15 s for five frames to establish baseline fluorescence before bleaching at maximum laser intensity. Images were then acquired every 15 s thereafter for 10 min to analyze fluorescence recovery at bleached spines. The Zen Definite Focus function was used to correct potential drifting of the imaging plane. To analyze FRAP of SEP-GluA1, manual ROIs were drawn over the bleached spines and one to two control (unbleached) spines for each cell and the ΔF/F0 was extracted using the multi measure tool. Baseline fluorescence (average of first five frames) was set to 100%, and the following frames were compared with that baseline, generating the FRAP curves in Figure 5. FRAP curves were fit using a nonlinear regression in GraphPad Prism 9.0 using the LOWESS function. The immobile fraction was calculated by subtracting the average of the fluorescence recovery value for the final five frames for each spine from 100%. All spines for each cell were averaged to create one n. cells were taken from two independent cultures.
To image calcium signals in WT and Ppt1−/− neurons, cells were transfected at DIV10 with GCaMP3 (see the acknowledgements footnote) using Lipofectamine 2000 (ThermoFisher) as described above. Cells were grown to DIV16–DIV18 and a subset of cells were treated with TTX (1 mm) for 48 h. Neurons were imaged with constant perfusion (∼2.5 ml/min) of Tyrode's solution without magnesium (imaging medium: 139 mm NaCl, 3 mm KCl, 17 mm NaHCO3, 12 mm glucose, and 3 mm CaCl2) using a LSM710 confocal microscope equipped with heated stage. Sixteen-bit videos were acquired at a resolution of 512 × 512 at approximately five frames per second at 10× magnification (with 3× digital zoom). Baseline acquisition, NASPM perfusion (1 μm), and washout periods were 4, 4, and 8 min, respectively; this corresponds to 1000 video frames, 1000 video frames, and 2000 video frames, respectively. For analysis, since NASPM persists in the bath for some time after infusion, the final 1000 frames (3000–4000) were always used for the washout period (i.e., frames 2000–3000 were excluded from analysis as a transition period between NASPM and washout, since washing is not complete during that time). All active synapses were encompassed with a minimal-sized circular (spines) or rectangular (shaft) ROI manually for each video by lab members blinded to condition and genotype. A typical video resulted in 120–200 ROIs (synapses). The “multimeasure” tool was applied to each ROI to generate a ΔF/F0 trace for each synapse. The raw fluorescence data for each synapse was then normalized to its own baseline and split into 1000 frame portions corresponding to baseline, NASPM, and washout conditions. The number of active frames for each period were counted by summing the frames reaching a threshold plus 3 × standard deviation (SD) for each synapse. A synapse was considered to be inhibited by NASPM if it met the following criteria: (1) the synapse had to be active, i.e., show at least six active frames (+3 × SD) during the baseline period; (2) the synapse showed a reduction in active frames in the NASPM period by >80%; and (3) the synapse recovered by showing at least five active frames in the washout period. The total activity plots (counted active frames) are plotted in Figure 6C,D, while the proportion of inhibited synapses (synapses which met the above criteria) is plotted in Figure 6E.
APEGS assay on primary cortical neuron lysates
The acyl-PEGyl exchange gel shift (APEGS) assay was performed as described by Kanadome et al. (2019) and in Koster et al. (2019). The protocol is nearly identical to the description above for in vivo samples, with slight modifications as recommended in the protocol by Kanadome et al. (2019). Specifically, the cortical neuron lysates were brought to 300 µg total protein in a final volume of 0.5-ml buffer A (PBS containing 4% SDS, 5 mm EDTA, protease inhibitors; instead of buffer B). Further, the initial protein reduction by TCEP incubation was performed at 55°C for 1 h (rather than at RT). Following the final chloroform-methanol precipitation, samples were solubilized in 60 µl of PBS containing 1% SDS and protein concentration was measured by BCA assay (Pierce). Samples were then brought to 10 µg protein in Laemmli buffer with 2% β-mercaptoethanol for immunoblot analyses as above. Quantification of palmitoylated versus nonpalmitoylated protein was conducted as described for in vivo samples.
In vivo calcium imaging
For in vivo calcium imaging, injection into neonatal mouse cortices was performed according to He et al. (2018). P0 or P1 pups underwent hypothermia-induced anesthesia (incubation on an ice block for 5–10 min) before being placed in the mouse stereotaxic frame (on a second ice block) equipped with specialized ear bars for holding pups (RWD Life Science Inc). AAV.CamKII.GCaMP6f.WPRE.SV40 (AAV9; Addgene) and pAAV-hSyn-mCherry (AAV2) were premixed in 1:1 ratio before being backloaded into a Neuros syringe (33 gauge, Hamilton). Alternatively, a Nanoject three automatic injector (Drummond Scientific Company) was used with a finely pulled borosilicate capillary micropipette. A total of 200 nl of the mixture was injected at a 15° angle ∼150 µm below the dura surface to transfect a large population of layer 2/3 excitatory cortical neurons with the viral particles. The needle was left in place for 1 min, then lifted by a minimal distance and left in place for an additional 30 s before being slowly removed to avoid reflux of solution along the needle track. Pups were immediately placed on a heating pad and monitored until they regained consciousness.
Cranial window implantation was performed as previously described with some modification (Holtmaat et al., 2009). Virus injected animals were anesthetized via isoflurane inhalation (4% induction, 1–1.5% maintenance) and placed in a stereotaxic frame (RWD Life Science Inc.). All procedures were performed under sterile conditions. Buprenorphine-SR (1 mg/kg, s.c.), meloxicam (2 mg/kg, s.c.), and dexamethasone (2 mg/kg, s.c.) were administered preoperatively. Hair from the scalp was removed using commercially available Nair by applying it with a cotton swab, waiting 3 min, and removing the hair with several clean cotton swabs. The scalp was then cleaned by sequential application of betadine-iodine (3×, allowing to dry in between each fresh application) and alcohol using alcohol pads (2×). Once the scalp was clean, an ∼2-cm incision was made along the posterior aspect of the midline and the margins of the incision were expanded to visualize the skull covering the left visual cortex by gently pushing aside the scalp tissue with sterile, fine tipped cotton swabs. Once the incision was large enough and centered on the left visual cortex, the connective tissue beneath the scalp were removed from the skull by applying of ∼25-µl epinephrine-lidocaine (via insulin syringe) and rubbing with sterile cotton swabs. The margins of the incision were also simultaneously dried with the swab. Once the margins of the tissue were dried, a layer of cyanoacrylate was applied to the edges and cured using minimal application of Zip Kicker cyanoacrylate catalyst (Zap). Next, the position of the headplate (CP-1, Narishige) was marked by positioning it so that the center was atop the visual cortex and marking its location with a fine tip marker. A layer of cyanoacrylate was then applied over the entire surface of exposed skull, leaving sufficient area (∼6 mm) so that glue does not cover the eventual drilling site, and the headplate was quickly secured with the glue to the desired position. After at least 5 min, allowing for the glue to cure, a 4- to 5-mm circular piece of skull was removed by drilling around the circle circumference using a dental drill (Marathon) and frequent soaking with sterile PBS to soften the skull. Once the drill site was thin enough, fine forceps were used to lift the medial edge and the circular bone fragment with a diameter of 4 mm was removed completely. The exposed brain tissue was covered with PBS-soaked sterile gelfoam (Surgifoam) to clear away any blood, which was minimal. Lastly, a 5-mm coverslip was glued to the top of the burr hole on the headplate. Animals were then immediately moved to a heating pad and monitored until they regained consciousness. Animals typically recovered fully within 15 min and showed little impact of the headplate implantation on normal functions such as eating and drinking (without any weight loss).
Experimental design and statistical analyses
Standard two-way ANOVA analyses were used in cases where WT and Ppt1−/−, treated and untreated groups were compared, as indicated in the figure legends. In cases where only pair-wise comparisons were required, standard Student's t tests were used (two-tailed). The number of animals and replicates performed for each experiment can be found in the corresponding figure legend. Similarly, the details of all statistical testing are also described in the figure legends. All statistical analyses were performed by GraphPad Prism 9.0.1.
Results
Developmental disruption of excitatory synaptic activity in Ppt1−/− visual cortex
We first conducted ex-vivo electrophysiological recordings from layer 2/3 visual cortical pyramidal neurons using an internal solution that enables concurrent acquisition of AMPAR-mediated and GABAA receptor (GABAAR)-mediated synaptic currents at a single-cell level (Flores-Barrera et al., 2017), which has previously been validated using pharmacology (Flores-Barrera et al., 2020). At postnatal day (P)14–P15 (before cortical critical period plasticity), both wild-type (WT) and Ppt1−/− mice showed similar levels of excitatory (PSC−60 mV) and inhibitory (PSC+15 mV) synaptic activity onto layer 2/3 pyramidal neurons (Fig. 1A,C). Interestingly, a reduction in the frequency of AMPAR-mediated synaptic currents emerged in Ppt1−/− visual cortical neurons at P28–P30, while the level of GABAAR-mediated transmission remained comparable to the WT group (Fig. 1B,D). This disruption was not associated with changes in the mean amplitude of the excitatory events (11.8 ± 0.9 pA in WT vs 10.7 ± 0.5 pA in Ppt1−/−, p = 0.3, unpaired t test), similar to findings from primary hippocampal neurons (Sapir et al., 2019).
Developmental AMPAR transmission is decreased in Ppt1−/− visual cortex. A, Quantification of sEPSC (−60 mV) and sIPSC (+15 mV) frequency from layer 2/3 visual cortical neurons in WT and Ppt1−/− mice at P14–P15. n = 6–7 cells, 3–4 animals/group. B, Quantification of sEPSC (−60 mV) and sIPSC (+15 mV) frequency from layer 2/3 visual cortical neurons in WT and Ppt1−/− mice at P28–P30. t test: ****p < 0.0001. n = 9–11 cells, 3–4 animals/group. C, Representative voltage-clamp recordings of sEPSCs and (D) sIPSCs from layer 2/3 visual cortical neurons at P28–P30. E, Representative immunoblots and quantification of total GluA1 levels in lysates from WT and Ppt1−/− visual cortices for ages P11–P60. Ppt1−/− data are normalized to the WT mean value at each age and comparisons were made within age. n = 2–4. F, Representative immunoblots and quantification of total GluA2 levels in lysates from WT and Ppt1−/− visual cortices for ages P11–P60. Ppt1−/− data are normalized to the WT mean value at each age and comparisons were made within age. n = 2–4. G, Quantification of the combined average of GluA1 and GluA2 levels in each occipital cortex lysate from WT and Ppt1−/− visual cortices for all ages examined. n = 19/group. H, Representative immunoblots and quantification of GluA1 levels in synaptosomes from WT and Ppt1−/− visual cortices for ages P11–P60. Ppt1−/− data are normalized to the WT mean value at each age and comparisons were made within age. n = 2–4. I, Representative immunoblots and quantification of GluA2 levels in synaptosomes from WT and Ppt1−/− visual cortices for ages P11–P60. Ppt1−/− data are normalized to the WT mean value at each age and comparisons were made within age. n = 2–4. J, Quantification of the combined average of GluA1 and GluA2 levels in each occipital cortex synaptosome from WT and Ppt1−/− visual cortices for all ages examined. n = 17/group. K, Representative immunoblots of APEGS-processed synaptosomes probing for GluA1, GluA2, and β-actin at P42. L, Quantification of the palmitoylated/nonpalmitoylated ratio (normalized to β-actin) of GluA1 and (M) GluA2 in APEGS-processed visual cortical synaptosomes. n = 3 mice/group. Data are mean ± SEM.
Despite this functional reduction in excitatory transmission, immunoblotting for GluA1 and GluA2 subunit levels revealed no significant change throughout development (P11, P14, P28, P33, P42, P60) between WT and Ppt1−/− visual cortices (Fig. 1E–J). This was true for both lysates (Fig. 1E–G) and synaptosomes (Fig. 1H–J). Further, we did not observe any baseline changes in GluA1 or GluA2 palmitoylation in Ppt1−/− synaptosomes (Fig. 1K–M).
Exaggerated homeostatic response following cortical silencing in Ppt1−/− neurons
We next asked whether the preferential disruption of excitatory AMPAR-mediated synaptic activity in Ppt1−/− cortical neurons is associated with a change in homeostatic plasticity (G. Turrigiano et al., 1998). To test this hypothesis, we implemented a designer receptor exclusively activated by designer drugs (DREADD)-based approach, targeting the local GABAergic system using an excitatory (hM3Dq) DREADD under the control of the Dlx5 promoter (Dimidschstein et al., 2017; Fig. 2A). The rationale for targeting local GABAergic neurons that they appear to be unaltered in young Ppt1−/− mice. Furthermore, this approach is commonly used to elicit robust suppression of cortical activity (Babl et al., 2019; Beltramo and Scanziani, 2019; Bennett et al., 2019; Blot et al., 2021), allowing us to focally induce synaptic upscaling in the visual cortex through enhanced GABAergic activity (Wu et al., 2021; Fig. 2A). Relative to the GFP-control group, activation of hM3Dq-DREADD caused an exaggerated increase of GluA1 levels in Ppt1−/− visual cortical synaptosomes (Fig. 2B). These results indicate that despite comparable levels of GluA1 at baseline, a challenge to the neural circuit generated a gross abnormality in the homeostatic regulation of AMPARs in Ppt1−/− neurons.
Homeostatic response of Ppt1−/− cortical neurons to chronic silencing is excessive upregulation of GluA1. A, Schematic of the DREADD-induced challenge paradigm and timeline (top). Representative images of DLX-hM3Dq injection in the left visual cortex (bottom right) of a WT mouse aligned to the Allen Brain atlas (bottom left). Scale bar = 50 µm for both low-magnification and high-magnification images. B, Representative immunoblot of synaptosomal GluA1 from WT, WT-DREADD, Ppt1−/−, and Ppt1−/−-DREADD occipital cortices and GFP-injected controls (top, all samples from same gel rearranged to match quantification) and quantification (bottom). n = 3–6 animals/group. Two-way ANOVA: interaction genotype × DREADD (F(1,17) = 8.740, **p = 0.0088); main effect of genotype (F(1,17) = 8.740, **p = 0.0088); main effect of DREADD (F(1,17) = 56.65, ****p < 0.0001). Tukey's multiple comparison indicated on graph: *p = 0.0280 WT versus DREADD-WT; ****p < 0.0001 Ppt1−/− versus DREADD-Ppt1−/−; **p = 0.0057 DREADD-WT versus DREADD-Ppt1−/−.
Synaptic upscaling of GluA1 is exaggerated in Ppt1−/− neurons in vitro
Considering the aberrant homeostatic regulation of GluA1 levels we observed in vivo, we next determined how loss of Ppt1 impacts AMPAR trafficking during homeostatic plasticity by inducing synaptic upscaling in vitro (O'Brien et al., 1998; G. Turrigiano et al., 1998). Specifically, we incubated WT and Ppt1−/− primary cortical neurons with vehicle control (Veh) or TTX (1 μm) for 48 h to induce synaptic upscaling, as previously reported (O'Brien et al., 1998; G. Turrigiano et al., 1998; Shepherd et al., 2006; Diering et al., 2014; Sanderson et al., 2018; Fig. 3).
Synaptic upscaling of GluA1 is exaggerated in primary cortical Ppt1−/− neurons. A, Representative images of surface immunolabeled GluA2 in DIV16 WT and Ppt1−/− cortical neurons, treated with vehicle control (DMSO, final concentration 0.1%) or TTX (1 μm) for 48 h. Scale bars for low-magnification images = 20 µm, for high-magnification images = 5 µm. B, Quantification of the number of surface immunolabeled GluA2 puncta over 20 µm of dendrite at a standard distance 50–100 µm from the soma. n = 16–23 cells/group across 2 independent cultures. Data are normalized to the WT vehicle control (Veh) condition. Two-way ANOVA: interaction genotype × treatment (F(1,4) = 5.008, ns); main effect of treatment (F(1,4) = 199, ***p = 0.0001); main effect of genotype (F(1,4) = 30.99, **p = 0.0051). Tukey's multiple comparison indicated on graph: **p = 0.0038 WT Veh versus WT TTX; *p = 0.0179 WT Veh versus Ppt1−/− Veh; **p = 0.0011 Ppt1−/− Veh versus Ppt1−/− TTX. C, Quantification of the percent area covered by surface immunolabeled GluA2 puncta over 20 µm of dendrite at a standard distance 50–100 µm from the soma. n = 16–23 cells/group across 2 independent cultures. Two-way ANOVA: interaction genotype × treatment (F(1,4) = 0.1787, ns); main effect of treatment (F(1,4) = 85.69, ***p = 0.0008); main effect of genotype (F(1,4) = 20.03, *p = 0.0110). Tukey's multiple comparison indicated on graph: **p = 0.0082 WT Veh versus WT TTX; *p = 0.0115 Ppt1−/− Veh versus Ppt1−/− TTX. D, Representative images of surface immunolabeled GluA1 in DIV16 WT and Ppt1−/− cortical neurons treated with vehicle control (DMSO, final concentration 0.1%) or TTX (1 μm) for 48 h. Scale bars for low-magnification images = 20 µm, for high-magnification images = 5 µm. E, Quantification of the number of surface immunolabeled GluA1 puncta over 20 µm of dendrite at a standard distance 50–100 µm from the soma. n = 3–5 cultures/group (total number of cells analyzed are listed for each group on the graph). Data are normalized to the WT vehicle control (Veh) condition. Two-way ANOVA: interaction genotype × treatment (F(3,18) = 5.620, **p = 0.0067); main effect of genotype (F(1,18) = 10.06, **p = 0.0053); main effect of treatment (F(3,18) = 52.74, ****p < 0.0001). Tukey's multiple comparison indicated on graph: **p = 0.0097 WT Veh versus WT TTX; ****p < 0.0001 Ppt1−/− Veh versus Ppt1−/− TTX; *p = 0.0218 WT TTX versus Ppt1−/− TTX. Data represent mean ± SEM. F, Quantification of the percent area covered by surface immunolabeled GluA1 puncta over 20 µm of dendrite at a standard distance of 50–100 µm from the soma. n = 3–5 cultures/group (total number of cells analyzed are listed for each group on the graph). Data are normalized to the WT vehicle control condition. Two-way ANOVA: interaction genotype × treatment (F(3,18) = 26.21, ****p < 0.0001); main effect of genotype (F(1,18) = 51.25, ****p < 0.0001); main effect of treatment (F(3,18) = 208.2, ****p < 0.0001). Tukey's multiple comparison indicated on graph: ****p < 0.0001 WT Veh versus WT TTX; ****p < 0.0001 Ppt1−/− Veh versus Ppt1−/− TTX; ****p < 0.0001 WT TTX versus Ppt1−/− TTX. G, Representative immunoblot of surface GluA1 and transferrin control following surface biotinylation assay in DIV16 cortical neuron cultures. Cells were treated with vehicle (DMSO, final concentration 0.1%), or TTX (1 μm) for 48 h. Note that all displayed immunoblot samples are from the same gel. H, Quantification of the surface GluA1 levels in WT and Ppt1−/−-treated and Ppt1−/−-untreated cortical neurons, normalized to transferrin for each experiment and expressed as percent change versus vehicle condition. n = 6 independent cultures. Note that all samples displayed are from the same gel. Two-way ANOVA: interaction genotype × treatment (F(2,10) = 7.061, *p = 0.0122); main effect of genotype (F(1,5) = 14.21, *p = 0.0130); main effect of treatment (F(2,10) = 34.77, ****p < 0.0001). Tukey's multiple comparison indicated on graph: *p = 0.0240 WT Veh versus WT TTX; ***p = 0.0002 Ppt1−/− Veh versus Ppt1−/− TTX; **p = 0.0038 Ppt1−/− TTX versus WT TTX. Data represent mean ± SEM. I, Representative immunoblots of GluA1, GluA2, and β-actin loading control for input samples (lysates). n = 6 cultures. Note that all displayed immunoblot samples are in fact from the same gel. J, Quantification of input GluA1 levels in WT and Ppt1−/−-treated and Ppt1−/−-untreated cortical neurons, normalized to β-actin for each experiment and expressed as percent change versus vehicle condition. n = 6 independent cultures. Two-way ANOVA: no interaction genotype × treatment (F(2,10) = 2.902, p = 0.1014); no main effect of genotype (F(1,5) = 2.432e-005, p = 0.9963); main effect of treatment (F(2,10) = 14.81, **p = 0.0010). Tukey's multiple comparison indicated on graph: *p = 0.0106 WT Veh versus WT TTX; ***p = 0.0001 Ppt1−/− Veh versus Ppt1−/− TTX. Data represent mean ± SEM. K, Quantification of input GluA2 levels in WT and Ppt1−/−-treated and Ppt1−/−-untreated cortical neurons, normalized to β-actin for each experiment and expressed as percent change versus vehicle condition. n = 6 independent cultures. Two-way ANOVA: interaction genotype × treatment (F(2,10) = 9.617, **p = 0.0047); no main effect of genotype (F(1,5) = 1.699, p = 0.2492); no main effect of treatment (F(2,10) = 3.316, p = 0.0786). Data represent mean ± SEM. ns = non significant.
Synaptic upscaling in cultured neurons is known to drive synaptic insertion of GluA2-containing AMPARs (Ibata et al., 2008; Pozo et al., 2012; Tan et al., 2015; Silva et al., 2019). Therefore, we first tested whether surface GluA2 levels were increased in upscaled (TTX-treated) WT and Ppt1−/− neurons using an immunofluorescence assay in which we labeled surface GluA2 under nonpermeabilizing conditions (Fig. 3A). Corroborating our in vivo electrophysiological findings (Fig. 1), Ppt1−/− cells demonstrated a decreased GluA2 puncta count compared with WT neurons, suggestive of fewer AMPAR-containing synapses (Fig. 3B). Similar to previous studies (Wierenga et al., 2005), TTX treatment induced upscaling of GluA2, which occurred to an equal degree in WT and Ppt1−/− neurons as measured by an increase in GluA2 puncta count and percent area covered by GluA2 puncta (Fig. 3B,C).
Next, we performed the same surface immunofluorescence assay for the GluA1 AMPAR subunit (Fig. 3D). Again, TTX-induced upscaling was evident in both WT and Ppt1−/− neurons; however, in contrast to our data examining GluA2, upscaling of GluA1 was significantly exaggerated in Ppt1−/− cells (Fig. 3E,F). We obtained similar results when upscaling was induced with the AMPAR-specific antagonist, CNQX (20 μm; Gong et al., 2007; Fong et al., 2015; Fig. 4A–C). To validate these immunofluorescence data, we analyzed synaptic upscaling of GluA1 using biochemical surface biotinylation assay (Yoshii et al., 2013) and found that TTX treatment-induced synaptic upscaling was again greater in Ppt1−/− cells (Fig. 3G,H). Immunoblots of inputs from the same cell lysates showed an increase in total GluA1 levels to an equal degree in both WT and Ppt1−/− neurons after upscaling (Fig. 3I,J), suggesting the differences in GluA1 surface expression is not because of increased protein abundance. We did not observe a significant increase in the total levels of GluA2 in either WT or Ppt1−/− under these conditions (Fig. 3K).
CNQX-induced exaggerated upscaling and impaired downscaling in Ppt1−/− cortical neurons. A, Representative images of surface immunolabeled GluA1 in DIV16 WT and Ppt1−/− cortical neurons, treated with bicuculline (20 μm) or CNQX (1 μm) for 48 h. Vehicle controls are displayed in Figure 3. Scale bars for low-magnification images = 20 µm, for high-magnification images = 5 µm. B, Quantification of the number of surface immunolabeled GluA1 puncta over 20 µm of dendrite at a standard distance 50–100 µm from the soma. n = 2–5 cultures/group (total number of cells analyzed are listed for each group on the graph). Data are normalized to the WT vehicle control (Veh) condition. Two-way ANOVA: interaction genotype × treatment (F(3,18) = 5.620, **p = 0.0067); main effect of genotype (F(1,18) = 10.06, **p = 0.0053); main effect of treatment (F(3,18) = 52.74, ****p < 0.0001). Tukey's multiple comparisons shown on graph: *p = 0.0218 WT Veh versus WT CNQX; ****p < 0.0001 Ppt1−/− Veh versus Ppt1−/− CNQX. C, Quantification of the percent area covered by surface immunolabeled GluA1 puncta over 20 µm of dendrite at a standard distance of 50–100 µm from the soma. n = 2–5 cultures/group (total number of cells analyzed are listed for each group on the graph). Data are normalized to the WT vehicle control condition. Two-way ANOVA: interaction genotype × treatment (F(3,18) = 26.21, ****p < 0.0001); main effect of genotype (F(1,18) = 51.25, ****p < 0.0001); main effect of treatment (F(3,18) = 208.2, ****p < 0.0001). Tukey's multiple comparisons shown on graph: *p = 0.0425 WT Bic versus WT Veh; ****p < 0.0001 WT Veh versus WT CNQX; ****p < 0.0001 Ppt1−/− Veh versus Ppt1−/− CNQX; ***p = 0.0004 WT CNQX versus Ppt1−/− CNQX. D, Representative immunoblot of surface GluA1 and transferrin control following surface biotinylation assay in DIV16 cortical neuron cultures. Cells were treated with bicuculline (20 μm) or vehicle (DMSO, final concentration 0.1%) for 48 h. Note that all samples displayed are from the same gel. E, Quantification of the surface GluA1 levels in WT and Ppt1−/− treated and untreated cortical neurons, normalized to transferrin for each experiment and expressed as percent change versus vehicle condition. n = 6 independent cultures. Two-way ANOVA: interaction genotype × treatment (F(2,10) = 7.061, *p = 0.0122); main effect of genotype (F(1,5) = 14.21, *p = 0.0130); main effect of treatment (F(2,10) = 34.77, ****p < 0.0001). Tukey's multiple comparison indicated on graph: *p = 0.0418 WT Bic versus WT Veh. F, Representative immunoblots of GluA1, GluA2, and β-actin loading control for input samples (lysates). n = 6 cultures. Note that all displayed immunoblot samples for a given protein are from the same gel. G, Quantification of input GluA1 levels in WT and Ppt1−/−-treated and Ppt1−/−-untreated cortical neurons, normalized to β-actin for each experiment and expressed as percent change versus vehicle condition. n = 6 independent cultures. H, Quantification of input GluA2 levels in WT and Ppt1−/−-treated and Ppt1−/−-untreated cortical neurons, normalized to β-actin for each experiment and expressed as percent change versus vehicle condition. n = 6 independent cultures. Two-way ANOVA: interaction genotype × treatment (F(2,10) = 9.617, **p = 0.0047); no main effect of genotype (F(1,5) = 1.699, p = 0.2492); no main effect of treatment (F(2,10) = 3.316, p = 0.0786). Tukey's multiple comparison indicated on graph: *p = 0.0112 WT Bic versus WT Veh.
GluA1 synaptic membrane retention is increased in upscaled Ppt1−/− neurons. A, Representative image of a SEP-GluA1 transfected, upscaled neuron (top) and a 20-µm segment of dendrite before, immediately after, and 10 min after bleaching of an individual spine (arrow, bottom). Scale bar = 20 µm for low-magnification images, 10 µm for high-magnification images. B, Time-dependent recovery of SEP-GluA1 fluorescence after photobleaching (FRAP) in WT and Ppt1−/− primary cortical neurons. Five-six spines were bleached for each cell and averaged for to make one n. Trace represents average of n = 5 cells across 2 independent cultures. C, Nonlinear fit of (single phase decay) of SEP-GluA1 fluorescence after photobleaching (FRAP) in upscaled WT and Ppt1−/− primary cortical neurons. Comparison of nonlinear fits: F(3,220) = 542.9, ****p < 0.0001. n = 5 cells across 2 independent cultures. D, Quantification of the immobile fraction (percent unrecovered compared with prebleach fluorescence) of SEP-GluA1 fluorescence after photobleaching (FRAP) for each cell. t test: *p = 0.0356. n = 5 cells across 2 independent cultures, 5–6 spines/cell. Data represent mean ± SEM.
CP-AMPAR upregulation is more robust in upscaled Ppt1−/− neurons. A, Representative calcium imaging (GCaMP) video still frames of vehicle control or TTX-treated (1 μm, 48 h) WT and Ppt1−/− cortical neurons. Scale bar = 20 µm. B, Representative ΔF/F0 traces for ten synapses in each condition (WT, Ppt1−/−, TTX-WT, and TTX-Ppt1−/−). Open circles denote NASPM-insensitive synapses and closed circles denote NASPM-sensitive synapses. C, Quantification of the average synaptic calcium activity (number of frames >3 SDs from baseline) for all cells from each condition before NASPM infusion (i.e., at baseline). Two-way ANOVA: interaction genotype × TTX-treatment F(1,16) = 6.349, *p = 0.0228; main effect of genotype = F(1,16) = 3.205, p = 0.1669; main effect of treatment F(1,16) = 37.15, ****p < 0.0001. Tukey's multiple comparison indicated on graph: *p = 0.0442 WT Veh versus WT TTX; ****p < 0.0001 Ppt1−/− Veh versus Ppt1−/− TTX; *p = 0.0304 WT TTX versus Ppt1−/− TTX. n = 4–6 cells/group, cells taken from at least 2 independent cultures. D, Quantification of the average synaptic calcium activity (number of frames >3 SDs from baseline) for all synapses in all cells from each condition before, during, and following NASPM (1 μm) bath infusion. Repeated measures two-way ANOVA: interaction genotype/TTX × NASPM F(3,16) = 15.74, ****p < 0.0001; main effect of genotype/TTX F(3,16) = 1.731, p = 0.2009; main effect of NASPM F(1,16) = 133.4, ****p < 0.0001. Bonferroni's multiple comparison test shown on graph for baseline, NASPM: **p = 0.0067 WT Veh; ****p < 0.0001 WT TTX; ****p < 0.0001 Ppt1−/− TTX. n = 4–6 cells/group, cells taken from at least 2 independent cultures. E, Quantification of the proportion of NASPM-sensitive synapses in WT, TTX-treated WT, Ppt1−/−, and TTX-treated Ppt1−/− cortical neurons expressed as the percentage of synapses inhibited during NASPM infusion. Two-way ANOVA: interaction genotype × TTX F(1,16) = 6.167, *p = 0.0245; main effect of genotype F(1,16) = 3.126, p = 0.0961; main effect of treatment F(1,16) = 82.29, ****p < 0.0001. n = 4–6 cells/group, cells taken from at least 2 independent cultures. Data represent mean ± SEM.
We also examined whether synaptic downscaling was affected by Ppt1. Indeed, while bicuculline (20 μm, 48 h) treatment induced downscaling of GluA1 in WT neurons, like in previous studies (Shepherd et al., 2006; Diering et al., 2014), synaptic downscaling was completely absent in Ppt1−/− neurons (Fig. 4A–E). This is despite no change in total levels of GluA1 between bicuculline-treated WT and Ppt1−/− neurons, although we did see a decrease in total levels of GluA2 in downscaled WT neurons (Fig. 4F–H).
Taking a ratio of the degree of upscaling of GluA1 versus GluA2 in WT and Ppt1−/− neurons reveals an outsized increase of GluA1 in upscaled Ppt1−/− cells (using percent area measure: 1.32 GluA1/GluA2 upscaling ratio in WT-TTX-treated cells vs 2.72 GluA1/GluA2 upscaling ratio in Ppt1−/−-TTX-treated cells). These results indicate that while GluA2 upscaling is comparable in WT and Ppt1−/− neurons, excessive GluA1 upscaling in Ppt1−/− neurons could lead to an exaggerated incorporation of GluA2-lacking AMPARs at glutamatergic synapses. Future studies are needed to establish whether the unique GluA1 scaling defect at glutamatergic synapses in Ppt1−/− mice is limited to a specific neuronal population (pyramidal neurons vs interneurons) since the approach we used (microtubule-associated protein 2 immunostaining) does not differentiate excitatory and inhibitory interneurons in cortical cultures, which can contain substantial GABAergic populations (Neale et al., 1983; Huettner and Baughman, 1986; Alho et al., 1988; Baltz et al., 2010).
GluA1 synaptic membrane retention is increased in upscaled Ppt1−/− neurons
To establish the mechanism underlying excessive synaptic expression of GluA1 during upscaling, we measured fluorescence recovery after photobleaching (FRAP) of the superecliptic (SEP)-GluA1 subunit (Kopec et al., 2006) in upscaled WT and Ppt1−/− cultured neurons following synaptic upscaling (Fig. 5A). This analysis revealed a slower recovery of photobleached SEP-GluA1 signal at individual synapses in Ppt1−/− cells compared with WT (Fig. 5B,C). Accordingly, Ppt1−/− neurons exhibited an increased immobile fraction of SEP-GluA1 15 min after photobleaching (Fig. 5D). These results point to a decreased mobility of GluA1-containing AMPARs once at the synaptic membrane of upscaled Ppt1−/− neurons.
CP-AMPAR upregulation is more robust in upscaled Ppt1−/− neurons
Previous studies specifically implicate calcium-permeable (CP-) AMPARs in synaptic upscaling (Desai et al., 2002; Goel et al., 2006, 2011; Goel and Lee, 2007). The increased GluA1/GluA2 upscaling ratio in Ppt1−/− neurons suggests a greater surface expression of CP-AMPARs in those cells, although our measurement of GluA1 levels (Fig. 3) does not exclude the upregulation of calcium-impermeable AMPARs. Therefore, we next discerned whether CP-AMPARs comprised the synaptic receptor population in overly upscaled Ppt1−/− synapses.
We performed live-cell calcium imaging in WT and Ppt1−/− primary cortical neurons transfected with GCaMP3 following a 48-h exposure to TTX or vehicle and perfused the CP-AMPAR-specific blocker NASPM into the bath (Fig. 6A; Movies 1, 2, 3, 4). Personnel blinded to the study manually selected all active synapses in each video and extracted their ΔF/F0 traces (Fig. 6B), which were binned into baseline (pre-NASPM infusion), NASPM (during NASPM infusion), and washout (following washout of NASPM) conditions for further analysis. In vehicle-treated neurons (no scaling), we found no difference in the number of synaptic calcium transients between WT and Ppt1−/− populations (Fig. 6C). Importantly, while TTX treatment increased the number of synaptic calcium transients in both WT and Ppt1−/− cells, consistent with the upscaling phenotype, upscaled Ppt1−/− neurons exhibited the highest number of baseline calcium transients (Fig. 6C). NASPM treatment abolished these differences between groups (Fig. 6D). Furthermore, the proportion of NASPM-sensitive synapses (see Materials and Methods) was similar between vehicle-treated WT and Ppt1−/− neurons, but upscaled Ppt1−/− neurons demonstrated significantly more NASPM-sensitive synapses than all other groups (Fig. 6E; Movies 1, 2, 3, 4). A limitation of the method is that our measurements likely overestimate the total number of CP-AMPAR containing synapses (e.g., the large number of NASPM sensitive synapses measured might be confounded by factors like the suppression of calcium activity across neighboring synapses by CP-AMPAR blockade at a single NASPM-sensitive site). Nevertheless, these data are consistent with the finding that GluA1 surface expression is exaggerated in upscaled Ppt1−/− neurons and suggest that a larger fraction of AMPARs are CP-AMPARs in these cells.
Live-cell calcium imaging of WT-vehicle-treated cell infused with 1 µm NASPM as indicated.
Live-cell calcium imaging of Ppt1-/--vehicle-treated cell infused with 1 µm NASPM as indicated.
Live-cell calcium imaging of WT-TTX-treated (1 µm, 48 h) cell infused with 1 µm NASPM as indicated.
Live-cell calcium imaging of Ppt1-/--TTX-treated (1 µm, 48 h) cell infused with 1 µm NASPM as indicated.
GluA1 palmitoylation is increased in Ppt1−/− neurons during synaptic scaling
To directly test whether GluA1 palmitoylation is associated with the enhanced incorporation of CP-AMPARs in Ppt1−/− neurons during synaptic scaling, we performed the Acyl-PEGyl Exchange Gel-Shift (APEGS) assay (Yokoi et al., 2016; Koster et al., 2019) in WT and Ppt1−/− neurons following downscaling, upscaling, or vehicle treatments (Fig. 7A). We measured the density of palmitoylated GluA1 (upper band in immunoblot) to the nonpalmitoylated fraction (lower band in immunoblot) to calculate the palmitoylated GluA1 ratio, for which the density plots are displayed (Fig. 7B). While we did not detect any significant change in GluA1 palmitoylation in WT cells following scaling, supporting a previous report (Yang et al., 2009), we observed a significant increase in the palmitoylated GluA1 ratio in upscaled Ppt1−/− neurons (Fig. 7C). In addition, total Ppt1 levels increased during synaptic upscaling in WT cells (Fig. 7A,D), implying that upscaling-dependent expression of Ppt1 typically prevents GluA1 subunits from uncontrolled palmitoylation and synaptic accumulation in WT neurons. This further supports the notion that excessive GluA1 insertion at Ppt1−/− synapses is attributable to aberrant palmitoylation.
Increase of GluA1 palmitoylation is associated with synaptic upscaling in Ppt1−/− neurons. A, Representative immunoblot of APEGS assay-processed cortical neurons lysates probing for GluA1 and β-actin. Representative blot for Ppt1 (input) also shown. Neurons were treated with bicuculline (20 μm), vehicle control (DMSO, final concentration 0.1%), or TTX (1 μm) for 48 h before APEGS assay. B, Representative immunoblot band density plots displaying distinct peaks for the palmitoylated (green arrow) and nonpalmitoylated fractions (orange arrow), which were quantified separately and expressed a ratio of palmitoylated/nonpalmitoylated GluA1. C, Quantification of the ratio of palmitoylated/nonpalmitoylated GluA1 levels, normalized to β-actin loading control. Two-way ANOVA: interaction genotype × treatment (F(2,24) = 4.211, *p = 0.0271); main effect of genotype (F(1,24) = 9.559, **p = 0.0050); main effect of treatment (F(2,24) = 11.80, ***p = 0.0003). Tukey's multiple comparison indicated on graph: **p = 0.0017 Ppt1−/− Veh versus Ppt1−/− TTX; **p = 0.0065 WT TTX versus Ppt1−/− TTX. n = 5 independent cultures. Data represent mean ± SEM. D, Quantification of the Ppt1 levels, normalized to β-actin loading control. Two-way ANOVA: interaction genotype × treatment (F(2,24) = 13.07, ***p = 0.0001); main effect of genotype (F(1,24) = 847.4, ****p < 0.0001); main effect of treatment (F(2,24) = 13.07, ***p = 0.0001). Tukey's multiple comparison indicated on graph: ***p = 0.0004 WT Veh versus WT TTX. n = 5 independent cultures. Data represent mean ± SEM.
In vivo imaging reveals coordinated calcium activity in the visual cortex of Ppt1−/− mice
To determine the circuit-level effects of dysregulated synaptic plasticity in Ppt1−/− mice, we employed in vivo two-photon calcium imaging in awake animals (Fig. 8A; Movie 5, Movie 6). We transduced layer 2/3 visual cortical neurons with GCaMP6f, implanted a cranial window over the left V1 area of injected animals around P21, and imaged visual cortical calcium activity at P28–P35 (recovery period from surgery was a minimum of 7 d).
In vivo imaging reveals coordinated calcium activity in the visual cortex of Ppt1−/− mice. A, Representative calcium imaging video still frames of layer 2/3 cortical neuronal somata from a WT and Ppt1−/− mouse. Scale bar = 50 µm. B, Representative ΔF/F0 traces for selected cells in each video. Note the coactivity of multiple cells in the Ppt1−/− brain demarked by gray boxes. All cells in each frame (not just selected cells) were used for further analyses in panels C and D. C, Correlation matrices of the deconvolved calcium activity (Persons r correlation values) of all cells in the representative videos (Movie 5, Movie 6). Each square represents 1 cell, as denoted in the top left. D, Averaged pair-wise correlation value (Fisher transformed) for all videos, from each animal (experimental data; Exp.) compared with average correlations produced from the same Ca2+ transient datasets randomized (Rand). Each point represents the average for one mouse. t test: *p = 0.0419. Square points denote the representative correlation matrices from panel C. n = 5 mice/group. Data represent mean ± SEM.
Representative spontaneous neuronal population activity in the visual cortex of an awake and behaving WT mouse using in vivo, two-photon calcium imaging.
Representative spontaneous neuronal population activity in the visual cortex of an awake and behaving Ppt1-/- mouse using in vivo, two-photon calcium imaging.
We monitored visual cortical neuron activity for 5 min epochs in WT and Ppt1−/− mice and extracted the ΔF/F0 trace for each active neuron in each video (representative examples are displayed in Fig. 8B; Movie 5, 6). The average activity levels in WT and Ppt1−/− visual cortical neurons was indistinguishable (8.863 ± 1.409 somatic spikes per cell in WT animals vs 7.772 ± 1.356 somatic spikes per cell in Ppt1−/− animals, p = 0.5753 by t test). However, whereas somatic calcium transients across neurons in WT animals typically varied in time (i.e., were temporally sparse), calcium transients in Ppt1−/− visual cortical neurons often occurred simultaneously with several other cells, followed by periods of relative silence (Fig. 8B; Movie 5, Movie 6). Indeed, analyzing the pairwise co-activity of all neurons in each video (all neurons in the field of view were considered) revealed that compared with WT, the calcium activity of Ppt1−/− visual cortical neurons demonstrated significantly increased co-activity (Fig. 8C). Accordingly, the average pairwise correlation (Fisher-corrected Pearson's r) of neuronal calcium activity was increased in Ppt1−/− animals compared with WT counterparts, an effect that was abolished by randomly shuffling the deconvolved somatic Ca2+ transient data (Fig. 8D). These results indicate that, despite a decrease in the number of AMPAR-containing excitatory synapses in Ppt1−/− cortical neurons (Fig. 1), these cells still demonstrate a propensity for coordinated calcium activity in awake animals.
Discussion
We demonstrate herein that loss of Ppt1 dysregulates AMPAR palmitoylation and trafficking, particularly when neurons are challenged with a homeostatic plasticity paradigm. We also show that impaired AMPAR trafficking contributes to disrupted excitatory synapse formation in the visual cortex of developing Ppt1−/− mice. Together, these findings are consistent with previous in vitro (Sapir et al., 2019) and in vivo (Koster et al., 2019) observations, adding to the growing evidence that homeostatic and CP-AMPAR plasticity are disrupted in a wide range of neurologic disorders (Noh et al., 2005; Pratt et al., 2011; Blackman et al., 2012; Qiu et al., 2012; Rajasekaran et al., 2012; Netzahualcoyotzi and Tapia, 2015; Quintana et al., 2015; Gilbert et al., 2017; Styr and Slutsky, 2018; Yamashita and Kwak, 2019).
The principal result of the current study is that lack of depalmitoylation by Ppt1 causes an aberrant upregulation of CP-AMPARs in the homeostatic response to neural circuit silencing. There are two proximal explanations from our data: (1) overpalmitoylation of GluA1 directly causes its increased surface expression during upscaling or (2) AMPAR-associated postsynaptic proteins are overly palmitoylated and indirectly act to disrupt the physiological turnover of CP-AMPARs, resulting in their surface retention during scaling. Gorenberg and colleagues recently reported that the extracellular cysteine of GluA1, C323, is likely to be the site of depalmitoylation by Ppt1 (Gorenberg et al., 2022). Although the trafficking effects of GluA1 palmitoylation at this site have not yet been studied, it is likely to act as a latent signal, whereby GluA1 is palmitoylated at C323 during trafficking to the synapse but forms a disulfide bond with C75 on exposure to the synaptic cleft (Cijsouw et al., 2018; Bechtel et al., 2020). Indeed, mutation of this residue to prevent its palmitoylation (C323A) completely abolishes glutamate-evoked responses in a heterologous expression system (Gorenberg et al., 2022). Therefore, one possible explanation for our findings is that excessive GluA1 palmitoylation at C323 because of loss of Ppt1 increases its trafficking to, or retention at the synaptic membrane. Similarly, palmitoylation also influences the turnover of specific proteins by inhibiting their ubiquitination, either directly or by limiting their endocytosis and exposure to ubiquitination enzymes (Linder and Deschenes, 2007). Considering GluA1 ubiquitination increases during synaptic downscaling (Scudder et al., 2014; Jewett et al., 2015), it is plausible that palmitoylation of GluA1 prevents such a signal during homeostatic upregulation. In this regard, GluA1 overpalmitoylation during synaptic upscaling may be directly responsible for excessive accumulation of CP-AMPARs in Ppt1−/− neurons (Fig. 7).
Beyond a direct mechanism, the aberrant trafficking of CP-AMPARs in Ppt1−/− neurons may also arise from abnormal palmitoylation of AMPAR-interacting postsynaptic molecules, including members of the adaptor protein 2 (AP2) complex, which were also recently identified as Ppt1 substrates (Gorenberg et al., 2022). An intriguing alternative is the A-kinase anchor protein 5 (Akap5), a critical postsynaptic signaling scaffold specifically implicated in the regulation of CP-AMPARs during synaptic scaling (Sanderson et al., 2018). Regardless of whether absent depalmitoylation of GluA1 itself or associated synaptic molecules is responsible for our findings, the broader conclusion stands, balanced palmitoylation underpins proper homeostatic regulation of AMPARs, which goes unchecked in the absence of Ppt1 activity.
Regarding the precise plasticity mechanism disrupted by loss of Ppt1, recent results (M.C. Bridi et al., 2018; Rodriguez et al., 2019; M. Bridi et al., 2022; Lee, 2023) suggest that homeostatic AMPAR and NMDAR plasticity are intertwined in a model termed sliding threshold metaplasticity (Bienenstock et al., 1982). The crux of metaplasticity is that neurons change their threshold for synaptic modification (strengthening or weakening) based on the recent history of incoming activity levels, which allows for compensation to prevent runaway Hebbian loops (Lee, 2023). Given this recent shift in understanding, we anticipate that our electrophysiological findings (Fig. 1), which diverge slightly from the classical synaptic scaling model (Desai et al., 2002), represent the outcome of failed metaplasticity in the developing Ppt1−/− brain. Indeed, our previous data demonstrate that NMDAR immaturity is detectable at a similar age to the AMPAR transmission deficits discovered here (Koster et al., 2019). This would also provide a mechanism by which disrupted NMDAR plasticity in Ppt1−/− neurons contributes to the propensity to upregulate CP-AMPARs, as observed in other systems (Christian et al., 2021). However, with extreme manipulations of neural activity like those we drove in vitro, synaptic scaling likely dominates (Lee and Kirkwood, 2019; Lee, 2023), and is therefore more likely to underpin our findings in those systems (Figs. 3–7). Future experiments will address whether blocking GluN2B can alleviate the excessive upregulation of GluA1, which would specifically implicate impairment of metaplasticity rather than synaptic scaling in Ppt1−/− neurons.
Similarly, although we demonstrate that visual cortical neurons favor a state of synchronized calcium activity in vivo (Fig. 8), this analysis was performed under basal conditions, suggesting CP-AMPAR dysregulation in Ppt1−/− neurons is only one potential underlying mechanism for its emergence. On one hand, a computational study shows that artificial deafferentiation of a population of synthetic neurons drives pathologic synchrony in their activity because of a homeostatic plasticity (Fröhlich et al., 2008), suggesting that this type of maladaptation can drive a runaway synchrony under some circumstances. On the other hand, developmental cortical synchrony (i.e., correlated calcium activity) is intimately tied to NMDAR function (Garaschuk et al., 2000; Okada et al., 2003), which is dysregulated in Ppt1−/− animals in such a way to bias cortical neurons toward increased correlated activity (Koster et al., 2019), again implicating metaplasticity (Lee, 2023). Moreover, while the influence of GABAergic circuitry on developmental calcium oscillations is not quite straightforward (Moody and Bosma, 2005), it is required for driving oscillatory activity in neural networks in vitro (Voigt et al., 2001), implicated in early developmental calcium propagation (Ben-Ari et al., 2007; Minlebaev et al., 2007), and contributes to appropriate development of glutamatergic synapses (Kirsch and Betz, 1998). Considering our electrophysiological recordings reveal an imbalance of excitatory and inhibitory circuitry in Ppt1−/− animals (Fig. 1), we cannot rule out abnormal circuit dynamics in the generation of cortical synchrony.
Despite the precise mechanism(s) driving this coordinated cortical calcium activity, the phenomenon itself runs counter to the developmental trajectory of healthy cortical network activity, which typically desynchronizes during development to expand their information encoding capacity (Golshani et al., 2009; Rochefort et al., 2009; He et al., 2018). We postulate that this pathologic network activity (synchrony) contributes to seizure activity with worsening pathophysiology in Ppt1−/− mice, as a similar failure to desynchronize is observed in a mouse model of Fragile X syndrome (Gonçalves et al., 2013).
Taken together, we interpret our results as a kindling model: loss of Ppt1 function holds immature NMDARs at synapses, pathologically lowering the plasticity threshold that drives synaptic CP-AMPAR accumulation, ultimately contributing to network synchronization. Excessive synchronous calcium influx drives excitotoxic cell death (Berliocchi et al., 2005; Hardingham and Bading, 2010; Koster et al., 2019) and challenges efferent brain regions to respond via homeostatic plasticity. Sequential scaling-synchronization-excitotoxicity cycles, which have also been proposed to contribute to the pathogenesis of Alzheimer disease (Small, 2008), drives a spreading neurodegeneration in parallel with seizure activity, as observed in CLN1 (Koster and Yoshii, 2019).
Unfortunately, the therapeutic landscape for CLN1 is still particularly limited. Targeting downstream processes in CLN1, like lysosomal waste buildup, has thus far provided inadequate relief to patients despite successfully clearing accumulations (Gavin et al., 2013; Levin et al., 2014). Considering the mounting evidence for a significant synaptic role for Ppt1, new treatments aimed at these pathways may prove beneficial in CLN1. For instance, targeting NMDAR function to improve pathologic synchronous oscillatory activity, as recently shown to be efficacious in Alzheimer disease and epilepsy models (Hanson et al., 2020), may prove beneficial in CLN1. Alternatively, PPT1-mimetics like N-tert-butyl-hydroxylamine (NtBuHA) are efficacious in Ppt1−/− mice (Sarkar et al., 2013), and recent data show that this small molecule reduces AMPAR transmission (Xia et al., 2019); although it remains to be directly tested how much of the benefit is imparted by correction of synaptic protein overpalmitoylation. Identification of the specific palmitoylating and depalmitoylating enzymes that act on molecules of interest, like AMPARs, will provide novel, more precise therapeutic options to balance aberrant palmitoylation. At minimum, our findings lend credence to the recent clinical use of the AMPAR antagonist Perampanel as a second line treatment for CLN1 (Augustine et al., 2021).
In conclusion, we demonstrate here with an interdisciplinary approach how destabilizing depalmitoylation disrupts synaptic plasticity by impairing AMPAR expression. These results urge fresh approaches to the translational effort to better treat this devastating neurodegenerative disease by mimicking the synaptic functions of PPT1. Even more, our data demonstrate that a disruption to the balance of a dynamic post-translational modification is sufficient to severely impair patterns of synaptic plasticity that are required for typical brain function.
Footnotes
This work is supported by startup funding awarded to A.Y. by the University of Illinois Chicago, Department of Anatomy and Cell Biology and by National Institutes of Health Gants R43AG072984 (to A.Y.) and R01MH086507 (to K.Y.T.). We thank Dr. Sandra Hofmann (University of Texas Southwestern) for the anti-PPT1 antibody, Dr. Froylan Calderon de Anda (Universitätsklinikum Hamburg-Eppendorf) for the GCaMP3 construct used to visualize calcium activity, and Dr. Daniel Dombeck (Northwestern University) and Dr. Peter Toth (University of Illinois Chicago) for technical support and useful discussion regarding two-photon microscopy. This manuscript is dedicated to Dr. Akira Yoshii, who sadly passed away during its preparation. Akira was a beloved colleague to many, including the current authors, for whom honoring his legacy is paramount.
The authors declare no competing financial interests.
- Correspondence should be addressed to Kevin P. Koster at kpkoster{at}uchicago.edu
