Potential glutamatergic and GABAergic false neurotransmitters in models of excitation-inhibition imbalance

Abstract

We previously determined that 2-Methylglutamate (2MeGlu) and 4-aminopentanoic acid (4APA) have neurochemical properties of glutamatergic and GABAergic false neurotransmitters (FNTs); here, we tested whether their activity impacts mouse models with excitation-inhibition (E-I) imbalance. We first screened racemic agents using models caused by E-I imbalance; rac-2MeGlu, but not rac-4APA, suppressed 81% of excitotoxicity in hippocampal slices and increased survival by 105% in Aldh5a1^−/− mice. Enantiomers with the least receptor activity were further tested in more complex models. R-4APA (50 mg/kg) worsened startle behaviors in the Shank3^−/− autism model while S-2MeGlu (50 mg/kg/d over 19 days) improved motor performance by 77% in MPTP-treated mice without changing dopaminergic neurotoxicity; neither agent improved motor function in a human α-synuclein overexpressing mouse. S-2MeGlu (10 mg/kg/d for 8 weeks), but not R-4APA, reversed the spatial working memory deficit in T41 (Thy-1 hAPP^Lond/Swe+ ) mice without significantly changing Aβ plaque density. Single-nucleus transcriptomics following the same chronic exposures in WT mice yielded positively enriched pathways related to protein handling and synaptic regulation in excitatory neurons with S-2MeGlu; R-4APA caused metabolic pathway negative enrichments in multiple cell types. Our data reveal distinct behavioral and transcriptomic impacts of S-2MeGlu and R-4APA and further support S-2MeGlu as a glutamatergic FNT.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-54218-4.

Introduction

Excitation-inhibition (E-I) imbalance occurs in many diseases of brain, including autism, Parkinson’s disease (PD), and Alzheimer’s disease (AD)^1–4. Indeed, multiple drug development programs pursue novel agents that interact with specific glutamatergic or GABAergic receptors with the goal of E-I rebalancing^5–7. However, targeting specific receptors has potential limitations including dozens of potential targets with highly adaptable expression that can lead to diminished impact over time^8–10.

We have pursued an alternative approach of modulating presynaptic endogenous neurotransmitters with alpha carbon methyl-substituted analogues^11,12, based on earlier reports of this modification to monoamines yielding false neurotransmitters (FNTs) that became highly successful drugs and neuroimaging agents^13–15. In our previous work^11,12, we expanded this approach with alpha methyl-substituted analogues of glutamate (2-methylglutamate or 2MeGlu) and GABA (4-aminopentanoic acid or 4APA). We found that 2MeGlu and 4APA are transported into brain, neurons, and synaptosomes where they displace the corresponding endogenous neurotransmitter and are released upon depolarization with kinetics equivalent to endogenous neurotransmitters. Interestingly, when assayed against a panel of over 30 glutamate and GABA receptors the S enantiomer of 2MeGlu (S-2MeGlu) has no detectable agonist, antagonist, or modulator activity; R-2MeGlu is a weak antagonist of GluNR2, S-4APA is a weak agonist (GABA_A α4β3δ, GABA_A α5β2γ2, and GABA_B B1/B2) or antagonist (GABA_A α6β2γ2) of multiple GABA receptors, and R-4APA is a weak agonist of GABA_A α5β2γ2^11,12. In aggregate, these neurochemical features support S-2MeGlu as a potential glutamatergic FNT and R-4APA as a potential partial GABAergic FNT.

Our previous in vivo work with these potential FNTs was performed exclusively in wild type (WT) mice^11,12. A single 100 mg/kg intraperitoneal (IP) injection of 2MeGlu or 4APA into WT young adult mice had plasma half-lives between 20 and 30 min and maximum brain concentration of approximately 250 pmol/mg protein for each without detected toxicity^11,12. Despite these encouraging pharmacokinetic and neurochemical features, the same dose of each compound had no effect on a broad battery of sensorimotor behaviors in WT mice^11,12.

Here we tested the hypothesis that ex vivo and in vivo activity of 2MeGlu and 4APA might be revealed under stressful conditions that involve E-I imbalance. First, we screened racemic (rac)−2MeGlu or rac-4APA using ex vivo and in vivo models caused by E-I imbalance. Next, we focused on the more robust potential FNT enantiomers, S-2MeGlu or R-4APA, in widely used mouse models of neurologic diseases characterized by E-I imbalance and so have existing data on glutamate or GABA receptor antagonists for comparison. Finally, we sought insight into brain cell types and pathways impacted by S-2MeGlu or R-4APA via single nucleus transcriptomics.

Results

Screening experiments in disease models caused by E-I imbalance

Our initial screen for concentration-response relationship (not shown) was performed using hippocampal slice cultures exposed to vehicle (Veh) or rac-2MeGlu for 24 h at which time kainic acid (KA, 2 µM) or vehicle (Veh) was added for an additional 24 h and then assayed for injury to pyramidal neurons using propidium iodide (PI) uptake. Suppression of KA-induced excitotoxicity by rac-2MeGlu had significant concentration-response (P = 0.0163, simple linear regression of 11 data points including Veh and 10 concentrations from 5 nM to 50 µM). One-way ANOVA of the same data was significant (P = 0.0037) with Dunnett’s multiple comparison test vs. Veh significant for 5 µM (P = 0.0286), 25 µM (P = 0.0134), and 50 µM (P = 0.0471) µM rac-2MeGlu. Based on these results, our next set of experiments with hippocampal slice cultures followed the exact same methods but now exposed to Veh, 5 µM rac-2MeGlu, 5 µM rac-4APA, or 10 µM memantine for 24 h prior to addition of KA (Fig. 1a; n = 6 to 15 cultures per group). Memantine is a NMDAR non-competitive antagonist that suppresses neuronal injury in this model¹⁶. Importantly, neither enantiomer of 2MeGlu nor 4APA is an agonist or antagonist of the KA receptor^11,12. Exposure to each agent alone without KA did not increase PI signal compared to Veh (P > 0.9999 for each). KA alone increased hippocampal pyramidal neuron injury approximately 17-fold (P < 0.0001). The agents varyingly suppressed pyramidal neuron injury compared to KA alone: 81% for rac-2MeGlu (P < 0.001) and 88% for memantine (P < 0.0001) but no significant effect by rac-4APA; suppression of KA-induced PI signal was not significantly different between memantine and rac-2MeGlu (P = 0.0962). These results showed rac-2MeGlu’s concentration-response for suppressing KA-induced excitotoxicity had comparable maximal neuroprotection as memantine in this ex vivo model.

Fig. 1.

Ex vivo and in vivo screens using rac-2MeGlu and rac-4APA in models caused by E-I imbalance. (a) Exposure of wild type (WT) C57Bl/6 mouse hippocampal slice cultures to kainic acid (KA, 2 µM), drug (5 or 10 µM), or vehicle (Veh) for 24 h markedly increased propidium iodide (PI) uptake in pyramidal layer neurons by KA (P < 0.0001) but none of the drugs alone when compared to Veh (n = 6 to 15 cultures per group). Exposure to KA plus one of the drugs suppressed pyramidal neuron injury at 24 h by 88% for memantine (****P < 0.0001) and 81% for rac-2MeGlu (***P < 0.001); exposure to KA and rac-4APA did not significantly change PI signal. Suppression of KA-induced injury was not significantly different between memantine and rac-2MeGlu but was between rac-4APA and rac-2MeGlu (****P < 0.0001). (b) Probability of survival of WT (Aldh5a1^+/+, top left), heterozygous (Het, Aldh5a1^+/−, bottom left), and knockout (KO, Aldh5a1^−/−, right) mice over 45 days. WT (n = 11), Het (n = 23), and KO (n = 19), both male (n = 26) and female (n = 27), were treated with Veh (n = 3, 7, and 6) or 50 mg/kg IP rac-2MeGlu (n = 3, 7, and 7) or rac-4APA (n = 5, 9, and 6) every other day. KO median survival was 21 days but Het and WT mice were > 46 days (P < 0.0001). rac-4APA did not significantly change median survival of KO mice. Median survival of KO mice treated with rac-2MeGlu was 43 days, significantly longer than KO mice treated with Veh (P < 0.005).

Mice genetically lacking succinate semialdehyde dehydrogenase (SSADH), Aldh5a1^−/− (KO) mice have profound perturbation of both GABAergic and glutamatergic neurotransmission^17,18. The consequences of this neurometabolic disorder are complex; Aldh5a1 KO mice fail to thrive yet have normal brain weight with onset of seizures soon after weaning that progress to status epilepticus and death within weeks (Supplementary Fig. 1)¹⁹. We screened for status epilepticus-free survival among 62 mice: WT (n = 14), Aldh5a1^+/− (Het) (n = 30), and Aldh5a1 KO (n = 18) injected IP with Veh (n = 16) or 50 mg/kg of rac-2MeGlu (n = 17) or rac-4APA (n = 29) every other day starting 3 days after birth. Seizures began post weaning in Aldh5a1 KO mice as reported¹⁹. Four Aldh5a1 KO/rac-2MeGlu mice were euthanized at day 43 because of onset of status epilepticus (6 of 6 Racine score for each); no Het or WT mouse regardless of treatment group had seizures (0 of 6 Racine score for all) and only one died (Het injected with rac-4APA) prior to termination of experiment on day 46. Survival analysis for these 62 mice had P < 0.0001 (Fig. 1b), with no significant difference when stratified by sex within Aldh5a1 KO mice (P = 0.1138) and no sex-based differences in survival for any group in response to rac-2MeGlu or rac-4APA. Median survival for Aldh5a1 KO/Veh mice was 21 days, replicating the work of others¹⁹; median survival of Aldh5a1 KO/rac-4APA mice also was 21 days. rac-2MeGlu approximately doubled median survival of Aldh5a1 KO mice to 43 days (P = 0.0092 compared to Aldh5a1 KO/Veh) without improvement in body weight, aligning with reports that used vigabatrin and other anti-epileptic drugs (AEDs) in this model^19–21.

In vivo mouse models of diseases characterized by E-I imbalance

We leveraged the extensive experience of contract research organizations and Stanford University core facilities that each used well-established, standardized protocols to insure robust data collection across a survey of multiple in vivo disease models. All investigators were blinded to agents’ identity and followed pre-specified primary endpoints and data analyses (Fig. 2); secondary endpoints and supportive data are shown in the supplementary figures. Finally, one of the disease models used, Line 61, has transgene insertion on the X chromosome so males exhibit more consistent phenotypes²²; for this reason, we focused on male mice in Line 61. Since one major goal was comparability with other compounds tested by the CROs’ standardized models (Shank 3^−/− and MPTP models) whose behavioral profiles are on data generated in male mice, we extended the use of male mice to the T41 model tested at Stanford to permit coherent comparison among our four behavioral models, recognizing that this will require use of both sexes in follow up studies focused on areas for potential translation.

Fig. 2.

In vivo mouse models of diseases characterized by E-I imbalance. All investigators were blinded to coded treatments and mice were evaluated by pre-specified behavioral tests and statistical analyses. (a) Shank3^−/− (KO) mouse model of autism: Four groups of 10 week-old male mice (n = 16 mice per group) were evaluated over 4 weeks (n = 16 mice per group): WT mice treated IP with vehicle (Veh) and Shank3 KO mice treated IP with Veh, 50 mg/kg S-2MeGlu, or 50 mg/kg R-4APA on the day of testing. Data are presented as mean ± SEM. Two-way ANOVA had dB Intensity P < 0.0001, Genotype/Treatment (P < 0.0001), and interaction P < 0.0001). Fisher’s post hoc test showed that startle response was decreased in Shank3 KO mice compared to WT at > 100 dB (**P < 0.01, ****P < 0.0001), with a significant further reduction in Shank3 KO/R-4APA mice compared to Shank3 KO/Veh mice at 110 to 120 dB (**P < 0.01). Non-significant paired comparisons are not indicated. (b) MPTP model in mice: Six groups of 9 week-old male C57Bl/6 mice (n = 13 to 15 per group) were exposed to Veh or 40 mg/kg/d MPTP IP on days 1 and 2. From days − 5 to day 14 mice were treated IP with Veh or S-2MeGlu or R-4APA at 5 or 50 mg/kg/d and then underwent automated assessment of 97 gait and balance parameters. The overall kinematic score, a composite index of the 97 parameters, was the pre-specified primary endpoint. One way ANOVA of overall kinematic score had P = 0.0056. Fisher’s post-hoc test compared MTPT/Veh group with each other group and was significant vs. Veh/Veh (**P = 0.0048) and vs. MPTP/S-2MeGlu at 50 mg/kg (*P = 0.0362). (c) T41 mouse model of Alzheimer’s disease cerebral Aβ amyloidosis. Male 16 week-old mice comprised six groups (n = 10 to 28 mice per group): WT mice with intraperitoneal (IP) injection of Veh and T41 mice with IP injection of Veh, 10 mg/kg of S-2MeGlu or R-4APA for 8 weeks. Y Maze data were analyzed using one sample t test with theoretical mean set to 50% (**P < 0.01, *P < 0.05).

Shank3^−/− (KO) mice are a model of autism spectrum disorder based on Phelan McDermid syndrome, characterized by synaptic dysfunction and abnormal behaviors consistent with altered neuronal circuit function and excitatory–inhibitory imbalance^23,24. A standardized behavioral test battery (Table 1) was performed over four weeks using four groups of mice (n = 16 per group): WT with Veh and Shank3 KO mice with Veh or 50 mg/kg of S-2MeGlu or R-4APA. Mice were injected IP once per week on the day of testing starting at 10 weeks of age. As expected, Shank3 KO mice (25.8 ± 1.1 g) were heavier than WT mice (23.9 ± 1.5 g) at baseline (P < 0.001)²⁴. Three behavioral tests showed no significant effect from Shank3 KO: total distance traveled in the urine open field test, number of grooming bouts, and number of sniffing events; no mouse displayed clasping behavior (not shown). Two behavioral tests showed a significant change in Shank3 KO/Veh compared to WT/Veh mice: increased duration of grooming (P = 0.0074) and decreased number of total body contacts (P = 0.0083); however, neither was significantly changed by treatment with S-2MeGlu or R-4APA (Supplementary Fig. 2). Finally, startle response to increasingly louder tones not only showed a significant reduction in Shank3 KO/Veh compared to WT/Veh (P = 0.0030 at 100 dB and P < 0.0001 at > 100 dB) but also a significant further suppression by R-4APA at 110 dB (P = 0.0056), 115 dB (P = 0.0032), and 120 dB (P = 0.0063); S-2MeGlu had no significant effect on startle response (Fig. 2a). These results show that a single IP dose of 50 mg/kg R-4APA further suppressed startle response, an unconditional reflex commonly used as a measure of habituation, sensitization, and anxiety²⁵, in this mouse model of a form of autism. This effect of R-4APA is similar to GABA_A agonists in this model²⁶, raising the possibility that R-4APA’s residual GABA_A agonist activity is responsible for the observed effect.

Table 1.

Mouse behavioral test information.

Model/tests	Description
Shank3 knockout (Psychogenics)	Autism spectrum disorder model characterized by synaptic dysfunction and abnormal behaviors consistent with altered neuronal circuit function and E-I imbalance. Responsive to: GABA agonists26.
clasping	Assesses muscular strength in limbs. Mice are held by the tail and gently lifted until the front paws just lift off the counter surface. The legs are observed to identify clasping or splay of limbs. Percent (per group) that show clasping is calculated.
grooming duration and bouts	Assessed in clean cages under red light conditions during the dark cycle. Mice are individually housed prior to testing. Grooming duration and bout number are quantified.
sniffing events	Subject and stimulus mice are introduced in a new arena; behavior is recorded for 10 min. Number and duration of sniffing interactions and number of body contacts are quantified.
startle response	Acoustic startle measures an unconditioned reflex response to auditory stimuli. Mice are placed in sound-attenuated startle chambers on a force transducer and habituated for 5 min with 70 dB white noise. Test sessions consist of 10 blocks of 11 trials, with stimuli ranging from 70–120 dB presented in random order (40 ms duration, 10–20 s inter-trial interval, mean 15 s). Responses are recorded for 150 ms and sampled every ms.
urine open field	Assesses recognition of a socially relevant olfactory stimulus and social communication. Male mice are pre-exposed to females and tested in a dimly lit open field with home cage bedding. On test day, following a 60 min habituation, female urine is applied to the arena and mice are reintroduced for 5 min. Total distance traveled is automatically recorded.
MPTP model (Charles River)	MPTP-induced Parkinson’s disease model characterized by loss of dopaminergic neurons and associated dyskinetic motor impairments. Responsive to: Metabotrophic glutamate receptor agonists30,31.
kinematic score	Assessed using a high-precision kinematic gait analysis platform. Mice traverse a corridor during voluntary locomotion, and movement is recorded with multi-angle high-speed cameras. Full-body tracking is performed using anatomical landmarks across limbs, joints, trunk, and head. ~90–100 spatiotemporal, postural, and interlimb coordination parameters, including gait and balance, angle range hip, and support three are extracted across locomotor cycles and integrated by PCA to generate a composite kinematic score.
Line 61 (Psychogenics)	Parkinson’s disease model based on Thy1-driven overexpression of human α-synuclein32,35, characterized by synaptic dysfunction and neuronal circuit alterations34.
tapered beam	The tapered beam test assesses fine motor coordination and early motor deficits. Mice traverse an inclined, narrowing beam toward a goal box, with performance recorded by video (3 trials per mouse, ≥ 30 s inter-trial interval). Animals are trained before testing. Primary endpoints include latency to turn, latency to traverse (max 120 s), and step-slip ratio, with foot slips quantified during traversal.
T41 (Stanford BFNL*)	Model of early-stage AD based on Thy1-driven overexpression of mutant human APP42, characterized by synaptic dysfunction and glutamate-dependent neuronal alterations38. Responsive to: memantine
Y-maze spontaneous alternation	Spatial working memory is assessed using the Y-maze spontaneous alternation task, which measures the tendency to explore a novel arm without prior training or reward. Mice are placed in the center of a 3-arm maze and allowed to explore freely for 5 min, with arm entries recorded by overhead camera. Spontaneous alternation is defined as consecutive entries into 3 different arms (e.g., ABC, BCA). The 1 st entry is excluded, and alternations are identified across overlapping triplets of consecutive entries. Alternation (%) is calculated as (# number of alternations/[total entries − 2]) × 100 and compared to the 50% chance level, which reflects random arm selection.

*Behavioral Functional Neuroscience Laboratory.

1-Methyl-4-phenyltetrahydropyridine (MPTP) is a selective dopaminergic neurotoxin that leads to secondary striatal E-I imbalance that then contributes to dyskinesias characteristic of PD^27,28. Six groups of WT C57Bl/6 mice (n = 13 to 15 per group, 9 weeks of age) were injected IP with Veh or MPTP (40 mg/kg on days 6 and 7) during a 19 day period when mice were injected IP daily with Veh or with S-2MeGlu or R-4APA at two doses (5 mg/kg or 50 mg/kg); mice then underwent automated assessment of 97 gait and balance parameters (Supplementary Fig. 3)²⁹. The pre-specified primary endpoint, a composite of these 97 measures called overall kinematic score, increased from 0.00 in Veh/Veh to 3.74 in MPTP/Veh mice (P = 0.0048) and was reversed to 0.85 in MPTP/50 mg/kg S-2MeGlu mice (P = 0.0362, Fig. 2b); lower dose of S-2MeGlu and neither dose of R-4APA significantly impacted the overall kinematic score. Secondary analysis using two-way ANOVA had P < 0.0001 for all 97 gait parameters, P = 0.0412 for group, and P < 0.0001 for interaction. Post hoc multiple comparisons highlighted four individual gait and balance parameters significantly impaired by MPTP (Veh/Veh vs. MPTP/Veh) and significantly improved by 50 mg/kg S-2MeGlu (MPTP/50 mg/kg S-2MeGlu vs. MPTP/Veh): Angle Range Hip (P < 0.0001 and P < 0.0001), Angle Range Ankle (P < 0.0001 and P < 0.001), Support Diagonal (P = 0.0062 and P < 0.0397), and Support Three (P = 0.0032 and P = 0.0037, Supplementary Fig. 3). Two of these secondary analyses (Angle Range Hip and Angle Range Ankle) also showed significant benefit from lower dose S-2MeGlu (P = 0.0078 and P < 0.0001) and higher dose R-4APA (P < 0.0001 and P < 0.001); however, lower dose R-4APA had no significant impact on any individual gait and balance measure. Importantly, all groups of MPTP treated mice had similarly reduced striatal concentrations of dopamine, DOPAC, and HVA (P < 0.0001 for each) demonstrating that neither S-2MeGlu nor R-4APA significantly impeded neurotoxicity from MPTP (Supplementary Fig. 3). These results support significant suppression of motor deficits following MPTP dopaminergic injury by high dose S-2MeGlu, a proposed glutamatergic FNT, and align well with similar outcomes with metabotropic glutamate receptor antagonists in this model^30,31.

Line 61 transgenic mice overexpress human α-synuclein under control of the mouse Thy-1 promoter as a synucleinopathy model of PD^32,33 characterized by synaptic dysfunction and neuronal circuit alterations³⁴. At 15 weeks of age, Line 61 mice (n = 13) showed significantly impaired performance on the tapered beam compared to untreated WT mice (n = 13): latency to turn (8.1 ± 2.1 s in WT vs. 38.2 ± 8.5 s in Line 61 mice, P = 0.0043) and slip-step ratio (0.10 ± 0.05 in WT vs. 0.33 ± 0.04 in Line 61 mice, P = 0.0028). Five groups of Line 61 mice (16 weeks old, n = 11 to 14 per group) underwent the same motor evaluation following IP injection with Veh or with R-4APA or S-2MeGlu (5 or 50 mg/kg/d) for 7 days³⁵. Neither latency to turn nor slip-step ratio was significantly changed by exposure to either compound at either dose (not shown). Immunohistochemistry (IHC) detected human α-synuclein in all Line 61 mouse brains. Percent immunoreactive hippocampal neurons was 52.1 ± 5.5% in Line 61/Veh similar to others’ results^32,33, and was not significantly different from Line 61/R-4APA mice or Line 61/S-2MeGlu mice at either dose (n = 10 to 14 in each group, P = 0.7007).

Transgenic mice that overexpress human mutant APP, also under control of the mouse Thy1 promoter, accumulate cerebral Aβ and exhibit synaptic dysfunction and glutamate-dependent neuronal alterations. This widely used approach models an early stage of the AD continuum that is likely amenable to rescue^36,37. Importantly, AD derives in part from Aβ oligomer activation of multiple glutamate receptors and is approved for treatment with memantine^36,37. Given that rac-2MeGlu had comparable efficacy as memantine in hippocampal slice cultures (vide supra), here we selected an exposure to S-2MeGlu comparable to the effective dose of memantine in a related transgenic mouse model of human mutant APP overexpression³⁸. Six groups of 16 week-old WT or T41 mice were injected IP with Veh or 10 mg/kg/d S-2MeGlu or R-4APA daily for 8 weeks (10 to 28 mice per group) and then assessed by a standardized behavioral test battery^39–41. As part of this behavioral battery, spatial working memory was assessed using the Y-maze spontaneous alternation task, in which alternation (%) reflects consecutive entries into three different arms and depends on the ability to remember recently visited locations; performance is compared to a 50% chance level, as random arm selection yields correct alternations in approximately half of cases. As expected, percent alternation for WT/Veh mice was significantly greater than chance (**P < 0.0085) and, consistent with our previous results that showed no behavioral impact of higher doses^11,12, remained significantly different than chance in WT mice injected with S-2MeGlu (**P = 0.0040) or R-4APA (*P < 0.0378, Fig. 2c). In contrast, percent alternation by T41/Veh mice was not different from chance (P = 0.1756), replicating their well-described spatial working memory deficit^42–44. Spatial working memory deficit in T41/S-2MeGlu mice was significantly different from chance (**P = 0.0021 by one-sample t test) with significant restoration compared to untreated (T41/Veh vs. T41/S-2MeGlu P = 0.0232 by Welch’s t-test) and not significantly different than untreated WT littermates (T41/S-2MeGlu vs. WT/Veh, P = 0.4587 by Welch’s t-test) but not R-4APA (P = 0.2380). IHC for Aβ plaques had significantly greater percent area of hippocampus occupied by chromogen product in T41/Veh mice (n = 3) compared to WT/Veh mice (n = 3, P < 0.0001) but, although trending lower, was not when significantly changed in T41/S-2MeGlu mice (n = 4, P = 0.0511, Supplementary Fig. 4)^45,46. These data support low dose, chronic S-2MeGlu to promote resilience to the spatial working memory deficit caused by mutant human APP overexpression and Aβ accumulation in mouse cerebrum.

Cell-type–specific transcriptional responses in cerebral cortex to S-2MeGlu or R-4APA exposure

We performed single-nucleus (sn) RNA-seq using murine frontal cerebral cortex from WT mice injected IP with S-2MeGlu or R-4APA at 10 mg/kg/d for eight weeks to mirror the lowest concentration that was effective in our survey of mouse models (Fig. 3). We selected WT mice to isolate transcriptomic changes of potential FNTs from the confounding effects of neurotoxin, gene ablation, or transgene expression that were necessary to create the disease models. Integration and clustering of nuclei identified major brain cell classes, including excitatory and inhibitory neurons, astrocytes, oligodendrocytes, and microglia, with consistent representation across treatment conditions (Fig. 3a). Differential gene expression analysis across major cell classes revealed distinct patterns of transcriptional impact by the two compounds (Fig. 3b). R-4APA caused detectable transcriptional changes in multiple neuronal and glial populations, while the transcriptional response to S-2MeGlu was largely restricted to excitatory neurons. Within excitatory neurons, R-4APA presented a substantially broader set of differentially expressed genes than S-2MeGlu.

Fig. 3.

Cell-type–specific transcriptional responses to S-2MeGlu and R-4APA in mouse cortex. (a) UMAP representation of integrated single-nucleus transcriptomes from mouse cerebral cortex across all treatment conditions, comprising a total of 39,770 nuclei that are colored by major cell class: excitatory neurons (Exc), inhibitory neurons (Inhib), astrocytes (Astro), oligodendrocytes (Oligo), and microglia (Micro). The integrated UMAP also is shown stratified by exposure on the left; annotations on the right illustrate the subcluster structure within each major cell class. (b) Number of significantly differentially expressed genes (Log2FC > 0.25; adjusted p-value < 0.05) detected in each major cell class for S-2MeGlu versus vehicle (solid bar) and R-4APA versus vehicle (striped bar). (c) Immediate early gene (IEG) module scores across major cell classes for vehicle, S-2MeGlu, and R-4APA exposed mice. Box plots show distributions of per-nucleus IEG module scores within each class. Significance denotes Wilcoxon rank-sum tests versus vehicle with Benjamini–Hochberg correction within each class (***adjusted p-value < 0.001; ns, not significant). (d) Gene set enrichment analyses (FGSEA) of excitatory neurons using Gene Ontology Biological Process terms. Plots show normalized enrichment score (NES) versus -log₁₀(adjusted P) for S-2MeGlu versus vehicle exposure (top) and R-4APA versus vehicle exposure (bottom). Each point represents a pathway, with point size proportional to the number of leading-edge genes. The top 5 significantly enriched pathways are annotated; grey points indicate pathways with adjusted P > 0.05.

Inspection of significant differentially expressed genes in excitatory neurons revealed a prominent representation of immediate early genes (IEGs) (Supplementary Table 1). Immediate early genes are rapidly and transiently induced in neurons in response to changes in synaptic activity and are widely used as markers of activity-dependent transcriptional responses⁴⁷. Based on this observation, we quantified IEG module scores across cell types to assess activity-dependent transcriptional responses to both compounds (Fig. 3c). S-2MeGlu and R-4APA significantly increased IEG scores in excitatory neurons relative to vehicle. In contrast, only R-4APA exposure, but not S-2MeGlu, significantly increase in IEG scores in inhibitory neurons; no significant change in IEG module scores were detected in glial populations with exposure to either potential FNT.

We next examined pathway-level changes for all major neuronal and glial classes using gene set enrichment analysis (FGSEA) to contextualize the transcriptional impact of each compound. Following R-4APA treatment, all major cell classes had treatment-dependent gene set enrichment with consistently negative enrichment, while S-2MeGlu showed more limited enrichment in just excitatory neurons and astrocytes (Supplementary Fig. 5). In excitatory neurons, S-2MeGlu exposure resulted in significantly enriched pathways, predominantly positive enrichment, and primarily associated with protein handling and synaptic regulation, including protein transport, protein folding, synapse structure, and synapse activity; these changes are consistent with S-2MeGlu exposure impacting a focused set of activity-related processes in excitatory neurons (Fig. 3d, top). In contrast, R-4APA exposure yielded a much broader pattern of pathway enrichment with the majority of significant pathways exhibiting negative enrichment (Fig. 3d, bottom); these were largely related to mitochondrial and metabolic functions, including oxidative phosphorylation, aerobic respiration, cellular respiration, and cytoplasmic translation, suggesting a transcriptional shift affecting core cellular metabolic programs.

Discussion

E-I imbalance is a feature in a broad range of neurologic diseases^1–4. Most have approached the challenge of E-I rebalancing through development of agents that target specific receptors. We previously used an alternative approach to develop presynaptic neurotransmitters with neurochemical features of FNTs, 2MeGlu for glutamate and 4APA for GABA^11,12. Given that we found both compounds gain ready access to brain and synaptic vesicles, it was somewhat surprising that neither altered WT mouse behavior at doses up to 100 mg/kg. However, the preliminary safety of these high doses encouraged us to study them further and test the hypothesis that disease models that involve E-I imbalance might provide the context to unveil neurobiological activity of these potential FNTs.

Our initial experiments were designed to screen rac-2MeGlu and rac-4APA for activity in ex vivo and in vivo models caused by E-I imbalance. While we predicted that a glutamatergic FNT would mimic a NMDA receptor antagonist in the ex vivo model, we did not have a strong expectation for the SSADH deficient mice because of their complex neurochemical changes in both glutamatergic and GABAergic systems¹⁷. In fact, in hippocampal slice cultures, rac-2MeGlu equivalently suppressed excitotoxicity as memantine, a widely prescribed NMDA receptor antagonist for multiple neurologic diseases including AD⁴⁸; rac-4APA did not suppress neuron injury in this model. In SSADH deficient mice, rac-2MeGlu (50 mg/kg) approximately doubled survival, similar to vigabatrin and other anti-epileptic drugs (AEDs)^19–21. In contrast, rac-4APA (50 mg/kg) was ineffective in this model. Together these screening experiments in models caused by E-I imbalance revealed fundamentally different neurobiological activity of these potential FNTs and encouraged more extensive testing of 2MeGlu that targets glutamatergic synapses and its decarboxylated analogue, 4APA, that targets GABAergic synapses.

We next investigated four widely used models of more prevalent neurologic diseases associated with E-I imbalance that are shown to respond to antagonists of glutamate or GABA receptors. We deliberately selected standardized protocols, including duration of exposure, to ensure robust data across multiple disease models and to facilitate direct comparison to published work with other drugs. Further, we revised down the IP doses used as we progressed through behavioral models in an attempt to survey a range of effective doses for each FNT candidate realizing that this limits direct comparison across our models. Our first two models were the Shank3 KO model of autism spectrum disorder and the MPTP model of the striatal dopaminergic degeneration characteristic of PD. The Shank3 KO model showed characteristic reduction in startle response^49,50, that was further suppressed by a single 50 mg/kg dose of R-4APA, like GABA_A receptor agonists²⁶, highlighting R-4APA’s retained GABA_A α5β2γ2 agonist activity and undermining its potential utility as a FNT. Extensive anatomic and pharmacologic data support that gait deficits after striatal dopaminergic afferent degeneration derive in part from dysregulated excitatory input to inhibitory medium spiny neuron dendrites and are the rationale for amantadine, the only FDA-approved glutamate antagonist for treating dyskinesias in people with PD^51,52. Consonant with these data, only the higher dose of S-2MeGlu (50 mg/kg/d x 19 d) reversed the overall kinematic score (pre-specified primary endpoint) by 77% with no significant effect by either dose of R-4APA. Secondary analyses suggested beneficial effect also from the lower dose (5 mg/kg/d) of S-2MeGlu for some individual gate and balance parameters. Our results in these two models demonstrated a clear difference in activity between S-2MeGlu and R-4APA in models characterized by E-I imbalance, suggested a lower bound for effective concentration of 5 mg/kg/d S-2MeGlu, and raised concern about R-4APA as a GABA FNT given its retained GABA_A agonist activity.

We used two models of neurodegenerative proteinopathy that have been widely employed in drug testing even if no longer considered the most advanced models of disease pathogenesis in humans. In addition to many comparator therapeutics, we also selected these mouse models because both overexpress human protein in mouse brain under control of the mouse Thy1 promoter: human α-synuclein in Line 61 mice and human mutant amyloid precursor protein in T41 mice. Line 61 mice had the expected pathological pattern of human α-synuclein accumulation in brain but their gait deficits were unresponsive to S-2MeGlu or R-4APA at the same 5 or 50 mg/kg IP doses used in the MPTP model, although with a shorter exposure duration. This is perhaps unsurprising since human α-synuclein aggregates in cell culture and in mice exert toxicity through incompletely understood mechanisms that involve prion-like properties that disrupt mitochondrial function, autophagy, and immune response without strong evidence for contribution by activation of glutamate receptors⁵³. In sharp contrast, Aβ overexpression leads to formation of oligomers that activate multiple ionotropic and metabotropic glutamate receptors, memantine is an approved treatment for people with AD, and multiple glutamate receptor antagonists as well as riluzole are under development for treatment of people with AD^54–56. The spatial working memory in T41 mice was restored only by S-2MeGlu (10 mg/kg/d over 8 weeks), approximating the lower effective dose from our MPTP experiment and validating results from extended exposure to memantine at the same dose in a related mouse model of AD³⁸. S-2MeGlu’s beneficial effect on memory did not alter Aβ plaque accumulation, again validating mematine results for mice at this age, and mirroring clinical and neuroimaging data from memantine-treated people with AD who show improved memory performance without reduction in cerebral Ab by PET imaging^38,57.

We determined the single-nucleus cerebral cortical transcriptomic changes by S-2MeGlu and R-4APA exposure in WT mice following the lowest dose regimen that had demonstrable behavioral effect to gain insight into molecular and pathway impacts in the absence of confounding by disease models. Following two months exposure, our results likely are the combination of direct effects and adaptive responses. S-2MeGlu exposure had a highly focused transcriptomic impact in excitatory neurons with significant positive enrichment of pathways involving protein handling and synaptic regulation, including IEGs. In contrast, exposure to R-4APA had much broader cellular impact that included excitatory neurons but also most prominently inhibitory neurons as well as glia subtypes with significant negative enrichment of core metabolic pathways. While these transcriptional results undermine confidence in R-4APA’s specificity for GABAergic neurons, they support focused impact of S-2MeGlu on excitatory neurons’ activity-dependent pathways over 8 weeks’ exposure and provide one measure of selective targeting of excitatory neurons by this potential glutamatergic FNT.

Our study has limitations. The disease models selected have well established pre-clinical testing protocols thereby facilitating comparison of our results with others’ who used glutamate or GABA receptor antagonists; however, the fidelity of these models to the human condition varies. Indeed, we selected the T41 transgenic mouse as a model of the earliest stages of AD for therapeutic intervention recognizing that it does not reflect changes in pathologic tau that occur later in the course of disease, similar to the approach of others⁵⁸. As noted, we deliberately shifted to lower doses but with varying duration of exposure dictated by the model’s standardized protocol to facilitate comparison with others’ results, accepting that our survey approach limits direct comparison across models. We observed a significant effect of S-2MeGlu at 5 mg/kg/d x 19 d in secondary endpoints in the MPTP model and at 10 mg/kg/d x 8 weeks in primary endpoints in T41 mice; however, the lowest effective dose and precise dose-response relationship will need to be refined in future work using models focused on the intended indication. S-2MeGlu is metabolically restricted by design and has a much longer half-life in brain than R-4APA, which is rapidly converted to multiple metabolites by as yet unclear mechanisms^11,12. We speculate that the retained GABA_A agonist activity and extensive metabolism of R-4APA may contribute to its much broader transcriptomic changes than S-2MeGlu. It is critical to note that E/I balance, glutamatergic receptor expression, and GABAergic interneuron composition varies among brain regions and so the translational relevance of hippocampal slice screening data to the cortically-focused transcriptomic analyses remains to be established. For biological reasons in one behavioral model and standardization of two other CRO models, our behavioral experiments were restricted to male mice. We think this is a justified limitation for a survey study that prioritized comparability with existing results; however, future work that focuses more deeply on specific disease models will need to include mice of both sexes. Finally, while we previously demonstrated that S-2MeGlu and R-4APA have neurochemical properties of FNTs, further work, including electrophysiologic measures, is needed to more fully understand the mechanisms of action of these potential FNTs in vivo.

Combined with our previous results^11,12, S-2MeGlu has neurochemical features of a glutamatergic FNT, focused transcriptomic impact on glutamatergic neurons in vivo, and activity in multiple mouse models with E-I imbalance that mirrors known actions of ionotropic and metabotropic glutamate receptor antagonists. In contrast, R-4APA, which retains GABA_A agonist activity in vitro, matched the known effects of some GABA_A agonists in vivo and had broad transcriptional impact on multiple brain cell types. Together, our results recommend S-2MeGlu as a potential glutamatergic FNT that is safe and effective in multiple mouse models with E-I imbalance.

Materials and methods

Compounds

As described previously^11,12, (R)−2-methylglutamic acid (R-2MeGlu) and (S)−2-methylglutamic acid (S-2MeGlu) were made to order by Concept Life Sciences (Cheshire, UK) from racemate with > 98% purity. Both racemic and (R)−4-aminopentanoic acid (R-4APA, cat. no. BBV-38374677, optical rotation value + 9.2°) were obtained from Enamine (Kyiv, Ukraine), also as described previously^14,15. Doses used in the current study were informed by previously performed pharmacokinetic experiments^11,12.

Immunohistochemistry (IHC) was performed on formalin-fixed paraffin-embedded tissue blocks using antibodies and protocols previously published by our group^45,46. These included (i) anti-human pS129- α-synuclein (Abcam, Cambridge, UK) and (ii) 6E10 (anti-human Aβ (Thermo Fisher, Fremont, CA). Images were analyzed with ImageJ (Fiji).

Animals

All experimental mouse protocols performed at Stanford University were approved by the Stanford University Administrative Panel on Laboratory Animal Care (APLAC; protocols #31890 #18466) and were carried out in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. All efforts were made to minimize animal suffering and to reduce the number of animals used. All mice were maintained on a 12/12 light/dark cycle at room temperature (20 to 23 °C) and relative humidity of ~ 50%. Food and water were provided ad libitum. C57BL/6 mice (20–25 g adult or 8–12 g post-weaning pups) reaching predefined humane endpoints, or at study termination, were euthanized by isoflurane inhalation overdose followed by a secondary physical method to ensure death, in accordance with APLAC-approved protocols. Experimental protocols for the in vivo studies conducted by Psychogenics, Inc. (Line 61 and Shank3 KO models) and Charles River Laboratories (MPTP model) were approved by their respective Institutional Animal Care and Use Committees (IACUCs) and carried out in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and applicable regulations.

Sex was selected according to experimental context and endpoint. Screening experiments in models caused by E–I imbalance (Aldh5a1 KO and hippocampal slice cultures) used mixed-sex cohorts or non–sex-specified preparations, as these assays rely on robust endpoints. Behavioral studies focuses on male mice because the Line 61 model carries an X-linked transgene that results in more consistent phenotypic expression in males²² and the standardized CRO testing in Shank3 KO and MPTP models also used male mice; based on this, we also used male T41 mice to enable comparison across our four behavioral models that are associated with E–I imbalance.

Hippocampal slice cultures

Hippocampi were collected from 5 to 6 day-old postnatal C57Bl/6 mouse pups (sex not recorded) following rapid decapitation with a sharp pair of scissors and sectioned at 250 μm using McIlwain tissue chopper^59,60. Hippocampal slices were placed on cell culture inserts using the air-medium interface method with culture media (50% MEM with Hank’s Salt and L-Gln Invitrogen 11575-032, 25% Hank’s Balance Salt Sigma H9269, 25% Horse Serum Invitrogen 26050-070, 25 mM HEPES Invitrogen 15630-080, 1% Penicillin/Streptomycin, 30mM or 5mM Glucose) in 5% CO2, 37 °C humidified incubator⁶¹. They were maintained in culture media plus 30 mM glucose for 10 days and then switched to culture medium plus 5 mM glucose for 3 days prior to experiments, which began on day 13 and used culture medium plus 5 mM glucose. Experiments lasted 48 h and began with 24 h exposure to 5 µM rac-2MeGlu, 5 µM rac-R4APA, or 10 µM memantine followed by the addition of 2 µM kainic acid (KA) or Veh for 24 h. Neuron injury following exposure was determined by propidium iodide (PI) uptake as detected by ImageXpress XLS (Molecular Devices) and quantified by MetaXpress¹¹.

Aldh5a1^+/- mouse survival

Aldh5a1^+/− mice were purchased from Jackson Laboratory and bred at Stanford to produce WT (Aldh5a1^+/+), heterozygous (Het, Aldh5a1^+/−), and knockout (KO, Aldh5a1^−/−) mice whose genotypes were determined by TransnetYX (Cordova, TN) using RT-PCR. Starting at 5 days of age, WT (n = 11), Het (n = 23), and KO (n = 19),, mice of both sexes (26:27 M: F) were injected every other day with Veh (n = 3, 7, and 6) or with 50 mg/kg rac-2MeGlu (n = 3, 7, and 7) or rac-4APA (n = 5, 9, and 6) by scientists in the Montine Lab blinded to test agent identity. Mice were weighed and evaluated by Racine seizure scale every other day and followed till death or sustained status epilepticus⁶². Mice were monitored daily for signs of distress including persistent seizures, impaired mobility, dehydration, or weight loss. Humane endpoints were predefined as sustained status epilepticus, weight loss exceeding 20% of baseline body weight, inability to access food or water, or a moribund condition. Animals reaching humane endpoints were euthanized immediately by CO₂ inhalation followed by cervical dislocation to ensure death. We observed expected differences in weight by sex and genotype (Supplementary Fig. 1); however, neither S-2MeGlu nor R-4APA exposure caused a significant change in body weight (not shown).

Four mouse lines were assessed for behavioral changes (Table 1):

Shank3^−/− mice (KO): With coded vials of S-2MeGlu and R-4APA provided by the Montine Lab, blinded scientists at PsychoGenics (Paramuis, NJ) injected IP male WT and Shank3 KO mice following pre-specified protocols, behavioral tests, and statistical analyses; all mice were bred and maintained at PsychoGenics. Previous survey data in WT mice showed no behavioral impact or toxicity but delivery to brain at 100 mg/kg of S-2MeGlu or R-4APA; here we selected one-half that dose for use in our first disease model. Shank3 KO mice (10 to 14 weeks old) were injected IP with Veh, 50 mg/kg S-2MeGlu, or 50 mg/kg R-4APA once per week on the day of behavioral testing; WT mice were injected with Veh. As expected, Shank3 KO mice (25.8 + 1.1 g) were heavier than WT mice (23.9 + 1.5 g) at baseline (P < 0.001)²⁴. However, all groups of WT and Shank3 KO gained weight at the same rate over the four-week experiment (not shown). There were no deaths during this experiment.

1-Methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) is a selective dopaminergic neurotoxin used widely to model dopaminergic degeneration in PD⁶³. With coded vials of S-2MeGlu and R-4APA provided by the Montine Lab, blinded scientists at Charles River Laboratories (CRL) following their 19 day standard protocol with pre-specified high-precision fine motor kinematic protocol and statistical analyses²⁹. Six groups of 9 week-old male C57Bl/6 mice (n = 13 to 15 per group) maintained at CRL were treated with Veh (group 1) or 40 mg/kg/d MPTP IP on days 6 and 7 (groups 2 to 6). Group 2 was injected IP from days 1 to day 19 with Veh while groups 3 to 6 were injected IP with two different coded test agents: 5 mg/kg/d or 50 mg/kg/d S-2MeGlu (groups 3 and 4) or 5 mg/kg/d or 50 mg/kg/d R-4APA (groups 5 and 6). The higher dose was selected to match the Shank3 KO mice and to explore possible activity in a 10-fold lower dose. One mouse in the MPTP/Veh group died. On day 19, CRL’s high-precision fine motor kinematic protocol quantified 97 individual gait and balance parameters and integrated them into an overall kinematic score; higher score indicates greater impairment. Mice were then sacrificed and striatum flash frozen for measurement of dopamine and its metabolites by HPLC with electrochemical detection.

Thy1-αSyn (Line 61) mice overexpress WT human α-synuclein under the mThy1 promotor to model neuronal accumulation of α-synuclein, a hallmark feature of PD^64–66. The Montine Lab provided coded vials of S-2MeGlu and R-4APA to scientists at PsychoGenics who followed pre-specified protocols, behavioral tests, and statistical analyses; five groups of 16 week-old male Line 61 mice were treated with Veh (n = 11) or either S-2MeGlu or R-4APA at 5 mg/kg/d or 50 mg/kg/d IP (n = 14 in each group) from day 1 to day 7. Dose was selected to match the MPTP mice. Motor function was assessed using the tapered beam test with video recording (3 trials per mouse) with pre-specified primary endpoint of latency to turn and step-slip ratio. Data from untreated 15 week-old WT and Line 61 mice validated the significant impact of transgene expression in these mice. There was no change in body weight or deaths over the 7-day experiment.

Thy 1-hAPP^Lond/Swe+ transgenic (T41) mice are a model of human Aβ accumulation, a hallmark feature of AD, used in preclinical testing for treatments for AD^42–44,67. Scientists at the Stanford Behavioral and Functional Neuroscience Laboratory (SBFNL) received coded vials with compound or Veh and then performed blinded, IP injections in 4 to 6 month-old facility-bred and -housed WT and T41 male mice following SBFNL’s pre-specified sequential behavioral testing protocol⁴². A dose of 10 mg/kg was selected based on the limited behavioral effects of 5 mg/kg in the MPTP model but the longer exposure duration in the T41 mice. All groups of mice gained weight normally over the 8-week experiment. There were no deaths in WT mice and six deaths among T41 mice distributed non-significantly across Veh, S-2MeGlu, and R-4APA treatment groups.

Nuclei isolation and snRNA-seq

Four month-old WT male C57BL/6J mice were injected IP daily with either Vehicle (normal saline, n = 2), S-2MeGlu at the lowest dose that had significant behavioral effect in our models (10 mg/kg, n = 3) or R-4APA (10 mg/kg, n = 3) for 8 weeks. A single nucleus suspension was prepared from cerebral cortex following the Omni-ATAC protocol⁶⁸, and nuclei sedimented for 20 min at 3,000 RCF and then diluted to 1,000 nuclei/ul by adding Resuspension Buffer (1X PBS with 1.0% BSA and 0.2 U/µl RNase Inhibitor). After isolation, coded samples of nuclei were captured and sequenced by the Stanford Functional Genomics Facility. Single nucleus capture was performed with 10X Genomics Controller device (Pleasanton, CA, United States) following manufacturer’s recommendations. Library preparation was performed according to the 10x Genomics Chromium Single Cell 3’ v3 Reagent Kit and sequenced on HiSeq 4000 instrument (Illumina, San Diego, CA, United States). Reads were subsequently processed using the 10X Genomics Cell Ranger 7.0.1 analytical pipeline with default settings and aligned to the mm10 genome.

SnRNA-seq data analysis, gene set enrichment analysis (GSEA) and IEG module score

Raw data gene–nucleus count matrices generated with Cell Ranger were imported into R using Seurat package⁶⁹. Droplets with > 5% mitochondrial RNA were excluded. Only nuclei with nCount_RNA ≥ 500 and ≤ 40,000; nFeature_RNA ≥ 250 and ≤ 6,000 were included. Datasets were individually normalized using the SCTransform with mitochondrial regression and integrated using Seurat v5 layer-based RPCA, followed by PCA, UMAP, and graph-based clustering. Cluster analysis of single-nucleus data used a graph-based clustering approach^69,70. Differentially expressed genes among clusters were identified using the Seurat “FindConservedMarkers” function, with the default Wilcoxon test, and their major class identity was assigned based on co-expression of multiple gene markers, according to the Allen Brain Institute mouse cerebral cortex taxonomy⁷¹. Clusters enriched for low-quality cells or mixed transcriptional signatures were excluded, and differential expression analyses were subsequently performed at the level of each refined major cell class using Seurat’s MAST framework. For each subset, genes were pre-filtered by detection rate (threshold: ≥1% of cells) and differential expression was tested for S-2MeGlu vs. Veh and R-4APA vs. Veh. MAST models included covariates such as sample identity (replicates) and mitochondrial transcript percentage as latent variables to control for sample effects and mitochondrial load. P-values were adjusted for multiple testing using a Bonferroni correction across genes tested within each comparison, and differentially expressed genes were defined as those with padj < 0.05 and |log2FC| ≥ 0.25 (Supplementary Table 1).

Gene set enrichment analysis was performed using FGSEA (fgsea R package⁷² on differential expression results obtained from MAST analyses. For each refined major cell class and comparison, genes were ranked using a signed statistic reflecting both effect size and statistical significance (i.e. sign(log2FC) × -log10(adjusted p-value)). Enrichment was assessed against Gene Ontology (GO) Biological Process gene sets from MSigDB using mouse annotations restricting gene sets to 20–500 genes. Because fgseaMultilevel was used, enrichment was estimated using adaptive multilevel sampling rather than a fixed number of permutations. Enrichment results are reported using normalized enrichment scores (NES) and multiple-testing–adjusted P values (Supplementary Table 2)⁷³. To reduce redundancy among enriched pathways, GO terms with adjusted p-value < 0.05 were clustered based on leading-edge gene overlap. Pairwise Jaccard similarity was computed between leading-edge gene sets, and similarity graphs were constructed retaining edges with Jaccard similarity ≥ 0.15. Pathway communities were identified using Louvain clustering (igraph package), and a representative term per community was selected based on the lowest adjusted p-value and highest absolute NES (Supplementary Fig. 5B, Supplementary Table 3).

Per-cell immediate early gene (IEG) module score was calculated using Seurat AddModuleScore function using a curated set IEGs (Supplementary Table 4)⁴⁷. Within each major cell class, IEG module scores were compared between each treatment and vehicle using two-sided Wilcoxon rank-sum tests; P values were adjusted for multiple comparisons within each class using the Benjamini–Hochberg procedure.

Statistical analysis

Data (other than for snRNA seq) were analyzed using GraphPad Prism 10 (San Diego, CA) with alpha set to 0.05. Data are presented as mean ± SEM unless otherwise indicated. Sample sizes are specified in each experiment. Hippocampal slice culture data, body and organ weights for Aldh5a1 experiments, and catecholamine concentrations for MPTP experiments were analyzed using one-way or two-way ANOVA with Tukey’s or Dunnett’s multiple comparison test as recommended; simple linear regression and one-way ANOVA plus Dunnett’s multiple comparison test was used for hippocampal slice culture data. Survival data were analyzed using the log-rank (Mantel–Cox) test and pairwise comparisons with Bonferroni’s correction. Data from behavioral models followed pre-specified standardized analyses that included two-way ANOVA for startle test (Psychogenics) and one-way ANOVA for overall kinematic score (Charles River Laboratories) using Fisher’s least significant difference (LSD) as the post hoc test to increase power in our behavioral survey. Fisher LSD dose not correct for repeat comparisons but rather protects from type 1 error by requiring an omnibus F statistic with P < 0.05; to increase rigor we required an omnibus F statistic with P < 0.01 before proceeding with Fisher’ LSD analysis. One sample t test with theoretical mean set to 50% was the pre-specified analysis for Y-maze data as done previously by us and unpaired t test with Welch’s correction for paired comparisons⁴².

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization: AP, TJM. Methodology: AP, TJM. Investigation: AP, JZ, NLS. Resources: CRG. MaS. Visualization: AP, TJM. Funding acquisition: TJM. Supervision: MeS, TJM. Writing (original draft): AP, TJM. Writing (reviewing/editing): AP, AWG, CRG, JZ, KSM, MaS, MeS, NLS, TJM.

Funding

Supported by the Bechtel Research Fund and the John and Gwen Smart Foundation.

Data availability

snRNA-seq raw data were deposited into the Gene Expression Omnibus database under accession number GSE235285.

Declarations

Competing interests

The authors declare no competing interests.