Introduction

Neurodevelopmental disorders (NDDs), including autism spectrum disorder (ASD), intellectual disability (ID), developmental delay (DD), schizophrenia (SCZ) and bipolar disorder (BPD), disrupt brain development and function. NDDs share considerable comorbidities and present with overlapping symptoms, including impairments in cognition, behavior, language, social and motor skills, emotions, and learning ability (1). In addition, the risk genes for NDDs overlap significantly, further suggesting shared mechanistic underpinnings and pathology for these disorders (2, 3).

TRIO encodes a large cytoskeletal regulatory protein with two guanine nucleotide exchange factor (GEF) domains for Rho family GTPases - GEF1 activates Rac1 and RhoG, while GEF2 activates RhoA (4–6). TRIO relays signals from cell surface receptors, acting on GTPases to coordinate cytoskeletal rearrangements critical for proper neurodevelopment (7–15). Trio knockout mice exhibit decreased survival, skeletal muscle defects, and severe defects in brain development (16–18). Selective ablation of Trio in either excitatory or inhibitory neurons alters their morphology, impairs synaptic signaling, and yields NDD-related behavioral defects (19, 20). Trio deficiency also leads to aberrations in long-term potentiation (LTP), as Trio-Rac1 signaling promotes AMPA receptor trafficking to increase synaptic strength (19, 21, 22).

Damaging de novo mutations and ultra-rare variants in TRIO are enriched in individuals with NDDs (23–30). Interestingly, nonsense variants spread throughout TRIO are enriched in individuals with SCZ (30, 31), whereas pathogenic missense TRIO variants in or surrounding the GEF1 domain are associated with ASD/ID (24, 28, 29). Variants in the Trio GEF1 domain that decrease Rac1 activity are associated with milder ID and microcephaly, whereas variants in the adjacent spectrin repeat 8 domain that increase Rac1 activity are associated with more severe ID and macrocephaly (25, 26, 32, 33). Rare missense variants in TRIO have also been observed in BPD, epilepsy, and other disorders, but studies to date are underpowered to establish a causal link with these disorders. How distinct heterozygous TRIO variants differentially impact mammalian brain development and lead to diverse phenotypes remains a fundamental and unresolved question.

We report here the comprehensive analysis of mice heterozygous for discrete Trio variants associated with different NDDs ASD-associated +/K1431M in the Trio GEF1 domain, SCZ-associated +/K1918X, leading to nonsense decay, and BPD-associated +/M2145T in Trio GEF2 domain. We show that distinct Trio NDD-associated variants differentially impair mouse behavior, brain development, and inhibitory and excitatory synaptic transmission. We show, for the first time, that Trio variants that impact GEF1 and GEF2 activity differentially impact presynaptic neurotransmitter release and synaptic vesicle replenishment. While the Trio K1431M GEF1 domain is impaired in Rac1 activation, unexpectedly +/K1431M mice exhibit increased Rac1 activity, associated with increased levels of the Rac1 GEF Tiam1. Acute Rac1 inhibition with NSC23766 rescued glutamate release deficits in +/K1431M variant cortex. Together, our data show how discrete heterozygous TRIO variants that differentially impact Trio biochemical activities yield divergent effects on behavior, neurodevelopment, and synaptic transmission.

Materials and methods

Animal work

Animal work was compliant with federal guidelines and approved by the Yale Institutional Animal Care and Use Committee. Age-matched male and female mice were used for behavioral experiments; males were used for electrophysiological and neuroanatomical studies to reduce potential variation due to estrus cycle (34).

Generation of Trio mutant mice and mouse genotyping

Mice heterozygous for Trio variants K1431M, K1918X, or M2145T were generated via CRISPR/Cas-mediated genome editing and maintained on a C57Bl/6 background. Details on allele design and genotyping are included in Supp. Methods.

In vitro GEF Assays

The K1431M point mutant was generated via site-directed mutagenesis of human WT TRIO GEF1 using the following oligos: 5’-cagcgaataacgatgtatcagctcc-3’ and 5’-ggagctgatacatcgttattcgctg-3’. Recombinant WT and K1431M human TRIO GEF1 and Rac1 were purified from bacteria as described (35). GEF activity was monitored by the decrease in fluorescent signal (λexcitation = 488 nm; λemission = 535 nm) as GTP was exchanged for BODIPY-FL-GDP on Rac1 over 30 min, as described (33, 35).

Body weight, brain weight, and head width measurements

Ear-to-ear head width of P42 mice was measured with calipers in anesthetized mice. Brain weight was measured after removing the olfactory bulbs and brain stem.

Mouse brain lysates and cortical synaptosome preparation

Whole brain lysates from P0-P1 or P35-P42 mice, and cortical synaptosomes from P39-P42 mice, were prepared as previously described (19, 36, 37), with minor modifications (described in Supp. Methods). Protein concentrations were determined with BCA assay (Pierce).

Western blot and quantification

Brain lysates were separated on SDS-PAGE gels, blotted to nitrocellulose membranes, stained with Ponceau S, blocked in 5% nonfat milk in TBS-T, incubated with primary antibodies overnight at 4°C, then with conjugated secondary antibodies at RT for 1 h (see Supp. Table 1 for list of antibodies). Images captured by a ChemiDoc Imaging System (Bio-Rad) were quantified in ImageJ. Signal intensity was normalized to Ponceau S, then to the WT average.

G-LISA

GTP-bound active Rac1 and RhoA levels in were measured using G-LISA kits according to manufacturer’s instructions (Cytoskeleton, Inc., BK128; Inc., BK124).

Behavioral tests

Behavioral tests were performed in mice 6-8 weeks of age (P42-P56), as previously described (19).

Animal perfusion and tissue processing

Mice were transcardially perfused with heparinized PBS followed by 4% paraformaldehyde (PFA) in phosphate buffered saline (PBS). Brains were postfixed in 4% PFA at 4°C for 24 h, then sliced coronally at 30 μm for Nissl stain and immunohistochemistry and 200 μm for dendritic arbor and spine analysis. For details on tissue processing, see Supp. Methods.

Electron microscopy

Electron microscopy of synapses in M1 layer 5 was performed as previously described (19) and synaptic features were quantified in ImageJ as previously described (38–41) from 5-18 fields of view (∼50 μm2) per mouse by a reviewer blinded to genotype.

Electrophysiological recordings

Acute slices were prepared from mice at P35-42 as previously described (19) with modifications noted here. Coronal slices of M1-M2 cortex were cut at 360 µm in ice-cold N-Methyl-D-glucamine-aCSF (NMDG-aCSF, in mM): 120 NMDG, 2.5 KCl, 7 MgSO4, 1.25 NaH2PO4, 0.5 CaCl2, 28 NaHCO3, 13 glucose, 7 sucrose, saturated with 95% O2-5% O2 at 300-320 mOsmol/L, pH 7.4) on a vibratome (Leica) and recovered in normal ACSF at 32°C for 30 min followed by 1 h recovery at RT. Recorded signals were acquired at 100kHz sampling rate and low pass filtered at 6 kHz. Cells were excluded if series resistance changed >20%.

AMPAR- and NMDAR-eEPSCs, miniature EPSCs (mEPSCs), Long term-potentiation (LTP) and were recorded and analyzed as previously described in (19). Gamma-aminobutyric acid receptor and glycine receptor miniature inhibitory postsynaptic currents (GABAR/GlyR-mIPSCs) were recorded at +15 mV.

Paired-pulse stimulation was applied and paired-pulse ratio (PPR) was calculated as described in (19). Calculations of the readily releasable pool (RRP) size, probability of release (Pr) and synaptic vesicles (SVs) depletion and RRP recovery (fractional) rate were made from recordings of AMPAR-eEPSCs under high frequency stimulation (HFS, 40Hz; 15 pulses) for each neuron as previously described (42–44) (details see in Supp. Methods).

Rescue experiments were performed as described above with addition of 100 µM NSC23766 into the recording solution with 5 min incubation period to allow for efficient slice penetration. All train data were processed and analyzed using OriginLab 10 software.

Quantitative Proteomics

Tissue from P21 mouse cortex (4 per genotype) was processed for tandem mass tag (TMT) peptide labeling as previously described (19) with minor modifications. Desalted 16-plex TMT-labeled peptides were separated by HPLC, and proteome analysis was performed using an EASY-nLC 1200 UHPLC coupled to an Orbitrap Eclipse mass spectrometer (Thermo Fisher Scientific). MS2 spectra were searched using Spectrum Mill (Agilent) against the UniProt Mouse Database. Technical details are described in Supp. Methods.

Gene Set Enrichment Analysis

Mouse UniProtIDs for identified proteins were converted to human orthologs and ranked by signed log10(nominal p-value) (with sign indicating direction of fold-change from WT). GSEA 4.3.3 (45, 46) was used to run the GSEAPreranked tests against all gene sets in the Human MSigDB Collection C2 v2023.2. SynGO 1.2 (47) was used to identify synaptic ontologies for differentially expressed proteins with nominal p-value <0.05.

Data Analysis

Statistical analyses were performed using GraphPad Prism 10. Data in bar graphs are presented as mean ±SEM, with individual data points graphed when applicable. Sample size ‘n’ is annotated within the bars or in the figure legend for each group. Distributions were tested for normality and outliers were removed before proceeding with statistical tests. Specific details of statistical tests performed, adjusted with post-hoc Bonferroni test for multiple comparisons (MC) where appropriate, are included in the figure legends. Significance was defined by a p-value less than 0.05: nsp<0.1; *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001.

Results

Generation of Trio variant mice and impact on Trio isoforms

We used CRISPR/Cas9 technology to generate mice bearing heterozygous Trio rare variant alleles associated with ASD (+/K1431M) and SCZ (+/K1918X), and a de novo mutation found in an individual with BPD (+/M2145T) (Fig. 1A,B). We chose these alleles for their discrete effects on TRIO levels or activity: K1431M impairs TRIO GEF1 activity in vitro up to 8-fold (Supp. Fig. 1A,B) (24), M2145T TRIO GEF2 has reduced ability to activate RhoA as a function of protein concentration in cells (24), and K1918X creates a premature stop codon between GEF1 and GEF2.

Genetically engineered mice with heterozygosity for K1431M, K1918X, or M2145T Trio variants have divergent effects on Trio protein expression, Rho GTPase activity, and NDD-like behaviors.

(A) Schematic of major Trio isoforms present in the adult mouse brain, with locations of engineered neurodevelopmental disease (NDD)-associated Trio variants: K1431M autism spectrum disorder (ASD)-associated missense variant in the GEF1 DH domain; K1918X schizophrenia (SCZ)-associated nonsense variant that lies just before the GEF2 domain; and M2145T bipolar disorder (BPD)-associated missense variant in the GEF2 DH domain. (B) Representative sequencing chromatograms of wild-type (WT), +/K1431M, +/K1918X, and +/M2145T mice. Arrows indicate heterozygosity for the variant alleles. (C) Representative immunoblots for Trio in P0 brain lysates using an antibody against Trio spectrin repeats (SR5-6). (D) Quantification of Trio protein levels from P0 brain lysates. Trio protein levels are reduced only in the brains of +/K1918X mice compared to WT controls (0.545 ±0.126 of WT level, p=0.0046). (E-H) Activity levels of Rac1 (E,G) and RhoA (F,H) in whole brain homogenates of neonate (P0, E-F) and adult (P42, G-H) Trio variant mice as measured by G-LISA assay. Rac1 activity is increased in +/K1431M mice relative to WT at both ages (1.106 ±0.027-fold at P0, p=0.0035; 1.509 ±0.175-fold at P42, p=0.0279) and decreased in neonate +/K1918X mice (0.908 ±0.0.032-fold, p=0.0230), with a trend towards increased activity in adult +/M2145T mice (1.438 ±0.183-fold, p=0.0843); meanwhile RhoA activity appears unchanged in all mice relative to WT, though there may be a trend towards decreased activity in +/K1918X neonates (0.840 ±0.074-fold, p=0.1292). (I,J) Activity levels of Rac1 (I) and RhoA (J) in synaptosomes isolated from P42 mouse cortex. Rac1 activity is increased in +/K1431M synaptosomes (1.125 ±0.107-fold, p=0.0023), while RhoA activity is decreased in +/M2145T synaptosomes (0.731 ±0.042-fold, p=0.0093) relative to WT. Data are presented as mean ±SEM. For (D-J), one-way ANOVA with post-hoc Bonferroni MC test identified differences from WT (nsp<0.1, *p<0.05, **p<0.01). Numbers of mice quantified per group are annotated inside the bar. (K) +/K1431M and +/K1918X male mice had decreased latency to fall off an accelerating rotarod compared to WT male mice. Two-way ANOVA with post-hoc Bonferroni MC test identified differences from WT for each individual trial (*p<0.05, **p<0.01, ***p<0.001, ****p<0.0001). Linear regressions identified differences from WT in slopes (WT 16.96 ±1.344; +/K1431M 7.270 ±2.019, p<0.0001; +/K1918X 10.61 ±1.444, p<0.0001; ####p<0.0001). n=40 WT; 10 +/K1431M; 16 +/K1918X; 13 +/M2145T male mice. (L) WT mice displayed normal social preference and spent more time with a stranger mouse (Str.) relative to an inanimate object (Obj.) in a three-chamber interaction test. Social interactions were impaired in +/K1431M male mice. (M) WT mice spent more time with a novel object (N) than with a familiar object (F). Novel object recognition was impaired in +/K1918X male mice. All data are presented as mean ±SEM. For (L-M), two-way ANOVA with post-hoc Bonferroni MC test identified differences from WT (nsp<0.1, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001). Numbers of mice quantified per group are annotated inside the bar. (N) Male +/K1918X mice exhibited increased nestlet shredding over 30 min (26.26 ±3.61% shredded vs WT 14.26 ±2.97%; p=0.0433), and +/K1431M mice exhibited a trend toward increased nestlet shredding (25.90 ±4.34% shredded, p=0.1038) compared to WT mice. One-way ANOVA with post-hoc Bonferroni MC test identified differences from WT (nsp<0.1, *p<0.05). n=19 WT; 10 +/K1431M; 15]+/K1918X; 9 +/M2145T male mice.

Mice heterozygous for these Trio alleles survived to adulthood. Mice homozygous for Trio K1431M and M2145T survived to adulthood, with genotypes from offspring of +/K1431M intercrosses observed in Mendelian ratios, but fewer than expected M2145T homozygotes obtained (Supp. Fig. 1C). K1918X homozygote pups were not observed, as expected for a null allele (16, 18). We focused on heterozygotes, as most rare damaging Trio variants are heterozygous in humans. Heterozygosity for the K1431M and M2145T alleles did not alter the levels of the predominant Trio isoforms (Trio9S, 263 kDa; Trio9L, 277 kDa; Duet, 145 kDa; cerebellum-specific Trio8 217 kDa (48, 49)) in the brain at postnatal day 0 (P0) (Fig. 1C-D) or P42 (Fig. 1A, Supp. Fig. 1D-L). Trio9 protein levels were reduced by ∼50% in the brains of +/K1918X mice at P0 (Fig. 1C-D) and P42, and we did not detect residual truncated protein (expected 217 kDa), suggesting this mutation leads to nonsense-mediated decay (Supp. Fig. 1D-L); levels of Trio8 and Duet were unaffected (Supp. Fig. 1D-L).

Trio variant alleles differentially impact active Rho family GTPase levels

Given the effects of these variants on TRIO GEF1/2 activities, we measured levels of active GTP-bound Rac1 and RhoA in brains of neonatal (P0) and adult (P42) Trio variant-bearing mice (Fig. 1E-H). In concordance with the ∼50% reduction in Trio levels in +/K1918X brains, Rac1 activity was decreased in P0 +/K1918X brains (91% of WT activity), with a trend toward decreased active RhoA levels (84% of WT, p=0.0865). By P42, +/K1918X brains did not differ from WT in Rac1 or RhoA activity.

As K1431M decreases TRIO GEF1 catalytic activity in vitro (Supp. Fig. 1A,B) (20, 24, 29), we anticipated Rac1 activity might be reduced in +/K1431M mice. Instead, active Rac1 levels were increased in +/K1431M whole brain lysates compared to WT controls at P0 ((Fig. 1E),111% of WT) and even higher at P42 ((Fig. 1G), 150% of WT). We also did not detect changes in active RhoA levels at the whole brain lysate level in +/M2145T mice at either age (Fig. 1F,H).

Because Trio, Rac1, and RhoA are enriched at synapses (15, 22, 24, 50), we also measured their activities in synaptosomes (Supp. Fig. 1M-P). Rac1 activity was significantly increased in +/K1431M crude synaptosomes from P42 cortex (112% of WT) (Fig. 1I), consistent with measurements in whole brain lysates. Consistent with its reduced GEF2 activity, RhoA activity was decreased in +/M2145T synaptosomes compared to WT (73% of WT) (Fig. 1J).

Distinct Trio variants differentially impact mouse behavior

NDDs affect learning and memory, compulsivity, motor coordination, and social skills, hence we assessed these skills using a diverse array of established behavioral tests (19) in mice bearing Trio variants. +/K1431M and +/K1918X mice of both sexes fell from an accelerating rotarod with reduced latency and improved more slowly in this skill over repeated trials relative to WT littermates (Fig. 1K, Supp. Fig. 1Q), while +/M2145T mice performed similarly to WT. No deficits in muscle strength were noted in any genotype using the Kondziela inverted screen test prior to rotarod testing, and no differences in locomotor or exploratory activity were seen in an open field test (data not shown). In a three-chamber social interaction test, +/K1431M mice of both sexes showed no preference for the stranger mouse over the object (Fig. 1L, Supp. Fig. 1R). +/M2145T mice of both sexes and +/K1918X males exhibited preference for the stranger mouse, similar to WT. In a novel object recognition test, +/K1918X mice of both sexes and +/M2145T females failed to discriminate between the novel and familiar objects, while +/K1431M mice and +/M2145T males exhibited normal discrimination between novel and familiar objects similar to WT (Fig. 1M, Supp. Fig. 1S). +/K1918X males exhibited significantly more compulsive nestlet shredding relative to WT, while +/K1431M males showed a trend toward increased nestlet shredding (Fig. 1N).

Overall, behavioral phenotypes overlapped between +/K1431M and +/K1918X mice, though there were distinct differences in social interaction and memory tests. +/M2145T mice exhibited the fewest phenotypes in measured tasks. Together, these data indicate that these Trio alleles differentially impact behavior.

Trio +/K1431M and +/K1918X mice have smaller brains

TRIO variants that reduce TRIO GEF1 activity are associated with microcephaly (25, 26, 33), so we assessed head and brain size in Trio variant mice (Fig. 2A-E, Supp. Fig. 2A-E). After adjusting for body weight, both the head width and brain weight (Fig. 2D,E) were reduced in adult +/K1431M mice relative to WT, consistent with the microcephaly seen in patients harboring TRIO GEF1-deficient alleles. Head width in +/K1918X mice was reduced, but in proportion to their body weight (head width-to-body weight ratios did not differ between +/K1918X and WT mice, p>0.95). However, +/K1918X mice still displayed a reduced brain weight relative to WT, suggesting a disproportionate reduction in brain mass relative to their already smaller head size. Meanwhile, +/M2145T mice did not differ from WT in either head width or brain weight when adjusted for body weight.

Trio +/K1431M and +/K1918X mice have smaller brain weights, but only +/K1918X brains have smaller less complex neurons.

(A) Ear-to-ear head width is reduced in P42 +/K1918X and +/M2145T compared to WT male mice (+/K1918X 12.125 ±0.074 mm, p=0.0096; +/M2145T 12.012 ±0.152 mm, p=0.0007; vs WT 12.414 ±0.049 mm). (B) Brain weight is significantly reduced in P42 +/K1431M, +/K1918X, and +/M2145T males compared to WT (+/K1431M 0.379 ±0.004 g, p=0.0006; +/K1918X 0.345 ±0.004 g, p<0.0001; +/M2145T 0.381 ±0.005 g, p=0.0016; vs WT 0.399 ±0.004 g). (C) Body weight is increased in P42 +/K1431M and decreased in +/K1918X males (+/K1431M 22.911 ±0.382 g, p=0.0034; +/K1918X 20.700 ±0.338 g, p=0.0153; vs WT 21.745 ±0.224 g). (D) Head widths normalized to body weight of P42 +/K1431M male mice were reduced 11.8% compared to WT mice (0.520 ±0.008 mm/g vs WT 0.589 ±0.011 mm/g, p=0.0001). Head width-to-body weight ratios were calculated per individual mouse, with mouse number per group annotated within the bar. (E) Brain weights normalized to body weight of P42 +/K1431M and +/K1918X male mice were reduced 9.9% and 9.2%, resp., compared to WT mice (+/K1431M 0.0165 ±0.0004, p=0.0002; +/K1918X 0.0167 ±0.0004, p=0.0015; vs WT 0.0184 ±0.0004). Ratios were calculated by dividing the mean brain weight (B) by the mean body weight (C) of overlapping but non-identical populations of mice (see (B) and (C) for mouse number per group). For (A-E), data are presented as mean ±SEM; only male mice are shown here (data for females in Supp. Fig. 2). One-way ANOVA with post-hoc Bonferroni MC test identified differences from WT (nsp<0.1, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001; n=number of mice per group, annotated within the bar). (F) Representative Nissl-stained image of a 30 μm coronal slice from a P42 WT brain. Magnification of the cortex (dotted black box) reveals cortical layers. (G) Total cross-sectional brain area of Nissl-stained coronal sections was reduced by ∼9% in P42 +/K1918X brains (43.21 ±0.577 mm2 vs WT 47.29 ±0.823 mm2, p=0.0045). (H) Total thickness of the cortex, in the region marked by the dotted black box in (F), is reduced by ∼8% in P42 +/K1918X brains (1.304 ±0.02262 mm vs WT 1.416 ±0.01417 mm, p=0.0021). For (G,H), data are presented as mean ±SEM. Ordinary one-way ANOVA with post-hoc Bonferroni MC test identified differences from WT (**p<0.01; n=number of mouse brains, annotated within the bar). (I) Representative maximum projection fluorescence image and corresponding dendritic arbor reconstruction of a motor cortex Layer 5 pyramidal neuron (M1 L5 PN) from a WT mouse. (J,K) Dendritic field size as measured by convex hull analysis of basal (J) and apical (K) dendrite arbor reconstructions of M1 L5 PNs of P42 Trio variant mice. +/K1918X trended toward smaller basal dendritic field size (0.1172 ±0.0078 mm2 vs WT 0.1368 ±0.0077 mm2, p=0.0933), and both +/K1918X and +/M2145T had significantly smaller apical dendritic field size (+/K1918X 0.5157 ±0.0169 mm2, p=0.0460; +/M2145T 0.4893 ±0.0285 mm2, p=0.0062; vs WT 0.6081 ±0.0319 mm2) compared to WT L5 PNsOrdinary one-way ANOVA with post-hoc Bonferroni MC test (B) identified differences from WT (nsp<0.1, *p<0.05, **p<0.01). (L-M) Sholl analysis revealed basal (L) and apical (M) dendritic arborization changes in Trio variant M1 L5 PNs compared to WT: both basal and apical arborization was reduced in +/K1918X, while proximal basal arborization was increased in +/K1431M. Two-way ANOVA (stacked) with post-hoc Bonferroni MC test identified differences from WT at each radius centered at the soma (*p<0.05, **p<0.01, ***p<0.001) For (J-M), n=23 neurons from 17 WT mice, 22 neurons from 15 +/K1431M mice, 20 neurons from 12 +/K1918X mice, 20 neurons from 14 +/M2145T mice.

Trio variant mice show limited changes in cortical organization

We examined whether the decreased brain weight and head size observed in +/K1431M and +/K1918X mice were associated with anatomical defects. Histological analyses did not reveal gross morphological defects (data not shown). We observed reductions in both total cross-sectional brain area and cortical thickness in +/K1918X brains (Fig. 2F-H), consistent with their smaller head size and brain weight. Significant decreases in cortical layer 2/3 and layer 5 thickness were observed in +/K1918X brains; but these were in proportion to the relative decrease in cortical thickness as compared to WT (Supp. Fig. 2F-H). No changes were observed in +/M2145T mice.

No differences in total cortical cell density or in layer-specific cell density were observed in Trio variant mice relative to WT, nor were numbers or proportions of cortical NeuN+ neuronal cells or PV+ inhibitory neurons altered relative to WT, although there were trends toward increased cell density in +/K1918X cortex and increased NeuN+ cell density in +/M2145T motor cortex (Supp. Fig. 2M-Q). These data suggested that the reduced brain size +/K1918X mice resulted from a loss of neuropil.

Trio variant heterozygotes exhibit alterations in dendritic arbors and synaptic ultrastructure

Altered dendritic arbor morphology and dendritic spine abnormalities are hallmarks of NDDs (51–57). Excitatory neuron-specific ablation of one or both Trio alleles decreased dendritic arborization, increased spine density, and yielded smaller synapses in cortex area M1 Layer 5 pyramidal neurons (M1 L5 PNs) (19).

Sholl analysis of M1 L5 PNs revealed significant reductions in both basal and apical dendritic arbor complexity and dendritic field size in P42 +/K1918X neurons, an increase in proximal basal arbor complexity in +/K1431M neurons, and a slight decrease in distal apical arbor complexity in +/M2145T neurons relative to WT (Fig. 2I-M). Path length analysis revealed that branch numbers were increased for lower order dendrites in +/K1431M neurons, while average branch lengths were decreased at higher order dendrites in +/K1918X neurons (Supp. Fig. 3F-K). None of the Trio variant heterozygotes exhibited altered dendritic spine density on M1 L5 pyramidal neurons compared to WT mice on either apical or basal arbors (Supp. Fig. 3L,M).

Electron microscopy of cortical area M1 L5 revealed that synapse density was significantly increased in +/K1918X mice compared to WT (Fig. 3A,B), possibly due to a net reduction in neuropil resulting from smaller dendritic arbors. Within synapses, postsynaptic density (PSD) length was slightly decreased in +/K1918X and +/M2145T mice, but not in +/K1431M mice (Fig. 3C). Cross-sectional presynaptic bouton area and spine head area were similar to WT in each Trio variant heterozygote (Fig. 3D,E).

Trio variants differentially impact synapse ultrastructure and synaptic vesicle distribution.

(A) Representative electron micrograph (EM) images from motor cortex layer 5 (M1 L5) of P42 WT and Trio variant mice. Post-synaptic regions are pseudo-colored in cyan; pre-synaptic regions in magenta. Scale bar = 0.5 μm. (B) Asymmetric synapses from all EM images of M1 L5 were quantified (from 18 fields of view per mouse, 5 mice per group). Synapse density was increased in +/K1918X mice (0.09205 ±0.004775 synapses/um2; vs WT 0.07633 ±0.003954 synapses/um2, p=0.0345). (C) PSD lengths were decreased in M1 L5 synapses of +/K1918X and +/M2145T mice (+/K1918X 0.2926 ±0.004652 um, p=0.0204; +/M2145T 0.2916 ±0.004922 um, p=0.0142; vs WT 0.3125 ±0.005612 um). (D,E) Presynaptic bouton and spine head areas of M1 L5 synapses were unchanged from WT in all Trio variants. For (C-E), number of measurements per mouse across all 5 mice per genotype are denoted in the corresponding figure. (F) Synaptic vesicles (SVs) counted per 100 nm of active zone (AZ) length in M1 L5 as a function of distance from the AZ. +/M2145T showed an increase in the readily releasable pool (RRP) identified as docked SVs (15 nm from AZ; 1.23 ±0.05 vs. WT 0.90 ±0.05) and an increase in tethered SVs (50 nm from AZ; 1.44 ±0.04 vs. WT 1.20 ± 0.05). +/K1918X and +/M2145T also showed an increase in the reserve pool of SVs (200 nm from AZ; 3.51 ±0.21 and 3.81 ±0.18, resp. vs WT 2.74 ±0.16). (G) Total releasable pool, calculated as the number of SVs at 15-150 nm from AZ per area of distribution (nm2). RRP was significantly increased in +/M2145T (0.257 ±0.007 vs WT 0.228 ±0.008), driven by increased docked and tethered SVs. All data are presented as mean ±SEM. Ordinary one-way ANOVA with post-hoc Bonferroni MC test identified differences from WT (*p<0.05, **p<0.01, ***p<0.001, ****p<0.0001). For (F,G), 35-50 synapses per mouse were analyzed.

Synaptic vesicle (SV) distribution was significantly altered in +/M2145T mice relative to the other genotypes (Fig. 3F,G). Within the total releasable pool 15-150 nm from the active zone (AZ), both docked SVs at 15nm comprising the readily-releasable pool (RRP) (40, 41) and tethered SVs at 50 nm were significantly increased in density in +/M2145T mice relative to the other genotypes (Fig. 3F,G). SVs distribution at 200 nm from the AZ which contribute to the reserve pool were also significantly increased in +/M2145T, suggesting an overall SV pool size. No differences in synaptic vesicle distribution were noted in +/K1431M mice, while +/K1918X showed modestly increased SVs at 200 nm from AZ relative to WT.

Synaptic transmission and plasticity are differentially impaired by distinct Trio variants

Loss of Trio function or disruption of TRIO GEF1 activity in slice culture decreases AMPAR levels at excitatory synapses (19, 21, 22, 29), while mice bearing a GEF1-deficient Trio allele exhibited decreased gamma-aminobutyric acid receptor (GABAR)- and glycine receptor (GlyR)-mediated inhibitory miniature current (mIPSC) frequencies in the prefrontal cortex (20). To explore how Trio variants impact synaptic function, we measured both miniature excitatory currents (mEPSCs) and mIPSCs in M1 L5 PNs in each of the Trio variant heterozygotes.

AMPAR-mediated mEPSC amplitudes were significantly increased in +/K1431M and +/K1918X mice relative to WT littermates, with no change in their frequencies (Fig. 4A-C). In contrast, AMPAR-mediated mEPSC amplitudes were unchanged in +/M2145T mice, but their frequencies were significantly increased. No significant changes in NMDAR mEPSC amplitudes were noted, while NMDAR mEPSCs frequencies were decreased in +/K1431M and increased in +/M2145T mice cortex (Fig. 4D-F). Notably, these findings correlated with gross alterations in the ratio of NMDAR/AMPAR-mediated evoked (e)EPSCs measured in M1 L5 PNs following stimulation in L2/3. All Trio variant heterozygotes showed decreased NMDA/AMPA ratios, indicating imbalances in NMDAR-versus AMPAR-mediated conductance (Supp. Fig. 4A,B). Significant decreases in mIPSC frequencies were noted in +/K1431M and +/M2145T mice relative to WT mice, with no change in amplitudes (Fig. 4G-I). +/K1918X mice exhibited increased mIPSC amplitude with no observed change in mIPSC frequency. Together, these data indicate that the Trio variants differentially impact excitatory and inhibitory transmission.

Trio variant mice exhibit deficits in synaptic signaling and LTP.

(A,D) Representative traces of miniature excitatory AMPAR-, NMDAR-mediated mEPSCs and (G) inhibitory postsynaptic currents mIPSCs in M1 L5 pyramidal neurons of WT, +/K1431M, +/K1918X and +/M2145T mice. (B) AMPAR-mediated mEPSC amplitudes (I) were significantly increased in +/K1431M (16.67 ± 1.04 pA) and +/K1918X (14.71 ± 0.92 pA) slices, with no observed changes in +/M2145T slices (13.90 ± 1.16 pA) compared to WT (11.25 ± 0.84 pA; n=17-25 neurons from ≥6-8 mice per group). (C) No significant changes in AMPAR mEPSC frequencies (q) were observed in +/K1431M and +/K1918X, while +/M2145T had an increase (2.20 ± 0.15 1/s; vs WT 1.55 ± 0.09 1/s). (E, F) NMDAR mEPSC frequencies were reduced in +/K1431M (0.89 ± 0.12 1/s; vs WT 1.3324 ± 0.11 1/s; n=9-13 neurons from ≥ 5-7 mice per group) and showed a slight but significant increase in +/M2145T mice (1.68 ± 0.10 1/s). (H, I) GABA/GlyR mIPSC amplitudes were significantly increased in +/K1918X vs WT (23.69 ± 2.89 pA; vs 15.86 ± 1.56 pA, resp.), while frequencies were decreased in +/K1431M and +/M2145T (0.94 ± 0.14 1/s and 1.64 ± 0.19 1/s resp.; vs WT 2.44 ± 0.20; n=16-26 neurons from ≥6-8 mice per group). For (B-I), data are presented as mean ± SEM. One-way ANOVA with post-hoc Bonferroni test identified differences from WT (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001). (J) Averaged representative traces of baseline and post-TBS eEPSCs currents in M1 L5 PNs of WT and Trio variant mice. (K) Normalized eEPSCs amplitudes measuring LTP in L5 PNs by TBS in L2/3 afferents in all genotypes showed a significant decrease in the initiation and no potentiation of the LTP in +/K1431M and +/K1918X, with the increase in initiation and potentiation of +/M2145T M1 L5 PNs compared to WT. LTP was induced at 0 min. RM two-way ANOVA with post-hoc Bonferroni MC test identified differences from WT (****p<0.0001; n=6-8 neurons from ≥4-5 mice per group).

Finally, we tested the ability of the L5 PNs to undergo long-term potentiation (LTP) following theta-burst stimulation of L2/3 afferents (Fig. 4J,K). While LTP was robustly induced and potentiated in M1 L5 PNs from WT mice, LTP induction and potentiation were deficient in slices from +/K1918X and +/K1431M mutant mice. In contrast, +/M2145T L5 PNs showed increased induction and prolonged potentiation of LTP compared to WT L5 PNs.

Excitatory neurotransmitter release is altered in Trio +/K1431M and +/M2145T heterozygotes

In addition to its postsynaptic roles, Trio localizes presynaptically and interacts with the presynaptic active zone scaffolding proteins Bassoon and Piccolo (14, 15, 22). To assess the impact of Trio variants on presynaptic function, we first measured the paired-pulse ratio (PPR). Synapses with low probability of release (Pr), e.g. L2/3 synapses onto L5 PNs, can undergo paired-pulse facilitation (PPF), a relative increase of synaptic transmission in response to the second of two closely-apposed stimuli. WT and +/K1918X M1 L5 PNs exhibited normal facilitation of eEPSC amplitudes that decreased with increased interstimulus interval (ISI) (Fig. 5A,B). PPF was significantly enhanced in +/K1431M M1 L5 PNs at short ISIs relative to WT, suggesting a possible reduction in glutamate Pr (Fig. 5A,B). The +/M2145T PPR curve was unusual, with significantly reduced PPF at short ISIs, yet clearly increased PPF at longer ISI (Fig. 5A,B) compared to WT. The decreased PPF at initial ISI in +/M2145T mice correlated with increased mEPSC frequency (Fig. 4A-C), suggestive of a possible increase in spontaneous glutamate Pr.

Trio GEF damaging variant mice have deficiencies in excitatory neurotransmitter release.

(A) Representative traces recorded from M1 L5 PNs of WT, Trio variant mice, and rescue experiment in acute slices of +/K1431M with 100 µM NSC23766 (NSC, Rac1 inhibitor) in response to paired-pulse stimulation in M1 L2/3. (B) Paired-pulse ratio (PPR) at varying interstimulus intervals (ISI), overlaid with a single exponential fit (except for +/M2145T data). An increase in the initial PPR was observed in M1 L5 PNs of +/K1431M mice (35 ms: 1.70 ± 0.089, p=0.003; 60 ms: 1.40 ± 0.07, p=0.046; 100 ms: 1.27 ± 0.05, p=0.031; n=20-34 neurons from ≥7-9 mice per group) with no changes in +/K1918X. In +/M2145T slices, we observed a decrease in initial PPR at shorter ISI (35 ms: 1.05 ± 0.06, p<0.0001; 60 ms: 0.97 ± 0.06, p=0.037) and an increase at longer ISI (100 ms: 1.36 ± 0.09, p=0.034; 200 ms: 1.18 ± 0.08, p=0.013) compared to WT (35 ms: 1.40 ± 0.04; 60ms: 1.21 ± 0.03; 100 ms: 1.13 ± 0.03; 200 ms 1.0 ± 0.02; 300 ms 0.96 ± 0.17). Acute application of NSC onto +/K1431M slices significantly shifted the PPR curve downwards at all points compared to untreated +/K1431M slices and showed no significant difference from WT (+/K1431M + NSC 35 ms: 1.25 ± 0.06, p<0.0001; 60 ms: 1.13 ± 0.052, p=0.0007; 100 ms: 1.02 ± 0.053, p=0.0017; 200 ms 0.91 ± 0.039, p=0.0043; 300 ms 0.88 ± 0.045, p=0.021). (C) Representative traces of AMPAR eEPSCs in M1 L5 PNs under HFS (15 pulses at 40 Hz) in L2/3. (D) AMPAR eEPSCn amplitudes normalized to eEPSC1 of the train revealed changes in the depletion rates during HFS in Trio variants compared to WT (tau decay (τd), WT: 2.70 s, +/K1431M: 3.19 s, +/M2145T: 4.79 s, +/K1918X: 3.52 s, +/K1431M + NSC: 2.68 s; n=12-15 neurons from 5-7 mice). (E) The estimated glutamate probability of release (Pr) was decreased in +/K1431M (0.13 ± 0.099) and increased in +/M2145T (0.26 ± 0.019), with no significant change in +/K1918X (0.15 ± 0.01) compared to WT M1 L5 PNs (0.19 ± 0.01; n=11-15 neurons from ≥ 5 mice per group); acute NSC application rescued Pr in +/K1431M (0.22 ±0.019; n=11-15 neurons from ≥ 5 mice per group). (F) The calculated size of the readily releasable vesicle pool (RRP) was increased in +/M2145T M1 L5 PNs compared to WT (665.7 ± 68.5 pA vs 415.8 ± 43.9 pA). RRP in +/K1431M synapses before or after treatment with NSC did not differ from WT (543.1± 64.4 pA; + NSC: 427.9± 79.2 pA vs 415.8 ± 43.9 pA) (G) Exponential fits of the fractional recovery plotted vs ISI. Time of recovery, measured by exponential tau recovery (τR), was significantly decreased in +/K1431M M1 L5 PNs (6.3 s, vs WT 1.7 s). +/K1431M exhibited an inability to fully recover to initial levels after ISI 18 s, while NSC application has allowed for full recovery at 18s. It improved but did not fully rescue tau recovery time in +/K1431M (5.2 s). Data are presented as mean ± SEM, with significant differences from WT tested using one-way ANOVA with post-hoc Bonferroni (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001).

We used high frequency (40Hz) stimulation (HFS) trains to quantitatively estimate glutamate Pr, RRP size, and rates of SV depletion and recovery. A plot of the normalized eEPSC responses to a HFS train stimulation again revealed facilitation upon the first 2-3 stimulations that was increased in +/K1431M and decreased in +/M2145T slices relative to WT, with no changes in +/K1918X slices (Fig. 5C,D). Initial facilitation was followed by decaying eEPSC amplitudes, reflecting SV depletion under HFS. All Trio variants exhibited a slower train decay rate relative to WT during HFS with +/M2145T depleting at half the rate of WT (τd, 4.8 s vs WT: 2.7 s) (Fig. 5D).

We used a ‘Decay’ method (42, 43) to estimate Pr and RRP size from HFS trains. Glutamate Pr in the L2/3-L5 synapses was increased in +/M2145T mice, while it was decreased for +/K1431M mice (Fig. 5E), consistent with the relative changes observed in PPF for these mice. RRP size was much larger in L2/3-L5 synapses of +/M2145T mice relative to WT (Fig. 5F), consistent with the increased SV distribution found in close apposition to the AZ (Fig. 3F,G). We tested the ability of the Trio variant heterozygotes to recover after train depletion by pairing HFS trains with a single stimulus at increasing intervals (0.01, 2, 6, 9, 12, 18s). The recovery rate (τR) was significantly slower in +/K1431M L5 PNs and they did not recover to their initial strength within 18 s, plateauing at 78% of maximal recovery compared to WT (Fig. 5G). Altogether, +/M2145T and +/K1431M L2/3-L5 synapses show a deficiency in both synchronous and spontaneous glutamate release as identified by HFS and frequency changes of mEPSCs.

Trio variant cortex display different proteomic signatures

We used comparative proteomics from P21 cortex to identify proteins and pathways that were differentially altered by the Trio variants. We quantified a total of 7,362 proteins, finding distinct differences in the cortical proteome for each genotype (Supp. Fig. 5, Supp. Table 2). Gene Set Enrichment Analysis (GSEA) (45, 46) revealed alterations in distinct functions for each Trio variant (Fig. 6A). Of note, the only enriched gene set specific to neurons was downregulation of the synaptic vesicle pathway in +/K1431M cortex; all other gene sets were not cell type-specific.

Trio variant mice show different molecular changes in the cortex involving presynaptic machinery and Rac1 GEFs.

(A) Bar graph illustrating the top enriched pathways (FDR q-value<0.2, *FDR<0.05) identified by gene set enrichment analysis (GSEA) for each Trio mutant mouse compared to WT, using all proteins quantified by mass spectrometry in P21 cortex (7362 proteins, n=4 mice per genotype), sorted by normalized enrichment score (NES). Pathways with a positive NES are upregulated compared to WT; pathways with a negative NES are downregulated compared to WT. [] denotes the gene set: [R] = Reactome, [WP] = WikiPathways, [K] = KEGG. (B,C) Bar graphs illustrating the top enriched (FDR q-value <0.001), (B) cellular components, and (C) biological processes identified by GSEA, using synaptic proteins from SynGO gene sets (n=1077 proteins). For a complete list of all enriched pathways identified by GSEA and SynGO, see Supp. Table 3. (D) Representative immunoblots for Munc18-1, synaptophysin (Syp), syntaxin-1a (Stx1a), and synaptotagmin3 (Syt3) in synaptosomes isolated from P42 cortex of WT and Trio variant mice. (E-H) Normalized intensity levels from immunoblots of select presynaptic proteins from P42 cortical synaptosomes. Munc18-1, Syp, and Syt3 are significantly increased in +/M2145T synaptosomes compared to WT; Syp is increased while Stx1a is significantly decreased in +/K1431M synaptosomes compared to WT. Ordinary one-way ANOVA with post-hoc Bonferonni MC test identified differences from WT (*p<0.05, **p<0.01, ***p<0.001; n=synaptosomes from 14 WT, 10 +/K1431M, 9 +/K1918X, and 7 +/M2145T male mice). (I) Representative immunoblots for Kalirin, Tiam1, and VAV2 in P42 cortical RIPA lysates of WT and Trio variant mice. (J-L) Normalized intensity levels from immunoblots of select RhoGEFs from P42 cortical lysates. Kalirin levels are unchanged in all Trio variant cortex compared to WT; Tiam1 levels are increased ∼47% in +/K1431M and increased ∼45% in +/M2145T cortex compared to WT; VAV2 is increased ∼34% in +/M2145T cortex compared to WT. Unpaired t-tests identified differences from WT (*p<0.05; n=6 mice per genotype).

We used SynGo (47) to investigate whether the Trio variants impacted synaptic functions. 1,067 of the 7,362 total quantified proteins were synaptic proteins listed in the SynGO geneset. When restricted to brain-specific genes, all three Trio variant heterozygotes showed enrichment of differentially expressed proteins (DEPs) in synaptic processes. Notably, +/M2145T upregulated DEPs and +/K1431M downregulated DEPs were significantly enriched at the presynapse, but with only +/K1431M downregulated DEPs showing significant enrichment for postsynaptic neurotransmitter receptor and synaptic vesicle cycling (Fig. 6B,C). Together, our proteomics data point to a significant deficit in presynaptic function in both +/K1431M and +/M2145T cortex, as well as a significant effect of +/K1431M on postsynaptic function.

Rho GEFs and synaptic regulatory proteins are altered in Trio +/K1431M and +/M2145T heterozygotes

Given our findings from proteomic analysis and electrophysiology, we measured levels of key presynaptic regulators, including synaptophysin (Syp), syntaxin binding protein1 (Stxbp1, also known as Munc18-1), syntaxin1a (Stx1a) and synaptotagmin3 (Syt3), which are crucial for synaptic vesicle (SV) tethering, docking, replenishment, and calcium-dependent replenishment, respectively. We also measured levels of presynaptic proteins in P42 cortical synaptosomes, where these proteins are enriched. Munc18-1, Syt3, and Syp levels were increased in +/M2145T synaptosomes relative to WT (Fig. 6D-H). Meanwhile, Stx1a levels were significantly decreased in +/K1431M synaptosomes compared to WT, with no significant changes in +/K1918X compared to WT mice.

The elevated Rac1 activity in +/K1431M brain lysates and synaptosomes (Fig. 1I) seemed at odds with previous reports that K1431M reduces TRIO GEF1 activity (Supp. Fig. 1A,B) (20, 24, 29). We hypothesized that homeostatic compensation in +/K1431M mice may alter expression of other RhoGEFs and GAPs. Indeed, levels of the Rac1 GEF Tiam1 were increased in both +/K1431M and +/M2145T P42 cortical lysates, while VAV2 levels were increased in +/M2145T P42 lysates (Fig. 6I-L). Levels of the Trio paralog Kalirin (58) were unaffected in the Trio variant mice at P42. Together, our proteomic analyses suggest that presynaptic functions are altered in +/K1431M and +/M2145T mice and may be driven by abnormal levels of crucial presynaptic regulatory proteins and changes in Rac1 and RhoA activity.

NSC23766, a Rac1-specific inhibitor, rescues neurotransmitter release in Trio +/K1431M heterozygotes

Rac1 negatively regulates synaptic vesicle replenishment and synaptic strength in excitatory synapses (15, 44). Rac1 activity was increased while Stx1a levels were decreased in +/K1431M cortex, in association with reduced synaptic strength and deficient vesicle replenishment in L2/3-L5 cortical synapses. We tested if the acute application of the Rac1 inhibitor NSC23766 (NSC) could rescue these deficits. Treatment of +/K1431M slices acutely with NSC normalized PPR in M1 L5 PNs at all ISIs to a WT pattern, suggesting a rescue of the decreased Pr (Fig. 5A,B). Application of NSC to +/K1431M slices also decreased time to depletion of signal upon HFS (Fig. 5C,D) and significantly increased Pr without effect on RRP size (Fig. 5E,F), consistent with prior work showing that changes in Rac1 signaling does not impact the RRP (15, 44, 59). Finally, we tested if acute Rac1 inhibition impacts the rate of recovery of the RRP following HFS train stimulation. The recovery curve exhibited improvement at later intervals with full recovery at 18s interval under NSC treatment, but while the recovery rate (τR) measured was improved, it remained significantly slower compared to WT (Fig. 5G). Overall, we demonstrate that presynaptic Trio GEF1-dependent Rac1 signaling is crucial for maintaining synchronous glutamate Pr and SV replenishment at cortical L2/3-L5 synapses.

Discussion

Large-scale genetic studies show significant overlap in risk genes for ASD, SCZ, and BPD, many converging on synaptic proteins (60–68). However, how variants in a single gene contribute to different NDDs remains a major unresolved question. Our study of mice heterozygous for Trio variants associated with ASD (+/K1431M), SCZ (+/K1918X), and BPD (+/M2145T) revealed that the lesions differentially affect Trio protein levels or GEF activity, yielding overlapping but distinct behavioral, neuroanatomical, and synaptic phenotypes. Our findings extend prior work demonstrating that Trio is critical for postsynaptic signaling and synaptic plasticity. We also demonstrate for the first time in mouse model that Trio is critical for glutamate release and synaptic vesicle recycling, and that NDD-associated variants differentially impact these pre- and post-synaptic roles.

Heterozygosity for Trio variants in mice yields phenotypes similar to those observed in NDDs

Individuals with mutations in TRIO present with a range of neurodevelopmental disorder-associated clinical features, including varying degrees of intellectual disability, altered head size, skeletal and facial features, and behavioral abnormalities (23, 25–28, 33, 69). Patients with missense or truncating variants in TRIO that reduce GEF1 activity have mild developmental delay and microcephaly (25, 26, 28, 33). Similarly, we found that heterozygosity for the GEF1-deficient K1431M missense or the K1918X nonsense variants significantly reduced brain weight and/or head size compared to WT mice, along with multiple behavioral impairments. Notably, while both showed impaired motor coordination and learning, only mice bearing the ASD-associated K1431M allele exhibited social interaction deficits.

Both +/K1431M and +/K1918X adult mice had reduced brain-to-body weight ratios compared to WT, but these were driven by different factors. The smaller brain size in +/K1918X mice was associated with a reduction in neuropil and reduced cortical thickness, similar to mice bearing excitatory neuron-specific ablation of one Trio allele (19) and paralleling the reduced gray matter volume and cortical thickness in patients with schizophrenia (70–74). Meanwhile, +/K1431M mice exhibited an overall increase in body weight leading to relatively decreased head width- and brain-to-body weight ratios in these mice. Adult +/K1431M mice had no change in neuropil or cross-sectional brain area, consistent with a prior study describing normal brain size in +/K1431M and K1431M/K1431M mice at E14.5 (20). Rac1 mediates glucose-stimulated insulin secretion from pancreatic islet beta-cells (75–80), which may explain how chronic alterations in Rac1 activity contribute to weight changes in Trio +/K1431M and +/K1918X mice. Most studies of Trio variants have focused on neuronal effects, but expression of Trio in other tissues could explain the increased body weight in these mice, as well as the musculoskeletal abnormalities associated with TRIO variation in humans (23, 25, 27, 32).

Trio variants differentially impact dendritic arbor structure

Rac1 and RhoA signaling is critical for dendrite development (10–13, 81). We found relatively subtle effects of Rac1/RhoA-altering Trio variants on cortical L5 PN dendrites. The reductions in dendritic arborization and length in +/K1918X neurons are consistent with reports of reduced gray matter volume and dendrite alterations in individuals with schizophrenia versus controls (82, 83), but were modest compared to NEX-Trio+/flmice lacking one copy of Trio in excitatory neurons (19). Reduced Trio function in other cell types, such as in inhibitory neurons or glia, or in neurons from other brain regions that project to cortical M1 L5 PNs, may ameliorate the phenotypes in excitatory neurons of +/K1918X relative to NEX-Trio+/fl mice. In addition, changes to L5 PN dendrites in Trio variant mice appeared to be regionally selective within the arbor: +/K1431M neurons had increased arborization only in proximal basal dendrites, while +/M2145T neurons had decreased arborization only in the most distal apical dendrites. These differences may reflect the differential spatiotemporal influence of Trio in regulating Rac1 versus RhoA activity. Alternatively, given our finding for presynaptic roles for Trio, these differences may reflect differential effects of the Trio variants on both excitatory and inhibitory afferent synaptic inputs which play critical roles in shaping the apical and basal dendrites of L5 PNs (84).

Trio variants impact brain Rho GTPase signaling

We show here that the K1431M variant significantly reduces TRIO GEF1 nucleotide exchange on Rac1 in vitro, consistent with previous reports (24, 29). Decreased Rac1 activity was observed in +/K1431M mice at embryonic day 14.5 in the ganglionic eminence, which is enriched for pre-migratory GABAergic interneurons (20). In contrast, we show an increase in active Rac1 levels in +/K1431M postnatal brains (at P0, P21, P42) and synaptosomes (at P42) from cortex, primarily composed of differentiated excitatory neurons. This increase aligns with observed phenotypes (e.g. reduced Pr) and reveals why reductions in brain volume, dendrites and spines, or AMPAR signaling anticipated from reduced Rac1 activity (85–89) were not observed in +/K1431M mice. We propose that the increased Rac1 activity we observed reflects homeostatic compensation in Rac1 regulation occurring between birth to adult ages, and identify changes in Rac1-specific GEFs, e.g. Tiam1 and Vav2, that may contribute to this compensation. Importantly, we did not find changes in Kalirin levels in the adult brain of these Trio mutant mice, suggesting that Kalirin does not compensate for loss of Trio GEF1 activity at this age.

We observed a significant reduction in active RhoA only in purified synaptosomes of +/M2145T brains, reflecting the synaptic compartment as a key locus of Trio function. In addition, despite +/K1918X mice having half the WT levels of Trio protein, we measured little to no change in active Rac1 and RhoA levels in +/K1918X brains. These findings are consistent with recent evidence that the spatiotemporally precise balance of Rac/Rho activity rather than absolute activity levels can mediate cytoskeletal rearrangements (7, 9, 90, 91). Alterations in the activity of additional potential TRIO substrates, such as RhoG (92), Cdc42 (6), and the neurodevelopmentally critical Rac3 (93–95) could also contribute phenotypes in these mice.

The NDD-associated Trio variants cause synaptic transmission, plasticity and excitatory/inhibitory imbalance

Overexpression of a TRIO K1431M variant with reduced GEF1 activity decreased AMPAR-mediated mESPC amplitudes in rat organotypic slices (21), while Rac1 activation increased AMPAR amplitudes by promoting synaptic AMPAR clustering (89). In +/K1431M mice we observed increased AMPAR current amplitudes without significant changes in NMDAR ESPC amplitudes in evoked and miniature EPSCs recordings, consistent with the increase in active Rac1 levels measured in synaptosomes of adult +/K1431M mice. AMPAR EPSC amplitudes were also increased in L5 PNs of +/K1918X mice, which have decreased Rac1 activity at P0 and normalized levels of Rac1 by P21 and P42 compared to WT. This temporal increase of Rac1 activity could be enough to drive a subtle but significant increase in AMPAR tone in +/K1918X.

mIPSC frequencies were decreased in +/K1431M and +/M2145T L5 PNs, while mIPSC amplitude was increased in +/K1918X slices. Sun et al. noted a similar deficit in inhibitory function in +/K1431M prefrontal cortex correlated with reduced interneuron migration to this region, including parvalbumin positive (PV+) neurons (20). In contrast, we did not observe reduced PV+ interneuron numbers in the motor cortex in any Trio variant mice, suggesting that Trio variants may impact the number of inhibitory synapses or transmission properties rather than solely interneuron migration. Overall, heterozygosity for Trio variants dysregulates excitatory and inhibitory synaptic transmission in different patterns, resulting in E/I imbalance, a known driver of NDD phenotypes.

Trio-deficient excitatory neurons are unable to undergo long-term potentiation (LTP) in mouse brain slices (19), which is crucial for working memory in mammals. During LTP ability of Rac1 suggested to be transiently activated and deactivated to regulate AMPAR (96–98). Both +/K1431M and +/K1918X L5 PNs exhibited reduced LTP induction and maintenance. The increased AMPAR resulting from already elevated Rac1 activity (+/K1431M mice) or the inability of reduced levels Trio levels to activate Rac1 (+/K1918X mice) may preclude LTP in these mice. +/M2145T mice showed a striking increase in the induction and maintenance of LTP, correlating with an increased glutamate Pr and SV pool size. The function of RhoA function in plasticity is unknown, but the decrease in RhoA activity measured in cortical synaptosomes may underlie the increase in LTP.

Trio GEF1 and GEF2-deficient variants lead to opposing defects in synaptic release of glutamate

In neuroendocrine, pancreatic beta, and mast cells, Trio GEF1 activity, Rac1 and RhoA are required for regulation of exocytosis (99–103). Recent work suggest that Trio GEF1 can act through Rac1 to regulate presynaptic processes – Rac1 colocalizes with SVs in the axonal boutons to negatively regulate action potential-dependent (synchronous) glutamate Pr and SV replenishment (15, 44, 59, 104). Using the Rac1 inhibitor NSC, we demonstrate that elevated Rac1 activity, possibly due to Tiam1 upregulation, underlies the reduction in synchronous glutamate Pr and SV replenishment without affecting RRP in +/K1431M mice.

Recent work on Rac1 knockout models shows conflicting results regarding their impact on spontaneous (action potential-independent) glutamate Pr, which is critical in synaptogenesis and plasticity (105, 106). Knockout of Rac1 at the Drosophila neuromuscular junction or in cultured mouse hippocampal neurons does not affect the frequency of spontaneous release (15, 59). However, conditional knock out of Rac1 in the Calyx of Held increased spontaneous Pr (44), aligning with the hypothesis that Rac1 activity negatively regulates neurotransmitter release by increasing assembly of actin filaments at the AZ that impede SV fusion. A significant decrease in NMDAR mEPSC frequency in +/K1431M mice suggests defects in spontaneous glutamate Pr in cortical synapses, consistent with increased Rac1 activity, while the increase in miniature AMPAR amplitudes could mask the decrease in AMPAR frequency. This relative impact of Rac1 on spontaneous release may reflect functional differences between these diverse synapse types.

RhoA activity is significantly reduced in cortical synaptosomes of +/M2145T mice, and this is associated with increased RRP size. While a specific function for RhoA in regulating presynaptic release is currently unclear, levels of Munc18, Syp, and Syt3, are all increased in +/M2145T mice and may contribute to the enhanced Pr and altered SV cycling. Taken together, we demonstrate that Trio GEF1 and GEF2 activities play crucial, distinct roles in the presynaptic SV cycle.

Conclusions

TRIO is a risk gene for several NDDs with different patterns of variants observed in different disorders. We show here that variants in Trio that lead to impaired Trio levels or GEF function cause both shared and distinct defects in behavior, neuroanatomy, and synaptic function that reflect variant-specific NDD clinical phenotypes. Our data also demonstrate, for the first time, the differential impact of distinct Trio lesions on glutamate Pr and SVs replenishment, along with alterations in presynaptic release machinery that contribute to these deficits. We propose that distinct Trio variants act via diverse mechanisms to disrupt normal brain function at the synaptic, circuit, and behavior level.

Acknowledgements

We are grateful to Xianyun Ye, Suxia Bai, Andrew Boulton, Chris Kaliszewski, and Xiao-Yuan Li for expert technical support and Bruce Herring, Dick Mains and Betty Eipper for formative discussions.

This work was supported by AHA Postdoctoral Training Grant 20POST35210428 (Y.I.), NIH Medical Scientist Training Program Training Grant T32GM136651 (A.T.J), NIH grants R56MH122449 (A.J.K.), R01MH133562 (A.J.K.), R01MH132685 (A.J.K.) and a Pilot Award from the Simons Foundation.

Additional information

Author contribution

YI, ATJ and AJK designed the study and wrote the manuscript. ATJ, TN, and AJK designed alleles and CRISPR strategy and AJK screened pups for germline transmission. TN and the Yale Genome Editing Center generated Trio CRISPR alleles. YI designed, conducted and analyzed all electrophysiology experiments. ATJ conducted dendrite reconstructions, dendritic spine analysis, measured brain and body weight, isolated crude synaptosomes and performed GLISAs. SF performed and analyzed all behavioral tests. ATJ collected samples; CMR, KN, and SAM measured protein levels in brain tissue. EEC perfused animals, CAG provided support, and ATJ and YI analyzed images from electron microscopy experiments. YI and ATJ conducted immunoblots and analysis. ATJ conducted Nissl stain and fluorescent immunohistochemistry; ATJ and MJV performed IHC analysis. MGC conducted GEF assays.

Conflict of interest

Authors declare no conflict of interest.