Introduction

Intellectual disability (ID) affects approximately 1–2% of the global population, posing substantial challenges to affected individuals, families, and healthcare systems1. Characterized by deficits in intellectual functioning and adaptive behaviors, including social skills and stress adaptation, ID results from a broad range of genetic mutations, many of which have been identified through advances in genomic technologies2. Notably, autosomal recessive ID (ARID), predominantly reported in Middle Eastern populations3, accounts for about 10% of ID cases in outbred populations4. Despite this, it remains less studied than autosomal dominant or X-linked forms due to its relative rarity in diverse outbred populations and genetic heterogeneity, which complicates sample collection and gene identification.

Tumor Suppressor Candidate 3 (TUSC3) is one of the genes implicated in ARID, with pathogenic variants identified in affected individuals5,6,7,8,9. TUSC3 is an endoplasmic reticulum (ER)-localized transmembrane protein, traditionally associated with N-glycosylation8. Furthermore, Zhou and Clapham identified that TUSC3 is required for cellular magnesium (Mg²⁺) uptake in mammalian cells and vertebrate embryonic development10. While this study identifies TUSC3 as essential for maintaining cellular Mg²⁺ levels10, its specific role in ER Mg²⁺ uptake and its contribution to neuronal dysfunction and ARID via ER stress remain poorly defined.

Mg²⁺ is essential for numerous enzymatic reactions, and its intracellular homeostasis is critical for neuronal function11,12. ER Mg²⁺ is crucial for ATP stabilization, chaperone activity, and protein folding. While direct evidence linking ER Mg²⁺ levels to neuronal dysfunction is lacking, studies have shown that culturing cells in low Mg²⁺ conditions triggers ER stress and activates the unfolded protein response13. Since chronic ER stress is implicated in various neurodevelopmental and neurodegenerative disorders14, a potential link between ER Mg²⁺ homeostasis and cognitive dysfunction remains to be elucidated. However, how TUSC3 regulates ER Mg²⁺ and its role in ARID remain unexplored.

To address this gap, we generated a TUSC3 knockout (KO) mouse model to investigate its role in ARID. We demonstrate that TUSC3 regulates ER Mg²⁺ homeostasis via ERMA, an ER magnesium transporter. TUSC3 deficiency disrupts ER Mg²⁺ uptake, triggers ER stress, and impairs global protein translation, leading to neuronal dysfunction and cell death. These deficits result in ID-like behaviors, including impairments in cognitive function, adaptive behaviors. Notably, magnesium supplementation restored ER Mg²⁺ balance, alleviated ER stress, and rescued cognitive and behavioral deficits of already developed ARID in TUSC3 KO mice, identifying magnesium supplementation as a potential therapy for TUSC3-related ID.

Results

TUSC3 is required for ER Mg²⁺ uptake in multiple cell types, including ARID patient-derived fibroblasts

To elucidate the molecular mechanism by which TUSC3 loss contributes to ARID, we first examined its known role as a non-catalytic subunit of the oligosaccharyltransferase complex involved in N-glycosylation15. We analyzed CNNM3, a Mg²⁺ uptake transporter known to undergo N-glycosylation16,17, in SH-SY5Y TUSC3 knockdown (KD) cells. While the N73A mutant of CNNM3 showed a clear band shift, CNNM3 WT exhibited no such shift in TUSC3 KD cells, indicating that its glycosylation was not affected (Supplementary Fig. 1a). We further analyzed glycosylation status of endogenous proteins, including NCAM1, N-cadherin, and GluA2, in primary hippocampal neurons and whole-brain lysates from our TUSC3 KO mice (characterized later in this study). Following PNGase F treatment, all proteins analyzed showed the expected band shift on the western blot, with no differences between WT and TUSC3 KO mice (Supplementary Fig. 1b, c). These results suggest that a predicted role of TUSC3 in N-glycosylation might not be linked to ARID, consistent with a previous report showing no significant changes in N-glycosylated proteins in blood samples from ARID patients8.

Next, given that TUSC3 regulates Mg²⁺ uptake in mammalian cells10, we examined this possibility by using an ER-targeting Mg²⁺ sensor, Mag-FRETER18,19, to monitor ER Mg²⁺ dynamics. Mag-FRETER contains an ER retention sequence (KDEL) and forms a compact structure upon binding Mg²⁺, generating a detectable FRET signal based on YFP (Fig. 1a)19,20. We confirmed ER localization of Mag-FRETER in SH-SY5Y cells (Supplementary Fig. 2a). Consistent with the previous findings19, L-lactate treatment triggered a rapid release of Mg²⁺ from the ER into the cytosol, followed by a reuptake phase where ER Mg²⁺ levels were restored (Fig. 1b). Real-time monitoring of ER Mg²⁺ dynamics revealed a sharp decrease at 21 s in both SH-SY5Y control and TUSC3 KD cells upon L-lactate treatment (Fig. 1b), accompanied by a concurrent rise in cytosolic Mg²⁺ levels (Supplementary Fig. 2b–d). The rate of Mg²⁺ release from the ER to the cytosol was comparable between SH-SY5Y control and TUSC3 KD cells (Fig. 1c). However, during the subsequent reuptake phase, SH-SY5Y TUSC3 KD cells exhibited significantly impaired Mg²⁺ uptake compared with control cells (Fig. 1d). Notably, reconstitution of SH-SY5Y TUSC3 KD cells with TUSC3-mRFP restored ER Mg²⁺ uptake to control levels (Fig. 1d, e). These findings suggest that while TUSC3 is dispensable for Mg²⁺ release from the ER, it is crucial for efficient Mg²⁺ uptake.

Fig. 1: TUSC3 is required for Mg²⁺ uptake into the ER.

a Schematic of the ER-targeted KDEL-Mag-FRETER system to assess ER Mg²⁺ dynamics. Created with BioRender.com. b–d SH-SY5Y shControl and shTUSC3#3-1 cells were transfected with Mag-FRETER for 48 h and analyzed for ER Mg²⁺ uptake following L-lactate treatment (10 mM). Circle 1 indicates ER Mg²⁺ release, and Circle 2 represents ER Mg²⁺ refilling (b, left). Representative confocal images at 21 s (b, right). Quantification of normalized ER Mg²⁺ release (Circle 1) (c). Quantification of normalized ER Mg²⁺ uptake (left) and uptake rate (right) (Circle 2) (n = 6, 7, 7 cells per group) (d). e Representative confocal images of SH-SY5Y shTUSC3#3-1 cells co-expressing Mag-FRETER and TUSC3-mRFP. Scale bar: 10 μm. f–i Non-ID and ID fibroblasts were transfected with Mag-FRETER alone (f) or co-transfected with TUSC3-mRFP (g) for 48 h and analyzed for ER Mg²⁺ uptake following L-lactate treatment (10 mM). Quantification of normalized ER Mg²⁺ release (Circle 1) (h), ER Mg²⁺ uptake (i, left), and uptake rate (i, right) (Circle 2) (n = 7 cells per group). Scale bar: 10 μm. j, k SH-SY5Y shControl and shTUSC3#3-1 cells were transfected with Mag-FRETER for 48 h, and supplemented with acute MgCl₂ treatment (10 mM, 200 s) to assess ER Mg²⁺ uptake (j), quantification of ER Mg²⁺ levels at 0 s and 200 s (n = 12, 14 cells per group) (k). l, m WT and TUSC3 KO primary cortical neurons were transfected with Mag-FRETER for 48 h and analyzed for ER Mg²⁺ uptake following acute MgCl₂ treatment (10 mM, 200 s) (l), quantification of ER Mg²⁺ levels at 0 s and 200 s (n = 10, 9 cells per group) (m). n, o Non-ID and ID fibroblasts were transfected with Mag-FRETER for 48 h and supplemented with acute MgCl₂ treatment (10 mM, 200 s) to assess ER Mg²⁺ uptake (n), quantification of ER Mg²⁺ levels at 0 s and 200 s (n = 10, 7, 10 cells per group) (o). One-way ANOVA followed by Tukey’s post hoc multiple comparison test. All data are presented as mean ± S.E.M. Source data are provided as a Source data file.

To assess the relevance of our findings to human disease, we analyzed fibroblasts derived from two patients with ARID harboring biallelic mutations in TUSC3 (ID-fibroblasts)7. Using the Mag-FRETER, we found that Mg²⁺ release from the ER to the cytosol in response to L-lactate was comparable between ID- and Non-ID fibroblasts (Fig. 1f–h). Similar to SH-SY5Y cells, however, Mg²⁺ reuptake into the ER was significantly impaired in both patient fibroblasts compared with non-ID fibroblasts (Fig. 1g, i). As expected, reintroducing TUSC3-mRFP into ID-fibroblasts rescued ER Mg²⁺ uptake deficits (Fig. 1g, i). This analysis was extended to other cell types, including mouse embryonic fibroblasts (MEFs) and the colon cancer cell line HCT116, where TUSC3 KO similarly resulted in reduced ER Mg²⁺ uptake (Supplementary Fig. 2e–j). These results underscore the essential and conserved role of TUSC3 in maintaining ER Mg²⁺ homeostasis across multiple cell types, including cells derived from ARID patients.

Magnesium supplementation rescues ER Mg²⁺ levels in TUSC3-deficient cells

When we measured basal ER Mg²⁺ levels, they were significantly reduced in TUSC3-deficient cells compared with WT cells (Fig. 1j, l, n). To evaluate the immediate capacity of ER Mg²⁺ uptake, we applied an acute MgCl₂ pulse (10 mM for ~200 s) to cells transfected with Mag-FRETER. Intriguingly, we found that ER Mg²⁺ levels rapidly increased in SH-SY5Y control cells, and by 200 s, TUSC3 KD cells also showed near-complete recovery to basal control levels, albeit with lower and slower Mg²⁺ uptake (Fig. 1j, k). A similar pattern was observed in TUSC3 KO primary cortical neurons, which exhibited reduced basal ER Mg²⁺ levels relative to WT neurons, but showed partial recovery upon MgCl₂ treatment (Fig. 1l, m). Likewise, ID-fibroblasts exhibited lower ER Mg²⁺ levels than non-ID fibroblasts, and their ER Mg²⁺ was restored by MgCl₂ treatment, albeit with delayed uptake kinetics (Fig. 1n, o). These findings suggest that, while TUSC3-deficient cells have reduced basal ER Mg²⁺ levels, supplementation with high concentrations of MgCl₂ can effectively restore ER Mg²⁺ homeostasis.

TUSC3 interacts with ER magnesium transporter ERMA to facilitate Mg²⁺ uptake

TUSC3 lacks a canonical Mg²⁺ transport domain and GxN motifs21,22, suggesting that it may function by modulating another transporter rather than directly mediating Mg²⁺ transport. ERMA (ER magnesium transporter A; formerly TMEM94) is a P-type ATPase recently identified as a key mediator of ER Mg²⁺ uptake, and P-type ATPases frequently require accessory subunits for optimal activity23. We therefore hypothesized that TUSC3 functions as a regulatory component of ERMA. To test this, ER Mg²⁺ release and uptake were examined in SH-SY5Y TUSC3 KD cells overexpressing ERMA-DsRed (Fig. 2a–d). ERMA-DsRed overexpression enhanced Mg²⁺ uptake in SH-SY5Y control cells, but this effect was attenuated in TUSC3 KD cells, failing to restore ER Mg²⁺ uptake to control levels (Fig. 2b–d).

Fig. 2: TUSC3 regulates ER Mg²⁺ uptake by interacting with ERMA.

a–d SH-SY5Y shControl and shTUSC3#3-1 cells were co-transfected with Mag-FRETER and ERMA-DsRed for 48 h and analyzed using confocal microscopy (a). Scale bar: 10 μm. Representative traces showing ER Mg²⁺ uptake following L-lactate (10 mM) treatment (b). Quantification of normalized ER Mg²⁺ release (Circle 1) (c). Quantification of normalized ER Mg²⁺ uptake (left) and uptake rate (right) (Circle 2) (n = 9 cells per group) (d). e, f Co-immunoprecipitation from whole brain lysates of WT mice showing endogenous interaction between TUSC3 and ERMA. Immunoprecipitation with anti-ERMA antibody pulled down TUSC3 (e), and reciprocal IP with anti-TUSC3 antibody confirmed the presence of ERMA (f). L.C.: light chain of immunoglobulin. g Schematic representation of TUSC3-mRFP deletion mutants. h Immunoprecipitation assays between ERMA-FLAG and various TUSC3-mRFP deletion mutants. i–k SH-SY5Y shControl and shTUSC3#3-1 cells were transfected with Mag-FRETER and indicated TUSC3-mRFP deletion mutants for 48 h and analyzed for ER Mg²⁺ release and uptake following L-lactate (10 mM) treatment (i). Quantification of ER Mg²⁺ release (Circle 1) (j) and ER Mg²⁺ uptake (left) and uptake rate (right) (Circle 2) (k) (n = 10, 9, 9, 6, 9, 4, 5 cells per group). One-way ANOVA followed by Tukey’s or Dunnett’s (k) post-hoc multiple comparison test. All data are presented as mean ± S.E.M. Source data are provided as a Source data file.

To determine whether TUSC3 interacts with ERMA, we performed co-immunoprecipitation (co-IP) assays. Co-IP assays clearly showed their interaction in HEK293T cells overexpressing TUSC3-mRFP and ERMA-FLAG (Supplementary Fig. 2k, l). Consistent with this, co-IP assays using whole brain lysates revealed the interaction between endogenous TUSC3 and ERMA (Fig. 2e, f). To map the TUSC3 domain critical for this interaction, deletion (Δ) mutants were generated based on TUSC3 topology (Fig. 2g). Co-IP assays revealed that Δ197–242 (TM1 and TM2) significantly reduced TUSC3-ERMA binding, whereas Δ222**–**297 (TM2 and TM3) had minimal effect (Fig. 2h), indicating that the TM1 domain is essential for the interaction. To evaluate the functional relevance of this interaction, SH-SY5Y TUSC3 KD cells were reconstituted with either TUSC3 WT or deletion mutants. Unlike WT and other deletion mutants, TUSC3 Δ197–242 failed to rescue ER Mg²⁺ levels (Fig. 2i–k). Thus, TUSC3 promotes ER Mg²⁺ uptake by interacting with ERMA, and this interaction critically depends on the TM1 domain.

TUSC3 deficiency evokes ER stress, translation impairment, and cell death via Mg² dysregulation

To determine how ER Mg²⁺ depletion in TUSC3-deficient cells contributes to ID phenotypes, we examined ER stress responses. Since Mg²⁺ is essential for protein folding and chaperone activity in the ER, its depletion can lead to accumulation of misfolded protein and activation of the unfolded protein response13. In the hippocampus and striatum of TUSC3 KO mice, GRP78/BiP was significantly upregulated, and activation of the PERK–eIF2α axis, including p-eIF2α and CHOP, was markedly increased (Fig. 3a–r). In contrast, the IRE1–XBP1s and ATF6 branches were less affected (Fig. 3f, i, o, r), suggesting that the PERK–eIF2α axis is the primary response of TUSC3 deficiency. CREB activity, a key regulator of learning and memory, and PI3K/AKT signaling24, its upstream modulator, were both reduced (Fig. 3g, h, p, q), linking ER stress to cognitive deficits in TUSC3 KO mice.

Fig. 3: TUSC3 loss induces ER stress via PERK-eIF2α pathway, leading to impaired protein translation and cell death.

a Representative western blot of ER stress proteins in the hippocampus of 4-month-old female WT and TUSC3 KO mice. b–i Quantification of GRP78 (b), p-PERK (T982) (c), p-eIF2α (S51) (d), CHOP (e), XBP1s (f), p-AKT (S473) (g), p-CREB (S133) (h), and ATF6 (i) normalized to β-actin (n = 3 mice per group). j Representative western blot analysis of ER stress proteins in the striatum of 4-month-old female WT and TUSC3 KO mice (n = 3 per group). k–r Quantification of GRP78 (k), p-PERK (T982) (l), p-eIF2α (S51) (m), CHOP (n), XBP1s (o), p-AKT (S473) (p), p-CREB (S133) (q), and ATF6 (r) normalized to β-actin (n = 3 mice per group). s Non-ID and ID1/ID2 fibroblasts were chronically treated with MgT (400 μM) for 48 h and analyzed by western blotting (left). Quantification of p-eIF2α (S51) levels normalized to eIF2α (right) (n = 3 independent biological replicates). t Non-ID and ID-derived fibroblasts were pretreated with or without MgT (400 μM) for 24 h, followed by puromycin (5 μg/ml) treatment for 2 h. Representative western blot using an anti-puromycin antibody (left), and puromycin incorporation was quantified and normalized to β-actin (right) (n = 3 independent biological replicates). u SH-SY5Y shControl and shTUSC3#3-1 cells were pretreated with MgT (400 μM) for 12 h, then treated with either DMSO, thapsigargin (2 μM), tunicamycin (2 μg/ml), A23187 (2 μM), or etoposide (25 μM) for 24 h. Cells were stained with propidium iodide and Calcein-AM, and double-positive cells were quantified and normalized to Calcein-AM-positive cells. Data represent three independent biological experiments (n = 3). The exact number of cells analyzed per condition is provided in the Source Data file. v Representative western blot of primary cortical neurons (DIV7) treated with vehicle or 4-PBA (10 mM) for 24 h (left). Quantification of indicated protein levels normalized to TUBA (right) (n = 3 independent biological replicates). Two-tailed unpaired t-test (b**–i, and k–**r); one-way ANOVA followed by Tukey’s post hoc multiple comparison test (s, t, v); two-way ANOVA followed by Tukey’s multiple comparison test (u). Data are presented as mean ± S.E.M. or S.D. (t, u). Source data are provided as a Source data file.

ER stress inhibits protein translation via the PERK–eIF2α axis, a mechanism implicated in neurodevelopmental disorders such as Down syndrome25,26. Using the Surface Sensing of Translation (SUnSET) assay27, we found that global protein translation was significantly reduced in TUSC3 KO MEFs, coinciding with the increase of GRP78, p-PERK, p-eIF2α, and XBP1s levels (Supplementary Fig. 3a–c). Similar findings were observed in ID-fibroblasts (Supplementary Fig. 3d, e). Notably, chronic magnesium supplementation with magnesium-L-threonate (MgT, 400 μM, 48 h), which has ability to cross the blood-brain barrier12, restored protein translation and reduced p-eIF2α levels in ID-fibroblasts (Fig. 3s, t), suggesting that Mg²⁺ supplementation mitigates both ER stress and translation impairment.

Since CHOP is a pro-apoptotic factor in ER stress response28, we tested to confirm whether TUSC3 deficiency increases cell death susceptibility. Compared with control cells, SH-SY5Y TUSC3 KD cells were more vulnerable to ER stressors (thapsigargin, tunicamycin, A23187), but not to the DNA-damaging agent etoposide (Fig. 3u, Supplementary Fig. 3f). Notably, magnesium supplementation effectively mitigated ER stress-induced cell death (Fig. 3u). To further assess the role of ER stress in TUSC3-related neuronal dysfunction, TUSC3 KO primary cortical neurons were treated with 4-phenylbutyrate (4-PBA), a chemical chaperone that alleviates ER stress29. Intriguingly, 4-PBA treatment markedly restored PSD-93, PSD-95, and GluA1 levels in TUSC3 KO neurons (Fig. 3v). These findings highlight that ER Mg²⁺- stress regulates synaptic activity in TUSC3-deficient cells, linking ER Mg²⁺ homeostasis to cognitive function. Targeting ER Mg²⁺-stress may serve as a therapeutic strategy for TUSC3-related ID.

TUSC3 KO mice display abnormal embryonic development

To assess the in vivo relevance of our cellular findings, we next generated a TUSC3 KO mouse model using CRISPR-Cas9-mediated targeting of Exon 2 in the Tusc3 gene, leading to a frameshift mutation (see Methods and Supplementary Fig. 4a for detailed methodology). Genotyping and western blot analysis confirmed the successful generation of TUSC3 KO mice (Supplementary Fig. 4b). Heterozygous Tusc3 mice were bred to generate homozygous KO offspring. However, only 48 of 568 offspring (8.6%) were homozygous KO, significantly deviating from expected Mendelian ratios (Supplementary Fig. 4c). Further analysis revealed embryonic lethality, with resorption observed as early as embryonic day (E)11.5 (Supplementary Fig. 4d). Next, we assessed growth phenotypes in surviving TUSC3 KO embryos and mice, which exhibited significantly lower body weights compared with WT littermates (Supplementary Fig. 4e, f). These results indicate incomplete embryonic lethality in TUSC3 KO mice, with surviving offspring showing persistent growth deficits compared with WT mice.

TUSC3 KO mice display intellectual disability phenotypes

To investigate whether TUSC3 KO mice represent ID phenotypes, we conducted a series of behavioral tests that assess learning, memory, stress-coping ability, and social interactions, which are hallmark features of ID2. At 4 weeks of age, TUSC3 KO mice showed no significant differences compared with WT mice in the Y-maze, novel object recognition (NOR), and social interaction tests, but exhibited a significant reduction in immobility time in the tail suspension test (TST) (Supplementary Fig. 5a–e), indicating early stress-coping deficits. At 3 months of age, however, cognitive, social, and stress-related impairments emerged in TUSC3 KO mice. Three-month-old TUSC3 KO mice exhibited significantly reduced spontaneous alternation rates in the Y-maze (Fig. 4a, b). In the NOR test, TUSC3 KO mice displayed impaired novel object discrimination (Fig. 4c–e). In the passive avoidance test, TUSC3 KO mice demonstrated significantly shorter latency to re-enter the dark chamber (Supplementary Fig. 6a). Collectively, these results indicate that TUSC3 deficiency leads to significant impairments in spatial working memory, recognition memory, and associative memory, underscoring a broad disruption in cognitive function.

Fig. 4: TUSC3 KO mice exhibit intellectual disability phenotypes.

a, b Schematic of the Y-maze test (a) and spatial memory performance in WT and TUSC3 KO mice (b). Created with BioRender.com. Schematic of the novel object recognition (NOR) test (c) and representative tracking plots in the open field in the NOR test (d). Created with BioRender.com. e The NOR test (discrimination index) for WT and TUSC3 KO mice. f, g Schematic of the forced swim test (FST), consisting of a 120 s adaptation phase followed by 240 s of immobility measurement (f) and immobility time in the FST, assessing stress-coping ability in WT and TUSC3 KO mice (g). Created with BioRender.com. h, i Schematic of the tail suspension test (TST), with immobility recorded over 360 s (h) and immobility time in the TST for WT and TUSC3 KO mice (i). Created with BioRender.com. j, k Schematic of three-chambered social interaction test (j) and representative tracking plots from the social interaction test, showing times spent in each chamber (k). Created with BioRender.com. l, m Social interaction preference in TUSC3 KO mice, assessed by time spent with the novel mouse (l) and sniffing time (m). n–q Schematic of three-chambered social preference test (n) and representative tracking plots of WT and TUSC3 KO mice in social preference test (o). Social preference for the novel mouse of WT and TUSC3 KO mice in the social preference test (p). Sniffing time data confirms this lack of preference (q). Created with BioRender.com. In all behavioral experiments, 4-month-old mice were used: WT: 9 (5 F, 4 M); KO: 10 (6 F, 4 M). Two-tailed unpaired t-test (b, e, g, i), Two-way ANOVA followed by Tukey’s post-hoc multiple comparison test (l, m, p, q). All data are presented as mean ± S.E.M. Source data are provided as a Source data file.

Since adaptive skills, including stress response, are frequently impaired in ID30, we next assessed stress-coping ability in TUSC3 KO mice using the forced swim test (FST) and TST. FST, traditionally used as a behavioral assay for depression31, is now widely recognized as a measure of stress adaptability32,33. In both FST and TST, TUSC3 KO mice exhibited markedly reduced immobility time, indicating deficits in stress-coping ability (Fig. 4f–i). Given that social behavior is another core domain affected in ID34, we next examined social interaction and recognition in TUSC3 KO mice. In the three-chamber social interaction test, TUSC3 KO mice spent significantly less time interacting with a novel mouse (Fig. 4j–m). Similarly, in the social novelty preference test, TUSC3 KO mice failed to show a preference for a novel mouse over a familiar one, highlighting deficits in social recognition and novelty response (Fig. 4n–q). Additionally, TUSC3 KO mice exhibited hyperactivity in the open field test, with increased locomotor activity but no significant changes in anxiety-like behavior (Supplementary Fig. 6b–d). They also displayed excessive self-grooming, suggestive of repetitive behaviors (Supplementary Fig. 6e, f). Together, TUSC3 KO mice exhibit progressive behavioral impairments characteristic of ID, with stress-coping deficits emerging by 1 month of age, followed by the onset of cognitive and social deficits at 3 months.

Synaptic structure and plasticity are impaired in TUSC3 KO mice

Abnormal synaptic signaling, impaired plasticity, and changes in spine morphology are key features of various psychiatric and neurological disorders35. Synaptic dysfunction is commonly observed in individuals with ID36. We thus sought to examine levels of scaffold proteins (PSD-93, PSD-95) and AMPA receptor subunit GluA1, which are critical components of synapse structure and function37. Western blot analysis revealed reduced levels of PSD-93, PSD-95, and GluA1 in TUSC3 KO mice compared with WT mice (Fig. 5a). Additionally, neuronal activity, measured by CREB phosphorylation38, was markedly reduced in TUSC3 KO mice. These molecular deficits were accompanied by a significant reduction in brain weight of TUSC3 KO mice compared with WT mice (Supplementary Fig. 7a). Consistently, levels of synaptic markers as well as phosphorylated CREB were also decreased by TUSC3 deficiency in primary hippocampal neurons and human neuroblastoma SH-SY5Y cells (Supplementary Fig. 7b–e). These results indicate that these reductions may contribute to the synaptic dysfunction observed in TUSC3 KO mice.

Fig. 5: Synaptic structure and plasticity are impaired in TUSC3 KO mice.

a Representative western blot of PSD-93, PSD-95, GluA1, and phosphorylated CREB (S133) in whole-brain lysates of 4-month-old WT and TUSC3 KO female mice (left). Quantification of indicated protein levels (right) (n = 3). b, c Representative confocal images of immunohistochemistry for TUSC3, synaptophysin (SYP), and MAP2 in the CA1 and CA3 hippocampal regions and the striatum of 4-month-old WT and TUSC3 KO female mice (b), and quantification of SYP intensity normalized to MAP2 (c) (n = 3). Scale bar: 50 μm. d, e Representative confocal images of immunohistochemistry for TUSC3, GluA1, and MAP2 in the CA1 and CA3 hippocampal regions and the striatum of 4-month-old WT and TUSC3 KO female mice (d), and quantification of GluA1 intensity normalized to MAP2 (e) (n = 3). Scale bar: 50 μm. f, g Representative dendritic images of primary hippocampal neurons transfected with EGFP (f). Scale bar: 50 μm. Sholl analysis of dendritic branching in WT and TUSC3 KO neurons (n = 20 cells per group) (g). h, i Representative confocal images of dendritic spines of primary hippocampal neurons (DIV 12) transfected with EGFP for 48 h (h). Violin plot quantifying dendritic spine density (spines per 10 μm) in WT and KO neurons (n = 20 cells per group) (i). Scale bar: 10 μm. j–l Representative traces of sEPSCs recorded from hippocampal slices of 2-month-old WT and TUSC3 KO mice (j), quantification of sEPSC amplitude (k), and frequency (l) in WT (n = 12 cells from 5 mice; 4 F, 1 M) and TUSC3 KO (n = 8 cells from 5 mice; 2 F, 3 M). m Representative traces (left) and quantification of the eEPSC amplitude (right) in WT (n = 8 cells from 5 mice; 4 F, 1 M), TUSC3 KO (n = 8 cells from 5 mice; 2 F, 3 M). n Representative traces (left) and EPSC-PPR (right) in WT (n = 8 cells from 5 mice; 3 F, 2 M), TUSC3 KO (n = 9 cells from 6 mice; 4 F, 2 M). Two-tailed unpaired t-test. Data are presented as mean ± S.E.M. (a, c, e, and k**–**n) or S.D. (g, i). Source data are provided as a Source data file.

To determine which brain regions are most vulnerable to TUSC3 deficiency, we examined the hippocampus and striatum—regions critically involved in cognitive function and adaptive behaviors such as social skills and stress coping39,40. Immunohistochemistry revealed a significant reduction in synaptophysin, a presynaptic marker41, in both the hippocampus CA1 region and the striatum of TUSC3 KO mice, whereas no change was detected in the CA3 region (Fig. 5b, c). GluA1 expression was similarly reduced in these regions (Fig. 5d, e). Next, we investigated abnormalities in synaptic structure in TUSC3 KO mice. Sholl analysis of primary hippocampal neurons showed a significant reduction in dendritic branching, as evidenced by fewer crossings in TUSC3 KO neurons compared with WT neurons (Fig. 5f, g). Moreover, dendritic spine density was markedly reduced in TUSC3 KO neurons (Fig. 5h, i).

We further investigated whether the structural abnormalities of the synapse in TUSC3 KO mice are associated with altered synaptic transmission using whole-cell voltage clamp recording. The spontaneous EPSC (sEPSC) amplitude in hippocampal CA1 neurons was significantly increased in TUSC3 KO mice, whereas sEPSC frequency remained unchanged (Fig. 5j–l). Evoked EPSC (eEPSC) amplitude was reduced, while the pair-pulse ratio (PPR) was unaltered (Fig. 5m, n). Although the unchanged PPR suggests intact presynaptic release probability under brief stimulation, FM1-43 dye unloading assays revealed delayed synaptic vesicle release in TUSC3 KO hippocampal neurons following sustained depolarization (Supplementary Fig. 8a–d), indicating that TUSC3 deficiency impairs synaptic vesicle release during prolonged stimulation and represents a form of presynaptic dysfunction not detected by standard PPR measurements.

We also performed similar recordings in CA3 pyramidal neurons. Unlike in CA1, sEPSC amplitude was unchanged in CA3; however, sEPSC frequency was significantly reduced in TUSC3 KO neurons compared with WT neurons (Supplementary Fig. 8e–g). This contrast shows region-dependent vulnerability of hippocampal circuits to TUSC3 loss, reflecting postsynaptic changes such as altered receptor expression or synapse strength in CA1 and presynaptic deficits of spontaneous neurotransmitter release in CA3. When we further analyzed miniature end-plate potentials (mEPPs) at the neuromuscular junction, no significant differences in amplitude or frequency were observed between WT and TUSC3 KO mice (Supplementary Fig. 8h–j), suggesting that synaptic transmission is largely preserved in peripheral synapses. Together, these results indicate that TUSC3 deficiency leads to structural synaptic abnormalities and region-specific disruptions in synaptic transmission, with both pre- and postsynaptic components being affected in a context-dependent manner. The lack of functional impairment at the neuromuscular junction further supports a selective vulnerability of central circuits to TUSC3 loss.

Magnesium supplementation ameliorates intellectual disability symptoms caused by TUSC3 deficiency

Given the role of ER Mg²⁺-stress and neuronal dysfunction in TUSC3-deficient cells, we investigated whether magnesium supplementation could rescue or prevent ID-like phenotypes in TUSC3 KO neurons and mice. In TUSC3 KO primary cortical neurons, we observed that chronic treatment of MgT (200 μM, 48 h) restored synaptic protein levels (PSD-93, PSD-95, GluA1) and CREB activity to WT levels (Supplementary Fig. 9a, b). To clarify whether MgT is able to restore ER Mg²⁺ levels, as previously demonstrated by MgCl₂ in vitro, we measured ER Mg²⁺ accumulation kinetics in SH-SY5Y TUSC3 KD cells using the Mag-FRETER assay. Whereas MgCl₂ treatment rapidly elevated ER Mg²⁺ levels within ~200 s, MgT exhibited a slower uptake profile during the 600 s imaging window. Intriguingly, ER Mg²⁺ levels reached comparable levels to MgCl₂ treatment approximately 8 h after MgT treatment (Supplementary Fig. 9c, d). These results indicate that both MgCl₂ and MgT can replenish ER Mg²⁺ levels, albeit with distinct kinetics.

We then assessed whether chronic magnesium supplementation via drinking water (910 mg/kg/day for 4 weeks) could reverse existing ID-like phenotypes in TUSC3 KO mice. Following treatment, a battery of behavioral tests was performed (Supplementary Fig. 10a). Hyperactivity in 5-month-old TUSC3 KO mice was significantly reduced to WT levels (Supplementary Fig. 10b). Cognitive performance in the Y-maze and NOR tests was markedly improved, reaching levels comparable to WT controls (Supplementary Fig. 10c, d). In the FST, TUSC3 KO mice showed impaired stress adaptation, as indicated by early removal and reduced immobility time. MgT treatment partially improved both survival and passive coping. Stress resilience in the TST was also significantly enhanced (Supplementary Fig. 10e, f). Social interaction and social preference behaviors were significantly improved following MgT supplementation (Supplementary Fig. 10g, h). At the molecular level, western blot analysis of the brain tissues further supported these findings. Synaptic protein levels in the striatum and h