A Farnesyltransferase Inhibitor Restores Cognitive Deficits in Tsc2^+/– Mice through Inhibition of Rheb1

Abstract

Tuberous sclerosis complex (TSC) is caused by mutations in Tsc1 or Tsc2, whose gene products inhibit the small G-protein Rheb1. Rheb1 activates mTORC1, which may cause refractory epilepsy, intellectual disability, and autism. The mTORC1 inhibitors have been used for TSC patients with intractable epilepsy. However, its effectiveness for cognitive symptoms remains unclear. We found a new signaling pathway for synapse formation through Rheb1 activation, but not mTORC1. Here, we show that treatment with the farnesyltransferase inhibitor lonafarnib increased unfarnesylated (inactive) Rheb1 levels and restored synaptic abnormalities in cultured Tsc2^+/− neurons, whereas rapamycin did not enhance spine synapse formation. Lonafarnib treatment also restored the plasticity-related Arc (activity-regulated cytoskeleton-associated protein) expression in cultured Tsc2^+/− neurons. Lonafarnib action was partly dependent on the Rheb1 reduction with syntenin. Oral administration of lonafarnib increased unfarnesylated protein levels without affecting mTORC1 and MAP (mitogen-activated protein (MAP)) kinase signaling, and restored dendritic spine morphology in the hippocampi of male Tsc2^+/− mice. In addition, lonafarnib treatment ameliorated contextual memory impairments and restored memory-related Arc expression in male Tsc2^+/− mice in vivo. Heterozygous Rheb1 knockout in male Tsc2^+/− mice reproduced the results observed with pharmacological treatment. These results suggest that the Rheb1 activation may be responsible for synaptic abnormalities and memory impairments in Tsc2^+/− mice, and its inhibition by lonafarnib could provide insight into potential treatment options for TSC-associated neuropsychiatric disorders.

SIGNIFICANCE STATEMENT Tuberous sclerosis complex (TSC) is an autosomal-dominant disease that causes neuropsychiatric symptoms, including intractable epilepsy, intellectual disability (ID) and autism. No pharmacological treatment for ID has been reported so far. To develop a pharmacological treatment for ID, we investigated the mechanism of TSC and found that Rheb1 activation is responsible for synaptic abnormalities in TSC neurons. To inhibit Rheb1 function, we used the farnesyltransferase inhibitor lonafarnib, because farnesylation of Rheb1 is required for its activation. Lonafarnib treatment increased inactive Rheb1 and recovered proper synapse formation and plasticity-related Arc (activity-regulated cytoskeleton-associated protein) expression in TSC neurons. Furthermore, in vivo lonafarnib treatment restored contextual memory and Arc induction in TSC mice. Together, Rheb1 inhibition by lonafarnib could provide insight into potential treatments for TSC-associated ID.

Introduction

Tuberous sclerosis complex (TSC) is caused by a heterozygous mutation in either the Tsc1 or Tsc2 gene, leading to the activation of mTORC1 signaling (Lipton and Sahin, 2014). TSC can involve any tissue or organ, but the CNS is one of the most commonly affected organs; its effects include epilepsy, subependymal giant cell astrocytoma (SEGA), and TSC-associated neuropsychiatric disorders (TANDs). TAND symptoms consist of cognitive, behavioral, and psychiatric disorders, such as intellectual disability, autism, and specific neuropsychological deficits (Curatolo et al., 2008, 2015). The role of the mammalian target of rapamycin (mTOR) pathway in brain development and function has been intensively studied both in vitro and in vivo, and mTOR inhibitors [everolimus and sirolimus (rapamycin)] are used to treat SEGA and angiomyolipoma in TSC patients. Thus far, however, everolimus has not been shown to significantly improve neurocognitive functions (memory, attention, executive function, and behavior) in TSC patients (Overwater et al., 2019).

Dendritic spines are the postsynaptic structure of excitatory synapses. The structural plasticity of dendritic spines is tightly correlated with synaptic function and memory. Indeed, a variety of cognitive disorders are accompanied by disease-related changes in dendritic spine morphology (Penzes et al., 2011). In rodent models of TSC, neuronal spine synapse formation was impaired; however, rapamycin did not restore mature spine formation (Yasuda et al., 2014). Thus, mTOR inhibition may not be effective for spine recovery in TSC neurons. This might be related to the reason why rapamycin failed to improve cognitive symptoms in TSC patients.

Tsc1 and Tsc2 form a complex that inhibits the small GTP-binding protein Rheb1, which activates mTORC1 in a GTP-dependent manner (Inoki et al., 2003). We previously showed that the activation of Rheb1, but not of mTORC1, causes synaptic abnormalities via the PDZ protein syntenin1, and that Rheb1 inhibition restored spine synapse formation in Tsc2^+/− neurons (Yasuda et al., 2014; Sugiura et al., 2015). Mutation in Rheb1 gene have been reported to cause intellectual disability in human patients (Reijnders et al., 2017), which is partly consistent with our observations. Thus, TSC–Rheb1 signaling may play an important role in dendritic spine and intellectual development.

The Rheb1 sequence terminates with CSVM, which corresponds to the CAAX box, which is farnesylated (A is an aliphatic amino acid; X is the C-terminal amino acid.). Farnesyltransferase (FTase) preferentially recognizes proteins whose sequences end in serine, methionine, glutamine, or alanine. This modification may localize Rheb1 to lysosomal membranes, followed by its activation (Clark et al., 1997). On the other hand, proteins with the CAAX box whose sequences end in leucine are also modified by geranylgeranyl protein transferase-1 (Casey and Seabra, 1996). K-Ras and N-Ras are not only farnesylated but also geranylgeranylated (Whyte et al., 1997). This geranylgeranylation may provide mutated Ras proteins with resistance to farnesyltransferase inhibitors (FTIs). However, Rheb1 is modified by only farnesylation and does not undergo geranylgeranylation (Clark et al., 1997), indicating that FTIs may efficiently inhibit activation of the Rheb1 protein rather than the Ras protein.

We demonstrated that treatment with FTIs, but not rapamycin, restored spine synapse formation in Tsc2^+/− rat neurons (Yasuda et al., 2014). Thus, Rheb1 could be a potential pharmacological target for the treatment of TAND symptoms. A direct inhibitor for Rheb1 was recently developed; however, its IC₅₀ is higher than those of FTIs (Mahoney et al., 2018), and the in vivo safety has not been characterized yet. Herein, we report the effectiveness of FTIs on cognitive deficits in a mouse model of TSC through the inhibition of Rheb1 and also show that this mechanism may confer a therapeutic advantage.

Materials and Methods

Mice.

Tsc2^+/− and Rheb1-floxed mice were generated previously (Kobayashi et al., 1999; Zou et al., 2011). CaMKII-Cre;Rheb1^flox/+ progeny were generated by crossing CaMKII-Cre mice with Rheb1^flox/+ mice. These progenies were then crossed with Tsc2^+/− mice to yield Tsc2^+/−;CaMKII-Cre;Rheb1^flox/+ mice. Only male animals were used in the behavioral experiments. Mice were of mixed genetic backgrounds (C57BL/6J and 129 Sv). All mice were group housed and maintained under a 12 h light/dark cycle. All animals were treated in accordance with the guidelines of the Animal Care and Use Committee of the Tokyo Metropolitan Institute of Medical Science and the Yamagata University School of Medicine.

Cell culture, transfection, and immunostaining.

HEK293 T cells were transfected with FLAG-Rheb1 and/or EGFP-syntenin expression vector. The cells were treated with 2 μm lonafarnib (catalog #S2797, Selleck) and 20 μm MG132 (catalog #M2400, LKT Labs). The primary culture and transfection of mouse hippocampal neurons (P 0–1) have been previously described (Yasuda et al., 2014; Sugiura et al., 2015). Cultured neurons were maintained in Neurobasal Medium (catalog #21103–049, Thermo Fisher Scientific) supplemented with B-27 supplement (catalog #17504–044, Thermo Fisher Scientific) and Antibiotic-Antimycotic (catalog #15240–062, Thermo Fisher Scientific). Neurons at 7 d in vitro (DIV7) were transfected with EGFP expression vector using Lipofectamine 2000 (catalog #11668–019, Thermo Fisher Scientific) according to the manufacturer instructions. Cultured hippocampal neurons were treated with 100 nm rapamycin (catalog #R0161, LKT Labs) or 2 μm lonafarnib (catalog # S2797, Selleck) for 7 d and fixed at DIV21.

Mouse cultured hippocampal neurons were fixed in 4% paraformaldehyde (PFA), blocked, permeabilized with BL solution (3% normal goat serum or 3% bovine serum albumin with 0.1% Triton X-100 in PBS), and incubated overnight at 4°C with one of the following primary antibodies: anti-GFP (chicken; 1:1000; catalog #AB16901, Merck Millipore) or anti-vesicular glutamate transporter 1 (VGluT1; guinea pig; 1:500; catalog #AB5905, Merck Millipore). Immunoreactivity was visualized using species-specific, fluorochrome-conjugated secondary antibodies. Spine and shaft synapse density analyses were performed as described previously (Yasuda et al., 2014; Sugiura et al., 2015). In brief, the relatively proximal region (∼150 µm from soma) of one dendrite was chosen randomly, and its length was measured. The number of VGluT1 puncta on the spine and shaft synapses of the dendritic segments was determined by counting the number of VGluT1 puncta that partially overlapped the GFP⁺ shaft or spines. Any VGluT1 puncta that did not touch a GFP⁺ dendrite were not included. In addition, individual dendritic spines were manually traced, and the maximum length and head width of each dendritic spine and spine density were measured. All quantification were conducted using MetaMorph software (Molecular Devices). For the cumulative plot graphs, we counted the number of spines below a certain length or width and calculated the percentage of the total number of measured spines.

Immunoblotting and antibodies.

HEK293 T cells, cultured hippocampal neurons, and mouse brains and livers were homogenized [40 mm Tris-HCl, pH 7.6, 1 mm EDTA, 1 mm EGTA, 0.1% Triton X-100, protease inhibitor cocktail tablet (Complete EDTA-free; catalog #11873580001, Roche) and 1 mm phenylmethanelsulfonyl fluoride (catalog #P7626, Sigma-Aldrich)] and centrifuged at 17,900 × g at 4°C. The resultant soluble and insoluble fractions were immunoblotted as described previously (Sugiura et al., 2015), and primary antibodies against FLAG (1:800; catalog #F3165, Sigma-Aldrich), Rheb1 [1:1000; catalog #MAB3426, R&D Systems (recognizing whole Rheb); 1:1000; catalog #13879; Cell Signaling Technology (preferably recognizing unfarnesylated Rheb)], α/β-tubulin (1:1000; catalog #2148, Cell Signaling Technology), phospho (p)-ERK1/2 (Thr202/Tyr204; 1:1000; catalog #9101, Cell Signaling Technology), ERK1/2 (1:1000; catalog #9102, Cell Signaling Technology), syntenin1 (1:400; catalog #sc-100336, Santa Cruz Biotechnology), p-S6 (Ser240/244; 1:1000; catalog #5364, Cell Signaling Technology), and S6 (1:1000; catalog #2317, Cell Signaling Technology), and HRP-conjugated secondary antibodies (1:6000; catalog #170–6515, BIO-RAD) were used.

Quantitative analysis of dendritic arc immunofluorescence.

Cultured hippocampal neurons (DIV15) were treated with lonafarnib (2 μm) or DMSO and incubated for 3 d, and then were exposed to vehicle (H₂O) or 10 μm (S)-3,5-dihydroxyphenylglycine (DHPG) (catalog #120020, Abcam) for 60 min. The neurons were fixed with 4% PFA, and were stained with anti-activity-regulated cytoskeleton-associated protein (Arc; mouse; 1:25; catalog #sc-17839, Santa Cruz Biotechnology) and anti-microtubule-associated protein 2 (MAP2; rabbit; 1:500; catalog #AB5622-I, Merck Millipore) antibodies as previously described (Sugiura et al., 2015). Images of stained neurons were collected with a confocal microscope (model LSM710, Carl Zeiss) with a 63× objective [numerical aperture (NA), 1.4]. Two-dimensional maximum projection images were reconstructed, and they were analyzed using MetaMorph software (Molecular Devices). The background corresponding to areas without cells was subtracted to make a “background-subtracted” image. Neurons were selected blindly based on the MAP2 fluorescence, and one dendrite was randomly chosen from one neuron. Twenty-micrometer-long regions along the dendrites (50–70 μm from soma) were manually outlined, and the average gray values (immunofluorescence integrated intensity) of each region were measured. The average value of H₂O-treated dendrites was defined as “base value,” and each value of DHPG-treated dendrites was divided by the base value of each condition.

Contextual fear discrimination test.

Eight- to 16-week-old mice were fear conditioned to the training context with three 0.1 mA shocks (2 s each) controlled by TimeFZ4 software (OHARA). Training sessions were 5 min in duration. Conditioned fear responses were assessed on the next day by TimeFZ4 software, which measured the percentage of time spent motionless during the test period (5 min session). To determine context-dependent conditioned responses, trained mice were assessed in the same training context and in a novel context (both were 5 min sessions). The novel context was created by varying the following conditions of the testing apparatus (training context vs novel context): room shape (rectangle vs triangle), wall opacity (clear vs opaque), floor material (metal bars vs plastic plate), lighting (150 vs 30 lux), and background white noise (65 vs 70 dB). All apparatus are manufactured by OHARA. For pharmacological experiments, the animals were orally administered lonafarnib (40 mg/kg) or vehicle [1% carboxymethyl cellulose (CMC; catalog #419273, Sigma-Aldrich), 0.9% NaCl, and 5% DMSO, 10 ml/kg].

Hot plate test.

Eight- to 16-week-old mice were placed on a plate heated to 55 ± 0.5°C (model NPH-M30N, Nissin Rika). A square acryl tube was placed on the plate to prevent evacuation of the mouse. Nociception was assessed by measuring the latency time for licking the hindpaws or jumping (the cutoff time was 60 s).

Diolistic labeling.

Eight- to 16-week-old mice were orally administered lonafarnib (40 mg/kg) or vehicle. On the next day, the mice were deeply anesthetized and perfused transcardially with PBS, followed by 4% PFA in 0.1 m phosphate buffer, pH 7.4. The brains were rapidly removed and postfixed for >24 h with 4% PFA; then, the brains were placed in PBS. Subsequently, 250-µm-thick coronal brain sections were obtained using a vibratome (model VT1200 S, Leica) and stored in PBS. Twenty-five milligrams of gold particles (catalog #1652263, BIO-RAD) was coated with 7.0 mg of the lipophilic dye DiI (catalog #D282, Thermo Fisher Scientific) and injected into Tefzel tubing, which was then cut into bullets. These particles were delivered diolistically at 80 psi using a Helios Gene Gun system (BIO-RAD). The slices were fixed again in 4% PFA for 15 min and mounted on slides in Vectashield Mounting Medium (catalog #H-1000, Vector Laboratories). Confocal images of dendrites were taken in stratum radiatum of CA1 domain. Each image was at least 100 μm apart, so one dendrite image should represent one neuron. Images were obtained with an confocal microscope (model LSM780, Carl Zeiss) using a 63× lens (Carl Zeiss; NA, 1.4) with sequential acquisition settings at the resolution of the confocal microscope (512 × 512 pixels). Three-dimensional stack images were acquired with the interval between images set to 0.407 µm. Stack images were maximally projected by the Zen program (Zeiss). Projected spine images were quantified with MetaMorph image analysis software (Molecular Devices) as follows: individual dendritic spines were manually traced, and the maximum length and head width of each dendritic spine and spine density were measured. For the cumulative plot graphs, we counted the number of spines below a certain length or width and calculated the percentage of the total number of measured spines.

Immunohistochemistry.

To measure Arc levels, individual mice resting in their home cages during their daytime sleep cycle were rapidly anesthetized and perfused with 4% PFA; then, the tissues were incubated in sucrose and sectioned. To block nonspecific binding, brain sections were incubated in 5% goat and donkey serum for 2 h at room temperature. The sections were incubated for 3 d at 4°C in mouse anti-Arc antibody (1:100, Santa Cruz Biotechnology) diluted in a blocking solution. Biotinylated anti-mouse secondary antibody (1:1000; Vector Laboratories) was diluted in a manner similar to the primary antibody and applied to the sections for 2 h. An Avidin-Biotin Complex (ABC) Kit (Vector Laboratories) and 3,3-diaminobenzidine tetrahydrochloride (catalog # D5637, Sigma-Aldrich) were used to visualize antibody labeling. Sections were imaged and analyzed with AxioVision (Carl Zeiss). Arc immunoreactivity was quantified in the entire granular cell layer of the dentate gyrus (DG) in dorsal hippocampal sections (−1.82 to −1.34 mm bregma). To obtain high-resolution photomicrographs of the entire DG within a brain section, multiple images of each brain section were acquired and automatically reconstructed. Regions of interest representing individual cells were visualized and quantified to obtain the total number of Arc-immunoreactive cells and the total area per section. These values were averaged per mouse and compared across genotypes.

To measure S6 and phospho-S6 levels, mice were orally administered lonafarnib (40 mg/kg) or vehicle. On the next day, the mice were deeply anesthetized and perfused transcardially with PBS, followed by 4% PFA in 0.1 m phosphate buffer, pH 7.4. The brains were rapidly removed and postfixed for >24 h with 4% PFA; then, the tissues were incubated in 20% sucrose for >20 h and sectioned by 30 μm thickness. Antigen retrieval was performed by a heat-induced epitope retrieval technique using microwave oven. In brief, brain sections were immersed in 10 mm sodium citrate buffer, pH 6.0, and boiled for 5 min, three times (10 min interval). Slides were cooled to room temperature for 30 min and washed with PBS for 10 min, three times. To block nonspecific binding, brain sections were incubated in PBS with 2% goat serum 0.1% Triton X-100 for 2 h at room temperature. The sections were incubated for 5 d at 4°C in anti-S6 antibody (1:50; catalog #2317, Cell Signaling Technology) and anti-phospho-S6 antibody (Ser240/244; 1:800; catalog #5364, Cell Signaling Technology) diluted in a blocking solution. Immunoreactivity was visualized using species-specific, Alexa Fluor 488- or Alexa Fluor 568-conjugated secondary antibodies (1:1000; Thermo Fisher Scientific). Secondary antibodies were diluted in PBS with 0.1% Triton X-100 applied to the sections for 2 h at room temperature. Sections were imaged with AxioImager Z.1 with the MosaiX option (Carl Zeiss). S6 and p-S6 immunoreactivity was quantified in the entire granular cell layer of the dentate gyrus images in dorsal hippocampal sections (−1.82 to −1.34 mm from bregma). MetaMorph software was used for the quantification. The whole granular cell layer was selected, and the averages of S6 and p-S6 intensity of the selected area were collected. The background intensity was collected from the area of the ventricle, and background intensity was subtracted from the granular cell layer intensity. The calculated p-S6 intensity was divided by the calculated S6 intensity, and the result was treated as the p-S6/S6 intensity value.

Experimental design and statistical analysis.

GraphPad Prism version 8.0 software was used for statistical analyses of the data. Detailed methods are described in the figure legends. Differences were considered significant at p < 0.05. All experiments requiring the use of animals, directly or as a source of cells, were subjected to randomization.

Results

Farnesylation inhibition restored dendritic spine formation in Tsc2^+/− mouse hippocampal neurons

We reported that Tsc2^+/− rat neurons showed impaired synapse formation, including reduced spine synapses or increased shaft synapses (Sugiura et al., 2015). To assess the efficacy of FTIs for the treatment of synaptic abnormalities in TSC mouse neurons, we tested whether FTI treatment restores proper synapse formation in cultured Tsc2^+/− mouse neurons. Treatment with the FTI lonafarnib (SCH66336) increased the dendritic spine width and decreased the spine length of Tsc2^+/− mouse neurons (Fig. 1a,b). Lonafarnib treatment almost completely restored spine synapse formation in Tsc2^+/− mouse neurons, whereas spine density was not altered by lonafarnib treatment (Fig. 1a,c,d). In contrast, treatment with the mTORC1 inhibitor rapamycin did not affect the spine width, length, or spine synapse ratio in either wild-type (WT) or Tsc2^+/− mouse neurons (Fig. 1e–h). Thus, pharmacological inhibition of Rheb1, but not mTORC1, may rescue proper synapse formation in Tsc2^+/− neurons.

Figure 1.

Treatment with FTIs and heterozygous knockout of Rheb1 restored spine synapse formation in Tsc2^+/− neurons. a, Confocal images of EGFP-expressing cultured hippocampal neurons from WT or Tsc2^+/− mouse. Immunodetection of VGluT1⁺ (red) presynaptic boutons on EGFP (green)-expressing WT and Tsc2^+/− neurons treated with DMSO (vehicle) or 2 μm lonafarnib. Scale bar, 5 µm. b, Cumulative probability distributions of the spine widths (left) and lengths (right) in a. Note that lonafarnib treatment increased the dendritic spine width (D = 0.5180; ***p < 0.0001) and decreased the dendritic spine length (D = 0.247; ***p < 0.001) compared with vehicle treatment in Tsc2^+/− neurons [spines/neurons, respectively; n = 541/12 (WT/vehicle), n = 363/10 (WT/lonafarnib), n = 525/12 (Tsc2^+/−/vehicle), and n = 328/8 (Tsc2^+/−/lonafarnib)]. c, No significant differences were observed in spine density in the condition of a [dendrites; n = 12 (WT, vehicle), n = 10 (WT, lonafarnib), n = 12 (Tsc2^+/−, vehicle), and n = 8 (Tsc2^+/−, lonafarnib); one-way ANOVA: F_(3,38) = 0.996, p = 0.405; two-way ANOVA: treatment × genotype interaction: F_(1,38) = 0.078, p = 0.781; genotype: F_(1,38) = 2.948, p = 0.094; treatment: F_(1,38) = 0.008, p = 0.931]. d, Quantification of VGluT1⁺ spines in a. The decrease of spine synapses in Tsc2^+/− was recovered by lonafarnib treatment [dendrites; n = 11 (WT, vehicle), n = 10 (WT, lonafarnib), n = 12 (Tsc2^+/−, vehicle), and n = 12 (Tsc2^+/−, lonafarnib); one-way ANOVA, post hoc Tukey's test, F_(3,41) = 36.13; ***p < 0.001, Tsc2^+/− + vehicle vs WT + vehicle; ***p < 0.001, Tsc2^+/− + vehicle vs Tsc2^+/− + lonafarnib; two-way ANOVA: treatment × genotype interaction: F_(1,41) = 32.93, p < 0.001; genotype: F_(1,41) = 38.15, p < 0.001; treatment: F_(1,41) = 32.50, p < 0.001]. e, Confocal images of EGFP-expressing cultured hippocampal neurons from a WT or Tsc2^+/− mouse. Immunodetection of VGluT1⁺ (red) presynaptic boutons on EGFP (green)-expressing WT and Tsc2^+/− neurons treated with DMSO (vehicle) or 100 nm rapamycin. Scale bar, 5 µm. f, Cumulative probability plots of the dendritic spine widths (left) and lengths (right) in e. Note that rapamycin treatment increased the dendritic spine length (D = 0.086, *p = 0.030) and did not alter the dendritic spine width (D = 0.028, p = 0.982) compared with those following DMSO treatment in Tsc2^+/− neurons [spines/neurons, respectively; n = 570/20 (WT/vehicle), n = 564/17 (WT/rapamycin), n = 579/18 (Tsc2^+/−/vehicle), and n = 551/18 (Tsc2^+/−/rapamycin); Kolmogorov–Smirnov test]. g, No significant differences were observed in spine density in the condition of e [dendrites; n = 20 (WT, vehicle), n = 17 (WT, rapamycin), n = 18 (Tsc2^+/−, vehicle), and n = 18 (Tsc2^+/−, rapamycin); one-way ANOVA: F_(3,69) = 0.167, p = 0.919; two-way ANOVA: treatment × genotype interaction: F_(1,69) = 0.381, p = 0.539; genotype: F_(1,69) = 0.009, p = 0.924; treatment: F_(1,69) = 0.111, p = 0.741]. h, Quantification of VGluT1⁺ spines in e. Rapamycin treatment did not alter the decrease of spine synapses in Tsc2^+/− neurons [dendrites; n = 10 (WT, vehicle), n = 8 (WT, rapamycin), n = 10 (Tsc2^+/−, vehicle), and n = 9 (Tsc2^+/−, rapamycin); one-way ANOVA, post hoc Tukey's test: F_(3,33) = 56.93; ***p < 0.001, Tsc2^+/− + vehicle vs WT + vehicle; p = 0.628, Tsc2^+/− + vehicle vs Tsc2^+/− + rapamycin; two-way ANOVA: treatment × genotype interaction: F_(1,33) = 0.094, p = 0.761; genotype: F_(1,33) = 163.4, p < 0.001; treatment: F_(1,33) = 3.931, p = 0.056]. Error bars indicate SD. ***p < 0.005, n.s.: not significant. D: the largest absolute difference between the two distribution functions across all values.

Lonafarnib suppressed the activation of Rheb1 but not Ras

We next sought to corroborate biochemical inhibition of Rheb1 farnesylation in FTI-treated neurons. FLAG-Rheb1 was expressed in HEK293 T cells, which were then treated with 2 μm lonafarnib or DMSO for 24 h. The treated cells were fractionated into soluble and insoluble fractions and immunoblotted with an anti-FLAG antibody. We estimated the farnesylation status of Rheb1 by a mobility shift assay; the farnesylated Rheb1 protein migrates faster in SDS-PAGE gels than unfarnesylated Rheb1 (Basso et al., 2005), and also unfarnesylated Rheb1 migrates slower than the nonfarnesylated Rheb1 mutant does (Wang et al., 2008). Lonafarnib treatment slowed the movement of the soluble FLAG-Rheb1 protein, suggesting that most soluble Rheb1 was unfarnesylated in the lonafarnib-treated cells (Fig. 2a). Among the insoluble fractions, a small amount of farnesylated Rheb1 remained in the untreated cells, but none was detected in the lonafarnib-treated cells (Fig. 2a; Clark et al., 1997). Thus, FTI treatment may increase unfarnesylated Rheb1 protein levels; however, total Rheb1 protein levels were reduced in lonafarnib-treated cells, probably via protein degradation (Sugiura et al., 2015).

Figure 2.

Rheb1 farnesylation and protein levels were modified by lonafarnib treatment. a, SDS-PAGE analyses of Rheb1 farnesylation. HEK293 T cells transfected with FLAG-Rheb1 were incubated with DMSO (lanes 1 and 3) or 2 μm lonafarnib (lanes 2 and 4), and their lysates were separated into supernatant (sup) and pellet (ppt) fractions. Top, Unfarnesylated (U) and farnesylated (F) FLAG-Rheb1 proteins are indicated. Bottom, Quantification of unfarnesylated and farnesylated FLAG-Rheb1 proteins in each fraction shown in the top panel (n = 8 independent experiments; two-tailed t test with Welch's correction: t_(13.84) = 15.090, ***p < 0.001, vehicle-treated condition vs lonafarnib-treated condition in sup-farnesylated; t_(9.182) = 9.030, ***p < 0.001, vehicle-treated condition vs lonafarnib-treated condition in sup-unfarnesylated; and t_(7.084) = 6.374, ***p < 0.001 vehicle-treated condition vs lonafarnib-treated ppt-farnesylated; three-way ANOVA: solubility × treatment × farnesylation interaction: F_(1,56) = 105.1, p < 0.001; solubility × treatment interaction: F_(1,56) = 4.895, p = 0.031; solubility × farnesylation interaction: F_(1,56) = 2.711, p = 0.105; treatment × farnesylation interaction: F_(1,56) = 288.8, p < 0.001; solubility: F_(1,56) = 207.6, p < 0.001; treatment: F_(1,56) = 22.35, p < 0.001; farnesylation: F_(1,56) = 29.14, p < 0.001. b, Top, Lonafarnib treatment also increased unfarnesylation of endogenous Rheb1 protein in HEK293 T cells. Bottom, Quantification of unfarnesylated and farnesylated Rheb1 proteins in each homogenate shown in the top panel (n = 7 independent experiments, two-tailed t test with Welch's correction: t_(6.829) = 14.394, ***p < 0.001, vehicle-treated condition vs lonafarnib-treated condition farnesylated; t_(6.004) = 11.953, ***p < 0.001, vehicle-treated condition vs lonafarnib-treated condition unfarnesylated; two-way ANOVA: treatment × farnesylation interaction: F_(1,24) = 326.2, p < 0.001; treatment: F_(1,24) = 2.204, p = 0.151; farnesylation: F_(1,24) = 10.56, p = 0.003). c, Top, Lonafarnib treatment drastically increased unfarnesylated Rheb1 levels in primary cultured hippocampal neurons. Bottom, Quantification of Rheb1 levels in each homogenate shown in the top panel (n = 7 independent experiments, two-tailed t test with Welch's correction: t_(11.943) = 12.186, ***p < 0.001). d, Lonafarnib did not influence the activation of mTORC1. Primary cultured hippocampal neurons were treated with DMSO or lonafarnib. Top, p-S6 levels in each supernatant were compared. Bottom, Quantification of the p-S6 levels in each homogenate shown in the top panel (n = 10 independent experiments; two-tailed t test with Welch's correction, t_(15.620) = 0.529 p = 0.604). e, Lonafarnib did not affect the activation of ERK1/2. Top, Primary cultured hippocampal neurons were treated with DMSO or lonafarnib. Phospho-ERK1/2 levels in each supernatant were compared. Bottom, Quantification of phospho-ERK1/2 levels in each homogenate shown in the top panel (n = 10 independent experiments, two-tailed t test with Welch's correction: t_(16.410) = 1.674, p = 0.113). f, Left, Six hours of lonafarnib treatment was sufficient to increase unfarnesylated Rheb1 in HEK293 T cells. Right, Quantification of unfarnesylated and farnesylated Rheb1 proteins in each homogenate shown in the top panel [n = 9 independent experiments, post hoc Dunnett's test: F_(4,40) = 196.495; ***p < 0.001 (0 h vs 6, 12, and 24 h); p = 0.160 (0 vs 3 h); two-way ANOVA: time point × farnesylation interaction: F_(4,80) = 213.0, p < 0.001; time point: F_(4,80) = 16.02, p < 0.001; farnesylation: F_(1,80) = 458.7, p < 0.001]. g, Lonafarnib treatment decreased unfarnesylated Rheb1 levels in a syntenin-dependent manner. Top, HEK293 T cells transfected with FLAG-Rheb1 and with mock or EGFP-syntenin were incubated with lonafarnib and either DMSO (lanes 1 and 3) or MG132 (lanes 2 and 4). Bottom, Quantification of unfarnesylated Rheb1 proteins in each homogenate shown in the left panel (n = 12 independent experiments, two-tailed t test with Welch's correction: t_(20.516) = 2.547, *p = 0.019, vehicle-treated groups, syntenin (–) and syntenin (+) conditions; t_(21.511) = 2.105, *p = 0.047, syntenin (+) groups, vehicle-treated vs MG132-treated conditions; two-way ANOVA, treatment × syntenin interaction: F_(1,44) = 2.154, p = 0.149; treatment: F_(1,44) = 2.231, p = 0.142; syntenin: F_(1,44) = 2.168, p = 0.148). Error bars indicate SD. *p < 0.05, ***p < 0.005, n.s.: not significant. sup: supernatant, ppt: precipitate.

To evaluate whether the endogenous Rheb1 protein is also affected by FTI treatment, we treated HEK293 T cells with lonafarnib or DMSO, and immunoblotted their extracts with anti-Rheb1 antibody. Again, the Rheb1 protein in lonafarnib-treated cells migrated more slowly than that in DMSO-treated cells (Fig. 2b). Then, we tested the effects of lonafarnib on the neuronal Rheb1 protein. We cultured hippocampal neurons from wild-type mice and treated them with 2 μm lonafarnib or DMSO for 24 h; however, we did not see any difference in the size of the Rheb1 protein between DMSO- and lonafarnib-treated cells with the same antibody. This may be because the mobility shift of farnesylated Rheb1 in mouse cells is not easily visualized because of close proximity. Therefore, we used the antibody that mainly recognized the unfarnesylated Rheb1 protein, and observed lonafarnib treatment caused a robust increase in unfarnesylated Rheb1 levels compared with those following DMSO treatment (Fig. 2c), indicating that lonafarnib treatment clearly upregulates unfarnesylated Rheb1 levels in cultured neurons.

FTIs may also act through the inhibition of mTORC1 signaling following the unfarnesylation of Rheb1. Therefore, we immunoblotted vehicle- or lonafarnib-treated neuronal lysates with anti-phospho S6 antibody. However, lonafarnib treatment did not affect S6 phosphorylation in cultured hippocampal neurons under our conditions (2 μm lonafarnib for 24 h; Fig. 2d). Previous literature shows that 1 μm lonafarnib treatment almost completely suppressed the phosphorylation of S6 in MCF cells (a human breast cancer cell line; Basso et al., 2005). This discrepancy could be because of cell type specificity (primary neurons vs breast cancer cells). mTORC1 signaling may be activated in cancer cells through Rheb1, whereas amino acid–Rag signaling might play more crucial roles in neuronal mTOR regulation. In fact, several RAG1-GAP mutations cause dysplasia in only the brain, but not in other organs, via the activation of mTORC1 (Iffland et al., 2019). Thus, the inhibition of Rheb1 farnesylation could have a less substantial effect on mTORC1 activity in neurons than in other cell types. Our result may also be in line with the observation that lovastatin, which also blocks protein prenylation, did not reduce the phosphorylation of S6 kinase or the S6 protein in hippocampal slices (Osterweil et al., 2013). FTIs may exert their effects via another pathway distinct from mTORC1 signaling.

FTIs inhibit the farnesylation of other signaling molecules, such as Ras. To test whether the effects of lonafarnib on TSC neurons were because of the inhibition of Ras function, we examined the phosphorylation status of ERK1/2, which is a downstream kinase of H-Ras signaling (Osterweil et al., 2013). However, we failed to detect any inhibition of ERK1/2 phosphorylation in lonafarnib-treated cultured neurons (Fig. 2e). Thus, Ras inhibition does not seem to be involved in the neuronal mechanism of lonafarnib action. This may be because the Ras protein can be alternatively prenylated by geranylgeranyl transferase-1 in the presence of an FTI (Whyte et al., 1997). These results suggest that the inhibition of Rheb1 may normalize synaptic abnormalities in Tsc2^+/− neurons in an mTORC1- or MAPK-independent manner.

Finally, we investigated the time required for lonafarnib to accumulate unfarnesylated Rheb1. We treated HEK293 T cells with lonafarnib, and collected their extracts at the indicated time point. Unfarnesylated Rheb1 could be detected 3 h after lonafarnib treatment, and a significant increase could be observed 6 h after the addition of lonafarnib (Fig. 2f). A gradual increase of unfarnesylated Rheb1 was observed thereafter. We assumed that it is better to wait at least 6 h after the administration of lonafarnib for additional experiments.

Lonafarnib treatment decreased Rheb1 protein levels in a syntenin-dependent manner

We previously demonstrated that Rheb1 activation released the PDZ protein syntenin, which bound syndecan-2 to impair synapse formation, in Tsc2^+/− neurons (Sugiura et al., 2015). We therefore assessed whether Rheb1–syntenin signaling (Sugiura et al., 2015) is involved in FTI-induced Rheb1 suppression. FLAG-Rheb1 was expressed with or without EGFP-syntenin in HEK293 T cells. The cells were then incubated with 2 μm lonafarnib for 24 h, along with 20 μm MG-132 or DMSO during the last 6 h. Unfarnesylated FLAG-Rheb1 levels were compared between the presence and absence of syntenin. Compared with the absence of syntenin, the coexpression of syntenin significantly reduced unfarnesylated Rheb1 levels in lonafarnib-treated cells (Fig. 2g). We have shown that the unfarnesylated Rheb1–syntenin complex appeared to be degraded by proteasome (Sugiura et al., 2015). Therefore, we treated the syntenin-expressing cell with a proteasome inhibitor, MG-132, and found that the unfarnesylated Rheb1 levels are not reduced in the presence of MG-132. Thus, lonafarnib could reduce the Rheb1 levels by increasing the unfarnesylated Rheb1, which might be prone to syntenin-dependent degradation.

Lonafarnib treatment increased unfarnesylated Rheb1 and restored synaptic morphology in the brain

To evaluate whether lonafarnib increases unfarnesylated Rheb1 in the brain, we administered lonafarnib (40 mg/kg) or vehicle to adult wild-type mice and dissected and extracted their livers and hippocampi 6 h later. Each soluble fraction was separated by SDS-PAGE and immunoblotted with anti-Rheb1 antibody. Lonafarnib treatment clearly increased unfarnesylated Rheb1 levels in the liver but not in the hippocampus (Fig. 3a,b). Because basal levels of unfarnesylated Rheb1 are higher in the hippocampus than in the liver, a small increase in Rheb1 level in the hippocampus might be less obvious than that in the liver. Therefore, we examined another farnesylated protein, HDJ2, instead of Rheb1 and detected a significant increase in mobility-shifted HDJ2 protein in the hippocampi of lonafarnib-treated mice compared with vehicle-treated mice (Fig. 3c). In contrast, phosphorylation of the S6 protein and MAP kinase was no different in the brains of vehicle- and lonafarnib-treated mice (Fig. 3d,e). Immunohistochemistry also showed that phosphorylation of S6 protein did not change by lonafarnib treatment (Fig. 3f,g).

Figure 3.

Lonafarnib treatment inhibited the farnesylation of a hippocampal protein and restored dendritic spine morphology in the hippocampi of Tsc2^+/− mice in vivo. a, Lonafarnib treatment increased unfarnesylated Rheb1 protein in the liver in vivo. Livers were dissected from WT mice treated with DMSO or lonafarnib. Top, Lonafarnib treatment drastically increased unfarnesylated Rheb1 levels. Bottom, Quantification of Rheb1 levels in each homogenate shown in the top panel (n = 7 independent experiments; two-tailed t test with Welch's correction: t_(11.659) = 5.576, ***p < 0.001). b, Lonafarnib treatment did not show a significant increase in unfarnesylated Rheb1 levels in the hippocampus in vivo. Top, Hippocampi were dissected from WT mice treated with DMSO or lonafarnib. Bottom, Quantification of Rheb1 levels in each homogenate shown in the top panel (n = 8 independent experiments, two-tailed t test with Welch's correction: t_(10.950) = 1.020, p = 0.330). c, Lonafarnib treatment increased unfarnesylated HDJ-2 protein in the hippocampus in vivo. Top, Hippocampi were dissected from WT mice treated with DMSO or lonafarnib. The HDJ-2 protein in each homogenate was immunoblotted. Unfarnesylated (U) and farnesylated (F) HDJ-2 proteins are indicated. Bottom, Quantification of the unfarnesylated HDJ-2 protein levels shown in the top panel (n = 7 independent experiments, two-tailed t test with Welch's correction: t_(10.860) = 3.795; ***p = 0.003). d, Lonafarnib treatment did not affect p-S6 protein levels in the brain. Top, Phosphorylated and total S6 proteins are indicated. Bottom, Quantification of the phospho-S6 levels shown in the top panel (n = 9 independent experiments; two-tailed t test with Welch's correction: t_(14.326) = 0.376, p = 0.713). e, Lonafarnib treatment did not affect the phosphorylation of MAP kinase in the brain. Top, Phosphorylated and total ERK1/2 proteins are indicated. Bottom, Quantification of the phospho-ERK1/2 levels shown in the top panel (n = 9 independent experiments; two-tailed t test with Welch's correction: t_(14.679) = 0.169, p = 0.868). f, Representative images of S6 (green) and p-S6 (red) staining in the DG of vehicle- or lonafarnib-administered WT mice. Scale bar, 300 μm. g, Quantification of relative p-S6/S6 intensity in the DG. Lonafarnib administration did not alter p-S6/S6 value in WT mice (n = 18/3 hippocampi/animals; two-tailed t test with Welch's correction: t_(3.13153) = 1.899, p = 0.150). Data are averaged and presented by mice. h, Representative DiI-labeled dendrites from the hippocampal CA1 neurons of WT and Tsc2^+/− mice treated with lonafarnib or vehicle. Scale bar, 2 μm. i, Cumulative probability plots of the spine width (left) and spine length (right) in h. Tsc2^+/− neurons showed a decreased spine width [D = 0.316; ***p < 0.001 (Tsc2^+/− + vehicle vs WT + vehicle)], and lonafarnib treatment rescued the aberrant spine shape [D = 0.2590, ***p < 0.001, Tsc2^+/− + vehicle vs Tsc2^+/− + lonafarnib; spines/neurons/animals, respectively: n = 1055/31/3 (WT + vehicle), n = 1155/30/3 (WT + lonafarnib), n = 1092/30/3 (Tsc2^+/− + vehicle), and n = 1039/32/3 (Tsc2^+/− + lonafarnib); Kolmogorov–Smirnov test]. j, Quantification of spine density in f [neurons/animals, respectively: n = 31/3 (WT, vehicle), 30/3 (WT, lonafarnib), n = 30/3 (Tsc2^+/−, vehicle), and n = 32/3 (Tsc2^+/−, lonafarnib); one-way ANOVA, F_(3,8) = 0.195 p = 0.897; two-way ANOVA: treatment × genotype interaction: F_(1,8) = 0.361, p = 0.565; genotype: F_(1,8) = 0.098, p = 0.763; treatment: F_(1,8) = 0.126, p = 0.732]. Data are averaged and presented by mouse type. Error bars indicate SD. ***p < 0.005, n.s.: not significant. D: the largest absolute difference between the two distribution functions across all values.

Next, to examine synaptic recovery by the FTI in the brain, we labeled hippocampal CA1 neurons in hippocampal slices prepared from vehicle- or lonafarnib-treated WT and Tsc2^+/− mouse brains with biolistic-DiI. Vehicle or lonafarnib was administered on the day before perfusion. The dendritic spines in vehicle-treated Tsc2^+/− neurons were significantly thinner than those in WT neurons; however, the spine length did not show a significant difference between WT and Tsc2^+/− mice, inconsistent with the results of neuronal cultures (Fig. 1b). This might be because of the difference in the surrounding environment; that is, sparse cultured neurons may more easily elongate and form dendritic spines than those in the brain, enhancing the difference in length. Lonafarnib treatment increased spine widths, but not spine density, in Tsc2^+/− neurons to a similar extent as WT neurons (Fig. 3h–j), suggesting that lonafarnib reaches the hippocampus through the blood–brain barrier and may restore synaptic morphology via an increase in the unfarnesylated Rheb1 protein in the neurons.

Farnesylation inhibition restored memory deficits in Tsc2^+/− mice

A previous study showed that Tsc2^+/− mice have an impaired ability to distinguish between familiar and novel contexts in a fear-conditioning task (Ehninger et al., 2008). Because lonafarnib restored synaptic morphology (Figs. 1, 3), we tested whether the FTI ameliorates cognitive deficits in Tsc2^+/− mice. For this behavioral test, mice are exposed to a footshock. The next day, contextual discrimination is tested by measuring the freezing time in the original context and in a distinct context. Before this discrimination test, we compared the pain thresholds of WT and Tsc2^+/− mice by the hot plate test, but we did not find any significant difference in pain threshold between the genotypes (Fig. 4a). For fear discrimination, WT mice that were orally administered vehicle could distinguish the contexts, whereas the vehicle-treated Tsc2^+/− mice could not (Fig. 4b). The oral administration of lonafarnib (40 mg/kg) 6 h before training eliminated the lack of context discrimination in Tsc2^+/− mice (Fig. 4b). Thus, the administration of lonafarnib before footshock (encoding) restored contextual memory in Tsc2^+/− mice. Next, to determine whether lonafarnib treatment after learning also improves memory retrieval, we administered lonafarnib to Tsc2^+/− mice after footshock (6 h before recall). However, these Tsc2^+/− mice could not discriminate different contexts (Fig. 4c), suggesting that the administration of lonafarnib before learning (encoding) is required in Tsc2^+/− mice during the learning phase to rescue the learning defect. Interestingly, it was not significant, but the administration of lonafarnib to Tsc2^+/− mice increased the freezing time in the original context, whereas the freezing time in distinct context was affected by whether lonafarnib was administrated before or after the footshock. These results suggest that Tsc2^+/− mice show a disability in memory encoding rather than memory retrieval, and that lonafarnib shows the efficacy to restore the synaptic encoding that may be required during the memory retrieval.

Figure 4.

Lonafarnib treatment restored the fear memory deficits of Tsc2^+/− mice in memory encoding. a, The hot plate test indicated no significant difference in the time for a nociceptive response between WT and Tsc2^+/− mice (n = 15 animals; two-tailed t test with Welch's correction: t_(27.861) = 0.432, p = 0.669). b, Schematic illustration of the experimental design for the contextual learning task. Impaired contextual fear memory in Tsc2^+/− mice was rescued by the oral administration of lonafarnib, whereas learning ability in WT mice was not altered by drug administration [n = 12 animals; two-tailed t test with Welch's correction: t_(19.247) = 2.559, *p = 0.019 (WT + vehicle); t_(20.393) = 2.546, *p = 0.019 (WT + lonafarnib); t_(21.978) = 1.592, p = 0.126 (Tsc2^+/− + vehicle); t_(18.779) = 2.341, *p = 0.030 (Tsc2^+/− + lonafarnib); three-way ANOVA: genotype × treatment × apparatus interaction: F_(1,88) = 0.280, p = 0.598; genotype × treatment interaction: F_(1,88) = 0.003, p = 0.954; genotype × apparatus interaction: F_(1,88) = 0.344, p = 0.559; treatment × apparatus interaction: F_(1,88) = 0.360, p = 0.550; genotype: F_(1,88) = 0.959, p = 0.330; treatment: F_(1,88) = 0.671, p = 0.415; apparatus: F_(1,88) = 20.68, p < 0.001; multiple comparison after three-way ANOVA: post hoc Sidak's test, p > 0.999 (same: WT + vehicle vs WT + lonafarnib); p > 0.999 (novel: WT + vehicle vs WT + lonafarnib); p = 0.990 (same: Tsc2^+/− + vehicle vs Tsc2^+/− + lonafarnib); p > 0.999 (novel: Tsc2^+/− + vehicle vs Tsc2^+/− + lonafarnib)]. c, Schematic illustration of the experimental design for the contextual learning task. Impaired contextual fear memory in Tsc2^+/− mice was not rescued by the oral administration of lonafarnib after footshock, whereas the ability of the WT mice to learn was not altered by drug administration [n = 12 animals; two-tailed t test with Welch's correction: t_(15.876) = 2.477, *p = 0.025 (WT + vehicle); t_(21.590) = 3.386, ***p = 0.003 (WT + lonafarnib); t_(18.761) = 1.301, p = 0.209 (Tsc2^+/− + vehicle); t_(21.316) = 1.076, p = 0.294 (Tsc2^+/− + lonafarnib); three-way ANOVA, genotype × treatment × apparatus interaction: F_(1,88) = 0.011, p = 0.916; genotype × treatment interaction: F_(1,88) = 5.018, p = 0.028; genotype × apparatus interaction: F_(1,88) = 2.352, p = 0.129; treatment × apparatus interaction: F_(1,88) = 0.157, p = 0.693; genotype: F_(1,88) = 0.060, p = 0.806; treatment: F_(1,88) = 9.355, p = 0.003; apparatus: F_(1,88) = 15.47, p < 0.001; multiple comparison after three-way ANOVA, post hoc Sidak's test: p > 0.999 (same: WT + vehicle vs WT + lonafarnib); p > 0.999 (novel: WT + vehicle vs WT + lonafarnib); p = 0.074 (same: Tsc2^+/− + vehicle vs Tsc2^+/− + lonafarnib); p = 0.157 (novel: Tsc2^+/− + vehicle vs Tsc2^+/− + lonafarnib); no outliers in the data, Grubb's test (α = 0.05)]. Error bars indicate SD. *p < 0.05, ***p < 0.005, n.s.: not significant.

The DHPG-induced increase in dendritic arc protein levels was reversed in Tsc2^+/− neurons by lonafarnib treatment

Several lines of evidence show that long-term depression (LTD) induced by the metabotropic glutamate receptor (mGluR)-selective agonist DHPG is impaired in mouse models of TSC (Auerbach et al., 2011; Bateup et al., 2011). In particular, Auerbach et al. (2011) demonstrated that a decrease in the translation of Arc caused mGluR-LTD suppression in TSC mice. Arc stimulates the trafficking of AMPA receptors and induces mGluR-LTD (Chowdhury et al., 2006; Waung et al., 2008). Therefore, we compared the DHPG-induced increase in dendritic Arc levels between cultured WT and Tsc2^+/− neurons. In WT neurons, DHPG stimulation significantly increased dendritic Arc levels compared with basal levels (Fig. 5a–c). However, DHPG treatment did not increase dendritic Arc levels in Tsc2^+/− neurons (Fig. 5b,c). Prior treatment of Tsc2^+/− neurons with lonafarnib restored the DHPG-induced increase in dendritic Arc levels (Fig. 5b,c). Together, FTI treatment recovered DHPG-mediated dendritic Arc expression in TSC neurons, suggesting that the FTI-dependent increase of Arc expression could restore the mGluR-LTD impairment in TSC mice.

Figure 5.

Arc protein synthesis in the dendrites of Tsc2^+/− neurons was restored by lonafarnib treatment. a, Schematic illustration of the experimental design to quantify the synthesis of the Arc protein. b, Representative images of Arc staining in WT and Tsc2^+/− hippocampal dendrites treated with DMSO (vehicle) or lonafarnib for 3 d and 60 min after DHPG or H₂O (vehicle) application. Scale bar, 5 μm. c, Quantification of the dendritic Arc staining in b. DHPG-treated Arc levels are divided by average H₂O-treated Arc levels [n = 5 (WT, DMSO, H₂O), n = 5 (WT, DMSO, DHPG), n = 5 (WT, lonafarnib, H₂O), n = 5 (WT, lonafarnib, DHPG), n = 5 (Tsc2^+/−, DMSO, H₂O), n = 6 (Tsc2^+/−, DMSO, DHPG), n = 5 (Tsc2^+/−, lonafarnib, H₂O), n = 6 (Tsc2^+/−, lonafarnib, DHPG), neurons, respectively; one-way ANOVA, post hoc Dunnett's test: F_(3,18) = 1.739, *p = 0.045 (WT, DMSO vs Tsc2^+/−, DMSO); *p = 0.040 (Tsc2^+/−, DMSO vs Tsc2^+/−, lonafarnib); two-way ANOVA, treatment × genotype interaction: F_(1,18) = 1.071, p = 0.315; genotype: F_(1,18) = 2.826, p = 0.110; treatment: F_(1,18) = 3.189, p = 0.091). *p < 0.05.

Lonafarnib treatment restored activity-regulated Arc expression by memory retrieval in the hippocampus

To verify restoration of activity-regulated plastic change in Tsc2^+/− mice, we examined Arc expression in the hippocampus in vivo. Arc is involved in spatial memory formation (Guzowski et al., 2001) and has been thought to reflect the regional specificity of space encoding using neuronal ensemble recordings (Vazdarjanova and Guzowski, 2004). Therefore, we compared the number of Arc-expressing neurons in the hippocampal DG between WT and Tsc2^+/− mice, when they generate contextual memory recall (Fig. 6a). As a basal control, WT and Tsc2^+/− mice were killed in their home cages (basal), and these mice had some basal Arc⁺ neurons in their DGs (Fig. 6b). As a retrieval group, WT and Tsc2^+/− mice first received a footshock and were then exposed to the same context 24 h later. Both vehicle- and lonafarnib-treated WT mice exhibited a clear increase in the number of Arc⁺ neurons in their DGs by fear memory retrieval, whereas vehicle-treated Tsc2^+/− mice did not (Fig. 6b,c). This suggests that when WT mice recall a memory, Arc may be induced in some populations of DG neurons, but this mechanism is impaired in Tsc2^+/− mice. The oral administration of lonafarnib (40 mg/kg) 6 h before footshock significantly increased the number of Arc⁺ neurons in the Tsc2^+/− mice after exposure to the same context. Thus, lonafarnib treatment may restore retrieval-related changes in Arc expression in Tsc2^+/− mice by improving memory encoding, which is consistent with the findings that lonafarnib restored DHPG-induced Arc synthesis (Fig. 5) and memories (Fig. 4b) in Tsc2^+/− neurons and mice, respectively. These results indicate that behavioral rescue by lonafarnib may be associated with restoration of the neural plasticity in the Tsc2^+/− mice.

Figure 6.

Lonafarnib-mediated recovery of memory encoding restored memory retrieval induced Arc expression in the hippocampi of Tsc2^+/− mice. a, Schematic illustration of the experimental design for the immunohistological analysis after the memory recall-inducing treatment. b, Home cage-treated (basal) and same-condition-treated (memory-recall-induced) Arc expression in the DG of vehicle- or lonafarnib-administered WT and Tsc2^+/− mice. Scale bar, 100 μm. c, Quantification of Arc⁺ neurons in the DG. Vehicle-treated WT mice exhibited an increase in Arc⁺ neurons in the DG after memory recall, whereas vehicle-treated Tsc2^+/− mice did not show a significant increase in Arc expression. Lonafarnib treatment restored Arc induction in the DG neurons of Tsc2^+/− mice [hippocampi/animals for home cage and same condition, respectively; n = 18/3 and 30/5 (WT + vehicle), n = 18/3 and 32/4 (WT + lonafarnib), n = 30/3 and 17/4 (Tsc2^+/− + vehicle), and n = 18/3 and 30/4 (Tsc2^+/− + lonafarnib); two-tailed t test with Welch's correction: t_(5.407) = 3.523, *p = 0.015 (WT + vehicle); t_(4.760) = 4.175, **p = 0.010 (WT + lonafarnib); t_(3.781) = 1.083, p = 0.343 (Tsc2^+/− + vehicle); t_(2.429) = 710.964, **p = 0.004 (Tsc2^+/− + lonafarnib); three-way ANOVA: genotype × treatment × induction interaction: F_(1,21) = 6.928, p = 0.016; genotype × treatment interaction: F_(1,21) = 6.760, p = 0.017; genotype × induction interaction: F_(1,21) = 1.565, p = 0.225; treatment × induction interaction: F_(1,21) = 5.055, p = 0.035; genotype: F_(1,21) = 27.39, p < 0.001; treatment: F_(1,21) = 7.522, p = 0.012; induction: F_(1,21) = 57.32, p < 0.001]. Data are averaged and presented by mouse type. Error bars indicate SD. *p < 0.05, **p < 0.01, n.s.: not significant.

Heterozygous knockout of Rheb1 also restored synaptic abnormalities and contextual memory impairments in Tsc2^+/− mice

Finally, in addition to pharmacological recovery, we validated that genetic reduction of Rheb1 is necessary for the restoration of synaptic abnormalities in Tsc2^+/− neurons. We generated excitatory neuron-specific heterozygous Rheb1 knock-out mice. Hippocampal neurons from CamkII-Cre (Cre), CamkII-Cre;Rheb1^flox/+ (Rheb1-flox), Tsc2^+/−;CamkII-Cre (Tsc2^+/−;Cre), and Tsc2^+/−;CamkII-Cre;Rheb1^flox/+ (Tsc2^+/−; Rheb1-flox) mice were cultured, and their spine morphologies were compared. Compared with Cre neurons, Tsc2^+/−;Cre neurons showed narrower and longer dendritic spines with a reduced spine synapse density. Heterozygous Rheb1 knockout in Tsc2^+/−;Cre (Tsc2^+/−; Rheb1-flox) mice significantly increased the dendritic spine width/spine synapse ratio (Fig. 7a–d). Collectively, pharmacological and genetic inhibition of Rheb1 may rescue proper synapse formation in Tsc2^+/− neurons.

Figure 7.

Heterozygous knockout of Rheb1 mimicked the lonafarnib-mediated recovery of spine malformation in Tsc2^+/− neurons and impaired fear memory in Tsc2^+/− mice. a, Confocal images of cultured hippocampal neurons from CamkII-Cre, CamkII-Cre; Rheb1^flox/+ (Rheb1-flox), Tsc2^+/−; CamkII-Cre (Tsc2^+/−; Cre), or Tsc2^+/−; CamkII-Cre; Rheb1^flox/+ (Tsc2^+/−; Rheb1-flox) mice. Immunodetection of VGluT1⁺ (red) presynaptic boutons on EGFP-expressing (green) neurons. Scale bar, 5 µm. b, Cumulative probability distributions of the spine widths (left) and lengths (right) in a. Note that heterozygous Rheb1 knockout increased the dendritic spine width (D = 0.281, ***p < 0.001) and decreased the dendritic spine length (D = 0.095, *p = 0.013) in Tsc2^+/− neurons [spines/neurons, respectively; n = 564/16 (CamkII-Cre), n = 558/14 (Rheb1-flox), n = 552/15 (Tsc2^+/−; Cre), and n = 562/17 (Tsc2^+/−; Rheb1-flox; Kolmogorov–Smirnov test)]. c, Quantification of spine density in a [dendrites; n = 16 (CamkII-Cre), n = 14 (Rheb1-flox), n = 15 (Tsc2^+/−; Cre), and n = 17 (Tsc2^+/−; Rheb1-flox); one-way ANOVA, post hoc Tukey's test: F_(3,58) = 0.548, p = 0.053 (CamkII-Cre vs Tsc2^+/−; Cre)]. d, Quantification of VGluT1⁺ spines in a. Decrease of spine synapses with VGluT1 puncta in Tsc2^+/−; Cre neuron was restored in Tsc2^+/−; Rheb1-flox neurons [dendrites; n = 8 (CamkII-Cre), n = 8 (Rheb1-flox), n = 10 (Tsc2^+/−; Cre), and n = 10 (Tsc2^+/−; Rheb1-flox); one-way ANOVA: post hoc Tukey's test: F_(3,32) = 14.061; ***p < 0.001, Tsc2^+/−; Cre vs CamkII-Cre; ***p < 0.001, Tsc2^+/−; Cre vs Rheb1-flox; ***p = 0.003, Tsc2^+/−; Cre vs Tsc2^+/−; Rheb1-flox]. e, Heterozygous Rheb1 knockout rescued the memory deficits of Tsc2^+/− mice. CamkII-Cre mice showed intact memory by freezing more in the familiar context than in the novel context (n = 12 animals; t_(21.315) = 3.648, ***p = 0.002), whereas Tsc2^+/−; Cre mice showed impaired contextual discrimination (n = 12 animals; t_(15.688) = 1.908, p = 0.075). Heterozygous Rheb1 knockout rescued the deficit in contextual discrimination (n = 12 animals; t_(20.108) = 2.797, *p = 0.011), and Rheb1-flox mice exhibited normal memory (n = 12 animals; t_(21.998) = 2.678, *p = 0.014), two-tailed t test with Welch's correction. Two-way ANOVA: genotype × apparatus interaction: F_(3,88) = 0.789, p = 0.503; genotype: F_(3,88) = 1.143, p = 0.336; apparatus: F_(1,88) = 30.88, p < 0.001). f, The hot plate test indicated no significant difference in the time for a nociceptive response among CamkII-Cre, Rheb1-flox, Tsc2^+/−; Cre, and Tsc2^+/−; Rheb1-flox mice (one-way ANOVA; n = 15 animals; F_(3,56) = 0.169, p = 0.917). g, Representative DiI-labeled dendrites from the hippocampal CA1 neurons of CamkII-Cre, Rheb1-flox, Tsc2^+/−; Cre, or Tsc2^+/−; Rheb1-flox mice. Scale bar, 2 μm. h, Cumulative probability plots of the spine width (left) and spine length (right) in g. Tsc2^+/−; Cre neurons showed a decreased spine width (D = 0.227; ***p < 0.001, Tsc2^+/−; Cre vs CamkII-Cre), and heterozygous Rheb1 knockout rescued the aberrant spine width [D = 0.179; ***p < 0.001, Tsc2^+/−; Cre vs Tsc2^+/−; spines/neurons/animals, respectively; Rheb1-flox; n = 1035/26/3 (CamkII-Cre), n = 1027/25/3 (Rheb1-flox), n = 1016/29/3 (Tsc2^+/−; Cre), and n = 1029/30/3 (Tsc2^+/−; Rheb1-flox); Kolmogorov–Smirnov test)]. i, Quantification of spine density in g [neurons/animals, respectively; n = 26/3 (CamkII-Cre), n = 25/3 (Rheb1-flox), n = 29/3 (Tsc2^+/−; Cre), and n = 30/3 (Tsc2^+/−; Rheb1-flox); one-way ANOVA, post hoc Tukey's test: F_{(3, 8)} = 0.175 p = 0.910]. Data are averaged and presented by mice. Error bars indicate SD. *p < 0.05, ***p < 0.005, n.s.: not significant. D: the largest absolute difference between the two distribution functions across all values.

We inferred that the inhibition of Rheb1 signaling benefits behavioral performance in Tsc2^+/− mice. To further validate this hypothesis, we tested contextual discrimination in Tsc2^+/−; Rheb1-flox mice, based on the notion that excitatory neuron-specific Rheb1 hetero-knockout mimics the reduced Rheb1 function with lonafarnib. Tsc2^+/−; Rheb1-flox mice recalled the context and froze longer than Tsc2^+/−;Cre mice with an equivalent pain sensation to that of other mice (Fig. 7e,f); however, the effects of excitatory neuron-specific Rheb1 knockout were milder than those of lonafarnib treatment. This might be because Rheb1 activation is not suppressed in other cells, such as astrocytes, which could have contributed to the cognitive deficits. Furthermore, the Rheb1 hetero-knockout rescued the dendritic spine width in the Tsc2^+/−;Cre mice, without affecting dendritic spine density (Fig. 7g–i). We further investigated whether pharmacological inhibition has additive effect on the genetic deletion of Rheb1; in other words, whether genetic deletion of Rheb1 ameliorates cognitive deficits in Tsc2^+/− mice as well as lonafarnib administration does. In histologic analysis, lonafarnib administration to Tsc2^+/−;Rheb1-flox mice did not increase spine width additionally in vivo (Fig. 8a–c). In behavioral analysis, lonafarnib administration to Tsc2^+/−;Rheb1-flox mice showed comparable results to those in vehicle-treated Tsc2^+/−;Rheb1-flox mice (Fig. 8d). Still, we cannot completely rule out the possibilities that lonafarnib affects other target proteins than Rheb1 to recover neuronal and behavioral disability; however, the main target protein of lonafarnib would be Rheb1 in Tsc2^+/− mice because genetic Rheb1 deletion and lonafarnib treatment have sufficient effects on neuronal morphology and function recovery. Together, pharmacological and genetic inhibition of Rheb1 may be crucial for the restoration of synaptic and memory retrieval in Tsc2^+/− mice.

Figure 8.

Lonafarnib administration had no or weak effect on heterozygous knockout of Rheb1 in the recovery of spine malformation in Tsc2^+/− neurons and impaired fear memory in Tsc2^+/− mice. a, Representative DiI-labeled dendrites from the hippocampal CA1 neurons of vehicle- or lonafarnib-administered Tsc2^+/−; CamkII-Cre (Tsc2^+/−; Cre) and Tsc2^+/−; CamkII-Cre; Rheb1^flox/+ (Tsc2^+/−; Rheb1-flox) mice. Scale bar, 2 μm. b, Cumulative probability plots of the spine width (left) and spine length (right) in a. Tsc2^+/−; Rheb1-flox + lonafarnib neurons showed a comparable spine width to Tsc2^+/−; Cre + lonafarnib neurons and Tsc2^+/−; Rheb1-flox + vehicle neurons [D = 0.033, p = 0.638 (Tsc2^+/−; Rheb1-flox + lonafarnib vs Tsc2^+/−; Cre + lonafarnib); D = 0.009, ***p < 0.001 (Tsc2^+/−; Rheb1-flox + lonafarnib vs Tsc2^+/−; Rheb1-flox + vehicle; Tsc2^+/−; Rheb1-flox + vehicle showed wider average; spines/neurons/animals, respectively: n = 1038/26/3 (Tsc2^+/−; Cre + vehicle), n = 1027/26/3 (Tsc2^+/−; Cre + lonafarnib), n = 1008/29/3 (Tsc2^+/−; Rheb1-flox + vehicle), and n = 1023/26/3 (Tsc2^+/−; Rheb1-flox + lonafarnib); Kolmogorov–Smirnov test]. c, Quantification of spine density in a [neurons/animals, respectively; n = 26/3 (Tsc2^+/−; Cre + vehicle), n = 26/3 (Tsc2^+/−; Cre + lonafarnib), n = 29/3 (Tsc2^+/−; Rheb1-flox + vehicle), and n = 26/3 (Tsc2^+/−; Rheb1-flox + lonafarnib); one-way ANOVA: F_(3,8) = 1.440, p = 0.302; two-way ANOVA, treatment × genotype interaction: F_(1,8) = 0.071, p = 0.797; genotype: F_(1,8) = 3.418, p = 0.102; treatment: F_(1,8) = 0.830, p = 0.389; data are averaged and presented by mouse type]. d, Schematic illustration of the experimental design for the contextual learning task. Tsc2^+/−; Rheb1-flox + lonafarnib mice showed a comparable phenotype of learning ability to Tsc2^+/−; Cre + lonafarnib mice and Tsc2^+/−; Rheb1-flox + vehicle mice (n = 12 animals; two-tailed t test with Welch's correction: t_(21.225) = 0.483, p = 0.638 (Tsc2^+/−; Cre + vehicle); t_(21.359) = 2.149, *p = 0.043 (Tsc2^+/−; Cre + lonafarnib); t_(20.661) = 2.444, *p = 0.024 (Tsc2^+/−; Rheb1-flox + vehicle); t_(21.799) = 2.462, *p = 0.022 (Tsc2^+/−; Rheb1-flox + lonafarnib); three-way ANOVA: genotype × treatment × apparatus interaction: F_(1,88) = 1.088, p = 0.300; genotype × treatment interaction: F_(1,88) = 1.235, p = 0.269; genotype × apparatus interaction: F_(1,88) = 0.770, p = 0.383; treatment × apparatus interaction: F_(1,88) = 0.644, p = 0.424; genotype: F_(1,88) = 1.000, p = 0.320; treatment: F_(1,88) = 0.829, p = 0.365; apparatus: F_(1,88) = 13.80, p < 0.001; multiple comparison after three-way ANOVA: post hoc Sidak's test: p = 0.998 (same: Tsc2^+/−; Cre + vehicle vs Tsc2^+/−; Cre + lonafarnib); p = 0.988 (novel: Tsc2^+/−; Cre + vehicle vs Tsc2^+/−; Cre + lonafarnib); p = 0.997 (same: Tsc2^+/−; Rheb1-flox + vehicle vs Tsc2^+/−; Rheb1-flox + lonafarnib); p = 0.735 (novel: Tsc2^+/−; Rheb1-flox + vehicle vs Tsc2^+/−; Rheb1-flox + lonafarnib)]. Error bars indicate SD. *p < 0.05, n.s.: not significant. D: the largest absolute difference between the two distribution functions across all values.

Discussion

Here, we demonstrated that the activation of Rheb1, but not mTORC1, is involved in key aspects of synaptic and behavioral abnormalities in TSC model mice. Because pharmacological and genetic inhibition of Rheb1 restored synaptic and behavioral impairments in TSC mice, we investigated the detailed mechanism. First, lonafarnib treatment converted Rheb1 to its inactive, unfarnesylated form without affecting mTORC1 or MAP kinase signaling. Second, lonafarnib treatment restored plasticity-related Arc expression to control levels in both TSC neurons and mice. Finally, lonafarnib treatment corrected impaired memory engrams in the hippocampi of TSC mice. Collectively, activation of Rheb1 may disturb proper synapse and circuit formation, resulting in cognitive deficits in TSC. FTI treatment could ameliorate TAND-related intellectual disability through the restoration of synapses and memory engrams by Rheb1 inhibition. We show that the TSC–Rheb1 pathway impacts neuronal synapses and circuits, and provides insight into neurodevelopmental processes and potential treatment options for TAND symptoms.

FTIs have been developed to inhibit farnesylation of Ras proteins, since post-translational farnesylation is necessary for the membrane targeting of Ras proteins. Rheb1 also requires post-translational prenylation for its proper localization and activity (Buerger et al., 2006). Rheb1 has been shown to be localized to lysosomal membranes via its C-terminal farnesylation (Sancak et al., 2010). Tsc2 is released from Rheb1-localized lysosomal membranes in a growth factor-dependent or amino acid-dependent manner (Demetriades et al., 2016), leading to the activation of Rheb1. Thus, the farnesylation-mediated Rheb1 localization may be critical for its activation. This study showed that the inhibition of Rheb1 farnesylation restored synaptic abnormalities and cognitive deficits in Tsc2^+/− neurons and mice, respectively, indicating that Rheb1 activation may be an underlying cause for the TAND symptoms of TSC patients.

FTIs can affect the activities of other signaling molecules; therefore, we cannot rule out the possibility that the beneficial effects of lonafarnib on cognition are because of the inhibition of other proteins. A variety of substrates other than Ras and Rheb1 are farnesylated, including lamin A, lamin B, RhoB, the centromere proteins CENP-E and CENP-F, and Rhes (Basso et al., 2005; Hernandez et al., 2019). Interestingly, lonafarnib was shown to ameliorate behavioral abnormalities in the tauopathy model via Rhes farnesylation (Hernandez et al., 2019). This contributing mechanism is lysosomal activation, which might be different from that of Rheb1 degradation (Sugiura et al., 2015). However, as lonafarnib-induced Rheb1 decreases are crucial for its actions (Fig. 2a), lysosomal involvement needs to be investigated. On the other hand, heterozygous knockout of the Rheb1 gene restored spine formation (Fig. 7a–d) and normal cognition (Fig. 7e,f) in Tsc2^+/− neurons and mice, respectively. In addition, lonafarnib administration to Tsc2^+/−;Rheb1-flox mice had no or weak additive effect on spine formation and learning ability (Fig. 8); therefore, the activation of Rheb1 signaling in Tsc2^+/− mice may be sufficient for the manifestation of cognitive deficits that these could be improved by FTI treatment. Furthermore, genetic deletion of farnesyltransferase has been shown to modify Alzheimer's disease neuropathology and to improve synaptic function and learning ability in animal models (Cheng et al., 2013; Jeong et al., 2021; Qu et al., 2021). Rheb1 was one of the substrate proteins that was enriched in the brains of farnesyltransferase knock-out mice (Qu et al., 2021).

Hyperactivity of the mTOR pathway in mice can lead to a myriad of phenotypes such as behavioral abnormalities. Several lines of evidence suggest that the mTOR pathway regulates long-term potentiation (LTP; Tang et al., 2002) and that increased mTOR signaling can change the induction of late-phase LTP, which most likely accounts for the learning and memory deficits in Tsc2^+/− mice. Thus, mTOR inhibitors can improve the cognitive deficits associated with TSC. In fact, high-dose rapamycin (2–5 mg/kg) treatment reversed the contextual memory deficits in Tsc2^+/− mice (Ehninger et al., 2008; Way et al., 2012). However, how upregulated mTORC1 signaling causes cognitive deficits in TSC mice remains unclear. Rapamycin did not restore spine synapse formation in Tsc2^+/− neurons under our conditions (Fig. 1e–h). Other studies also showed that rapamycin treatment caused modest to robust increases in spine length in Tsc1 or Tsc2 knock-down neurons (Tavazoie et al., 2005; Meikle et al., 2008). Therefore, rapamycin could improve TSC-related cognitive deficits without restoring synaptic morphology. It might be possible that FTI restores memory deficits via partial inhibition of mTORC1 signaling, but its steady-state activity was not suppressed by lonafarnib treatment (Fig. 3d). Apart from cognitive function, rapamycin can prevent epilepsy in a mouse model of TSC (Zeng et al., 2008), whereas we failed to see that lonafarnib prevented PTZ-mediated seizure exacerbation (data not shown), indicating that FTI would exert its efficacy for ameliorating memory dysfunction in Tsc2^+/− mice.

In contrast, lonafarnib restored the cognitive deficits in Tsc2^+/− mice, probably through the correction of synaptic morphology and Arc expression. Impaired Arc induction might disturb mGluR5-dependent long-term plasticity in TSC neurons (Auerbach et al., 2011). Additionally, a reduction in the dendritic spines of memory engram cells was shown to be well correlated with memory impairments in a model of Alzheimer's disease (Roy et al., 2016). Therefore, impaired spine formation in Tsc2^+/− mice might cause distorted engram connections in the DG, resulting in memory deficits.

We administered lonafarnib to the mice 6 h before footshock (learning) and tested their memory recall 24 h after footshock. As Tsc2^+/− mice froze with the familiar context 30 h after lonafarnib administration, its effects appear to persist for at least 30 h. However, lonafarnib treatment after encoding did not increase discrimination ability in Tsc2^+/− mice. Therefore, the administration of lonafarnib before learning is necessary for memory encoding. We speculate that pretreatment with lonafarnib can restore synaptic abnormalities by encoding (footshock plus context) and that its effects can persist until recall. The dendritic spine width at 24 h after administration was significantly greater in lonafarnib-treated mice than in vehicle-treated mice (Fig. 3h,i). Although the T_1/2 (biological half-life) of lonafarnib in the blood is approximately several hours, its effects in the brain might continue longer through the maintenance of dendritic spine morphology. We could not find significant increases of unfarnesylated Rheb1 and decreases of total Rheb1 in the lonafarnib-administered mouse brain by Western blot, but lonafarnib treatment recovered abnormal neuronal morphology and memory-encoding dysfunction, indicating lonafarnib could penetrate the blood–brain barrier. This might suggest that Rheb1 in the neurons is easier to be unfarnesylated by lonafarnib than that in the glial cells, and unfarnesylated Rheb1 is rather sensitive to proteasomal degradation during biochemical sampling in vivo.

Clinical investigations of TSC patients have been performed with the rapamycin analog everolimus. SEGA regression was demonstrated in a double-blind, placebo-controlled phase III trial. In another study, the seizure frequency was lower in TSC patients with refractory seizures who were treated with everolimus than in those treated with placebo as a control (Krueger et al., 2013). However, everolimus treatment has not yet improved neurocognitive symptoms so far (Overwater et al., 2019). This failure might be because everolimus was not administered at a critical window of efficacy (Ehninger et al., 2009; Way et al., 2012). In addition, some adverse effects of everolimus, such as stomatitis, pharyngitis, pneumonia, vomiting, and decreased appetite, have been reported (Franz et al., 2013; French et al., 2016). On the other hand, lonafarnib has been used to treat progeria and hepatitis D in clinical trials, and both studies provided evidence for the efficacy and relatively low toxicity of lonafarnib (Koh et al., 2015; Gordon et al., 2018). Because neuronal Rheb1 expression is upregulated by epileptic seizures (Yamagata et al., 1994) and high expression of constitutively active Rheb1 in the brain displayed spontaneous, recurrent seizures (Hsieh et al., 2016; Nguyen et al., 2019), epilepsy-related cognitive problems could also be a therapeutic target for lonafarnib. Our observations may corroborate how Rheb1 protein can inform biological mechanisms underlying common traits in the neurodevelopmental disorders. We speculate that targeted administration of lonafarnib, perhaps during critical postnatal neurodevelopmental windows, might be of significant benefit to patients. Its safety profile may promote this drug to a promising candidate for treatment of intractable cognitive deficits in TSC patients.