ABSTRACT
Fragile X syndrome (FXS) is caused by mutations in the FMR1 (fragile X mental retardation 1) gene. It is a significant form of heritable intellectual disability with comorbidity of other symptoms such as autism. Due to the lack of efficacious medication, repurposing the existing FDA-approved drugs may offer an opportunity to advance clinical intervention for FXS. Analysis of the whole-genome transcription signatures predicts new therapeutic action of vorinostat to correct pathological alterations associated with FXS. We further find that the administration of vorinostat restores object location memory and passive avoidance memory in the Fmr1 knockout (KO) mice. For the non-cognitive behavioral symptoms, vorinostat corrects the autism-associated alterations, including repetitive behavior and social interaction deficits. In the open field test, vorinostat dampens hyperactivity in the center area of the arena. Surprisingly, vorinostat does not affect the abnormally elevated protein synthesis in Fmr1 KO neurons, suggesting different outcomes from correcting behavioral symptoms and specific aspects of cellular pathology. Our data reveal the therapeutic effects of the FDA-approved drug vorinostat in a mouse model of FXS and advocate efficacy testing with human patients.
INTRODUCTION
Fragile X syndrome (FXS) is a genetic disorder caused by mutations in the FMR1 (fragile X mental retardation 1) gene. As the most frequent mutation, the increased number of the CGG trinucleotide repeat in the 5’ non-coding region hampers gene transcription and leads to a significant reduction or lack of FMRP (fragile X mental retardation protein) expression. The main symptoms of FXS patients include cognitive disability, hyperactivity, and autism-related behavior 1,2. Based on the pathological and mechanistic studies with animal models of FXS, strategies targeting specific molecular and cellular alterations have shown notable therapeutic efficacy in preclinical studies 3-7. Nevertheless, efficacious medication is not available yet.
Toward achieving clinical treatment for FXS, the development of brand-new drugs may easily take more than ten years. Comparing to de novo drug development, repurposing the existing FDA-approved drugs, of which the toxicity has already been tested in humans, and the detailed pharmacology and formulation are available, provides unprecedented opportunity to cut the amount of money, time, and effort. The main traditional approaches of drug repurposing usually depend on the outcome of high-throughput screening 8 or knowledge of drug structure and mechanism of action. More recently, computational comparison of drug-induced transcriptome profiles has been proposed as a non-structure based in-silico screening of similarity drugs 9-11. By using such a computation approach, our recent study predicts that the FDA-approved drug vorinostat may have therapeutic effects to correct FXS-associated symptoms 12. However, the prediction requires empirical validation.
Vorinostat, also known as suberanilohydroxamic acid (SAHA), is currently used to treat cancers, including cutaneous T cell lymphoma (CTCL) and advanced non-small-cell lung carcinoma (NSCLC). Although, as a small chemical compound, it should have promiscuous pharmacological activities, its inhibition activity against class I, II, and IV histone deacetylases (HDAC) is well recognized. Vorinostat also shows neuroprotective effects in the central nervous system and is suggested to treat neurodegeneration such as Alzheimer’s disease and Huntington disease 13-15. As histone acetylation-induced epigenetic changes are involved in activity-dependent plasticity and learning 16, a well-recognized outcome of vorinostat treatment is the enhancement of memory and cognition 15.
In this study, we examined the effects of vorinostat in the Fmr1 KO mice. We found that vorinostat corrected deficits in object location memory and passive avoidance memory. Interestingly, it also corrected non-cognitive behavioral symptoms, including repetitive behavior, social interaction deficits, and a specific aspect of hyperactivity. Our study provides evidence to support a new therapeutic action of vorinostat that is predicted by an unbiased transcriptome-based computational approach. It also advocates future tests with human clinical trials.
MATERIALS AND METHODS
Comparison of vorinostat-induced transcriptome changes with other drugs in the Connectivity Map (CMap) database
The search for drugs/compounds that induce similar transcriptome changes to that of vorinostat in the CMap database was performed as described 12. In brief, we first obtained microarray data sets of MCF7, PC3, and HL60 cells treated with vorinostat and the corresponding vehicle controls. The differential gene expression analysis was conducted, and the top 500 up-regulated and down-regulated genes were retained as the signature for further query 12,17,18. The transcriptome signature of vorinostat was uploaded to the CMap query page and used to search for compounds that induce similar or oppositional transcriptome changes in the corresponding cell line (i.e., MCF7, PC3, and HL60) 9. The returning results of drugs/compounds with a p-value of less than 0.05 in each cell line are presented in Supplementary Tables 1, 2, and 3.
Animals
Young adult male mice at 2.5 to 3.5 months of age were used for behavioral examinations. Postnatal day 0 mice were used to obtain primary hippocampal neurons. The Fmr1 knockout (KO) and their wild type (WT) littermates are on the C57BL/6 background. The Institutional Animal Care and Use Committee approved all procedures.
Behavioral examinations
For the examination of object location memory, mice were habituated to the training chamber without any object for 10 min on three consecutive days. Mice were then trained by a 10-min exposure to the training chamber holding two objects at different locations (Fig. 2a), during which the mice freely explored the chamber and interacted with the objects. Twenty-four hours after training, the trained mice were tested by a 10-min re-exposure to the same chamber with one object at the same location and one object at a new location (Fig. 2a). During training and testing, time spending in interacting with each object was recorded. The % of preference was determined by the time spent with each object divided by the total time spent with both objects.
For the examination of passive avoidance memory, mice were trained as described in our previous studies 19. During training, mice received a mild foot shock (0.7 mA for 2 sec) immediately after entering the dark chamber, stayed in the dark chamber for 30 sec, and were then returned to their home cage. Twenty-four hours later, the trained mice were reintroduced to the training chamber; crossover latency (i.e., the time elapsed until the mice crossed over and entered the dark chamber) was recorded. When there was no crossover beyond 600 sec, the examination was terminated, and a crossover latency of 600 sec was used for those specific mice.
Mouse activity in the light/dark box, the open field, and the 3-chamber social interaction test was examined as described in our previous studies 3,12,19.
Drug administration
Vorinostat (Sigma-Aldrich) and trichostatin A (Sigma-Aldrich) were prepared in the vehicle (10% DMSO) and i.p. injected into mice at 50 mg/kg and 10 mg/kg, respectively. The chosen doses were equivalent or higher than doses used in previous in vivo studies, showing that such treatments are sufficient to improve cognitive functions in mice 16,20-22. For the examination of object location and passive avoidance memory, drugs were administered 30 min before training. For all other behavioral paradigms, drugs were administered 30 min before the examination. The control groups were treated with vehicle injection.
Examination of protein synthesis and histone acetylation in neurons
Primary hippocampal neurons were obtained from postnatal day 0 WT and Fmr1 KO mice and maintained in vitro. To determine protein synthesis with the SUnSET method 23, DIV (days in vitro) 14 neurons were incubated with 5µg/ml puromycin (Sigma, Cat #P8833) for 30 min and then harvested in Buffer H (50 mM β-glycerophosphate, 1.5 mM EGTA, 0.1 mM Na3VO4, 1 mM DTT). After the determination of total protein concentration, the samples were examined by Western blot with anti-puromycin antibody (KeraFAST, Cat # EQ0001, 1:1000). For the determination of histone acetylation level, samples collected from DIV 14 WT and Fmr1 KO neurons were examined by Western blot with antibodies against total and acetylated histone proteins H2B, H3, and H4 (Cell Signaling, 1:1000). The relative amount of loading was determined by β-actin. The intensity of the immuno-signal was analyzed by the ImageJ software (NIH, MD, USA).
Data collection and statistics
Mice were randomly assigned to vehicle and drug treatment groups, which were not disclosed before data analysis. Mice from multiple litters were used to avoid pseudo repeats. Data with normal distribution were analyzed by two-sided Student’s t-test or ANOVA. The crossover latency data for passive avoidance testing did not show normal distribution and were analyzed by Fisher’s exact test. All data are expressed as mean ± SEM.
RESULTS
In-silico screening of vorinostat similarity drugs in the CMap database
Transcriptome landscape reflects a particular aspect of molecular outcome in physiological and pathological conditions 24-26. We recently found that transcriptome changes in the Fmr1 KO neurons can successfully predict therapeutic interventions. Among the predicted drugs, an FDA-approved antipsychotics trifluoperazine causes transcriptome alteration oppositional to that caused by FMRP deficiency. It was further demonstrated that trifluoperazine corrects the key FXS-associated symptoms in the Fmr1 KO mice 12. Moreover, computational analysis of the trifluoperazine-induced transcriptome signature revealed other similarity drugs, predicting that those similarity drugs may be repurposed to treat FXS. Among the top 10 similarity drugs, vorinostat is the third-ranked FDA-approved drug following two antipsychotics 12.
Here, we further used vorinostat-induced transcriptome perturbations as a query to identify similarity drugs in the CMap database (Broad Build 02 database, http://www.broadinstitute.org), which contains over 7000 reference gene signatures altered by 1309 compounds/perturbagens. As the gene signatures were characterized in three major cell lines (i.e., MCF7, PC3, and HL60) in the database, we performed a computational analysis to search for similarity drugs within each cell line. The transcriptome signature of vorinostat in each cell line identified drugs showing significant positive and negative similarity scores (i.e., similarity mean with a p-value of less than 0.05) (Supplementary Table 1, 2, and 3). Notably, there is an overlap of similarity drugs among the three cell lines, indicating a certain degree of conservation of transcriptional responses to vorinostat. Among the top 10 ranked compounds (Fig. 1a, 1b, and 1c), 4 drugs are the common vorinostat similarity drugs identified from all 3 cell lines (Fig. 1d). These 4 similarity drugs include 2 HDAC inhibitors (i.e., trichostatin A and valproic acid), trifluoperazine, and a known PI3K (phosphatidylinositol 3-kinase) inhibitor LY-294002. These data further support that, based on their effects on transcriptome signature, vorinostat and trifluoperazine may have a similar action. To validate the in-silico prediction of drug action, we examined the effects of vorinostat in a mouse model of FXS.
Vorinostat corrects cognitive deficits in the Fmr1 KO mice
Recapitulating the intellectual deficits in human FXS patients, the Fmr1 KO mice show compromised cognitive function 2,19. We first examined the effects of vorinostat on object location memory (Fig. 2a). During training, the vehicle- and vorinostat-treated WT and Fmr1 KO mice showed comparable preference to objects at location A and B (genotype effect: F1,70 = 0.001, p = 1.000; drug effect: F1,70 = 0.001, p = 1.000; location effect: F1,70 = 0.133, p = 0.716; Fig. 2b). During testing, there was a significant effect of location preference (genotype effect: F1,70 = 0.001, p = 1.000; drug effect: F1,70 = 0.001, p = 1.000; location effect: F1,70 = 53.032, p = 0.001; Fig. 2c). The vehicle-treated WT but not Fmr1 KO mice showed preference to the object at the new location (i.e., location C), indicating that object location memory is compromised in Fmr1 KO mice (Fig. 2c). In contrast, both the vorinostat-treated WT and Fmr1 KO mice showed preference to the object at location C (Fig. 2c), indicating significant location memory formation.
We next examined the effects of vorinostat on passive avoidance memory (Fig. 2d). During training, there was a genotype effect but no drug effect on crossover latency (genotype effect: F1,30 = 8.514, p = 0.007; drug effect: F1,30 = 0.022, p = 0.884; genotype X drug interaction: F1,30 = 1.358, p = 0.253; Fig. 2e). The vehicle-treated wild type (WT) and Fmr1 KO mice showed similar crossover latency during training (Fig. 2e). Following a single vorinostat administration, Fmr1 KO mice showed less crossover latency than WT mice (Fig. 2e) during training. Still, there is no significant difference between vehicle- and vorinostat-treated Fmr1 KO mice (Fig. 2e). When tested 24 hours later, the vehicle-treated Fmr1 KO mice showed less crossover latency than the vehicle-treated WT mice (Fig. 2e), indicating impaired passive avoidance memory. The acute and single administration of vorinostat did not improve memory in the Fmr1 KO mice (Fig. 2e). We next treated mice with repeated vorinostat administration by daily intraperitoneal (i.p.) injection for 2 weeks. There was a genotype effect but no drug effect on behavior during training (genotype effect: F1,32 = 10.201, p = 0.003; drug effect: F1,32 = 0.511, p = 0.480; genotype X drug interaction: F1,32 = 0.054, p = 0.817; Fig. 2f). Repeated vorinostat restored passive avoidance memory in the Fmr1 KO mice to the WT level, as indicated by the improved crossover latency during testing (Fig. 2f).
Therapeutic effects of vorinostat on repetitive behavior and hyperactivity in the Fmr1 KO mice
In the light/dark box test, the vehicle-treated Fmr1 KO mice showed a higher number of repetitive transitions between the light and dark chambers (Fig. 3a1). Administration of vorinostat normalized this hyperactive and repetitive behavior in the Fmr1 KO mice (genotype effect: F1,39 = 12.932, p = 0.001; drug effect: F1,39 = 4.469, p = 0.041; genotype X drug interaction: F1,39 = 4.364, p = 0.043; Fig. 3a1). Regardless of the treatment, the Fmr1 KO and WT mice showed similar preference and spent comparable time in the light chamber (genotype effect: F1,39 = 0.156, p = 0.695; drug effect: F1,39 = 0.042, p = 0.840; genotype X drug interaction: F1,39 = 5.410, p = 0.025; Fig. 3a2).
In the open field test, the vehicle-treated Fmr1 KO mice showed more locomotion activity than the vehicle-treated WT mice in the whole arena (genotype effect: F1,34 = 25.844, p = 0.001; drug effect: F1,34 = 0.573, p = 0.454; genotype X drug interaction: F1,34 = 0.685, p = 0.414; Fig. 3b). They also showed higher locomotion activity in the center area (genotype effect: F1,34 = 19.996, p = 0.001; drug effect: F1,34 = 3.910, p = 0.056; genotype X drug interaction: F1,34 = 3.457, p = 0.072; Fig. 3c). Following vorinostat administration, the Fmr1 KO mice did not show changes of locomotion in the whole arena (Fig. 3b); they showed a reduction of locomotion in the center area (Fig. 3c).
Therapeutic effects of vorinostat on social deficits in the Fmr1 KO mice
In the 3-chamber social interaction test, all groups of mice spent similar time in the social chamber (genotype effect: F1,36 = 0.000, p = 0.995; drug effect: F1,36 = 0.858, p = 0.360; genotype X drug interaction: F1,36 = 0.443, p = 0.510; Fig. 3d1)(genotype effect: F1,24 = 0.518, p = 0.479; drug effect: F1,24 = 0.122, p = 0.730; genotype X drug interaction: F1,24 = 0.033, p = 0.857; Fig. 3e1) as well as in the non-social chamber (genotype effect: F1,36 = 0.262 p = 0.612; drug effect: F1,36 = 0.695, p = 0.410; genotype X drug interaction: F1,36 = 0.016, p = 0.899; Fig. 3d1)(genotype effect: F1,24 = 0.490, p = 0.491; drug effect: F1,24 = 0.025, p = 0.875; genotype X drug interaction: F1,24 = 0.583, p = 0.453; Fig. 3e1). The vehicle-treated Fmr1 KO mice spent less time in direct interaction with the stranger mouse target (Fig. 3d2 and 3e2). While a single vorinostat administration had no significant effect (genotype effect: F1,36 = 137.711, p = 0.001; drug effect: F1,36 = 4.386, p = 0.043; genotype X drug interaction: F1,36 = 0.010, p = 921; Fig. 3d2), repeated daily vorinostat treatment for 2 weeks improved social interaction in the Fmr1 KO mice to the WT level (genotype effect: F1,24 = 29.767, p = 0.001; drug effect: F1,24 = 3.219, p = 0.085; genotype X drug interaction: F1,24 = 8.127, p = 0.009; Fig. 3e2).
Effects of vorinostat on protein synthesis
The abnormally elevated protein synthesis has been recognized as a prominent aspect of cellular pathology associated with FXS 1,2. We confirmed that the Fmr1 KO neurons display a higher level of new protein synthesis than WT neurons (genotype effect: F1,30 = 13.060, p = 0.001; drug effect: F2,30 = 0.386, p = 0.683; genotype X drug interaction: F2,30 = 0.067, p = 0.935; Fig. 4). However, vorinostat failed to suppress protein synthesis in both Fmr1 KO (F2,30 = 0.384, p = 0.684) and WT neurons (F2,30 = 0.068, p = 0.934) (Fig. 4).
Trichostatin A does not affect the behavioral outcomes in the Fmr1 KO mice
As the main known pharmacological action of vorinostat is HDAC inhibition, we wondered whether the observed therapeutic efficacy is due to a general effect of HDAC inhibition and can be achieved by the administration of a different HDAC inhibitor. Trichostatin A is a well recognized HDAC inhibitor 27 and a top-ranked similarity drug of vorinostat (Fig. 1). Following vehicle and trichostatin A treatment, we examined WT and Fmr1 KO mice with object location memory (Fig. 5a) and light-dark box tests (Fig. 5b). Comparing to WT mice, the trichostatin A-treated Fmr1 KO mice still showed impaired object location memory (location effect: F1,56 = 0.106, p = 0.746; genotype X location interaction: F1,56 = 0.054, p = 0.817; Fig. 5a1) (location effect: F1,56 = 11.296, p = 0.001; genotype X location interaction: F1,56 = 15.294, p = 0.0001; Fig. 5a2) and more transitions between the light and dark chambers (genotype effect: F1,27 = 24.565, p = 0.0001; drug effect: F1,27 = 1.291, p = 0.266; genotype X drug interaction: F1,27 = 0.025, p = 0.875; Fig. 5b1) (genotype effect: F1,27 = 2.127, p = 0.152; drug effect: F1,27 = 1.437, p = 0.241; genotype X drug interaction: F1,27 = 0.012, p = 0.913; Fig. 5b2). We further found that levels of the acetylated histone proteins H2B, H3, and H4 are normal in the Fmr1 KO neurons (Fig. 5c). These data suggest that the therapeutic effect of vorinostat is unlikely due to its inhibition activity against HDAC.
DISCUSSION
While there is no efficacious therapeutics, de novo drug development for FXS treatment is still in its infancy and encountered significant obstacles 28. One alternative and efficient approach is to repurpose the existing FDA-approved drugs 29. In this study, we used the CMap drug-induced transcriptome database to reconfirm vorinostat as a similarity drug to trifluoperazine, which has been recently found to correct FXS-associated symptoms in a mouse model 12. The therapeutic efficacy of vorinostat in correcting a variety of behavioral symptoms is validated with the Fmr1 KO mice.
The use of a holistic analysis of transcriptome signature to predict therapeutic strategy for neurological disorders has been recently proposed but not empirically examined 25. One application is to compare the disease-associated transcriptome signature with drug-induced transcriptome signatures. The value of this application is implicated by that transcriptome signature associated with bipolar disorder and schizophrenia can predict drugs, and some of the predicted drugs are already in clinical use to treat these disorders 24,25. As the first empirical attempt, we screened the CMap database with the FXS-associated transcriptome signature and identified trifluoperazine as a therapeutics to treat symptoms in the Fmr1 KO mice 12. Another important application of the transcriptome-based therapeutic prediction is to compare the transcriptome signatures associated with different drugs and chemical compounds 9,10. Using the trifluoperazine-induced transcriptome change as query identified vorinostat as a top-ranked similarity drug 12. In this study, using the vorinostat-induced transcriptome change as a query also identified trifluoperazine as a top-ranked similarity drug. The transcriptome similarity predicts that vorinostat may have similar therapeutic effects to that of trifluoperazine and be useful to treat specific symptoms associated with FXS. Empirically, this study validates the transcriptome-based approach to identify new drug action and repurpose vorinostat.
It is important to note that the vorinostat- and trifluoperazine-induced transcriptome changes share similarities but are not identical. Depending on the cell types, scores (i.e., similarity mean) underlying the similarity between vorinostat and trifluoperazine are 0.282, 0.307, and 0.19 (Fig. 1 and Supplementary Table 1, 2, and 3). As a score of 1 reflects being identical, and a score of -1 reflects being oppositional, it is anticipated that vorinostat and trifluoperazine should have both common and different pharmacological actions. As far as the correction of behavior symptoms is concerned, vorinostat and trifluoperazine have similar but not identical effects. For example, while a single administration of trifluoperazine rescues passive avoidance memory and social deficits 12, correction of these deficits requires repeated dosing of vorinostat.
It has been recognized that HDAC inhibitors may be considered to improve learning and memory in animal models of cognition impairment 15,16,21. The effect of HDAC inhibitors on non-cognitive functions such as repetitive behavior, hyperactivity, and social interaction has not been recognized and appreciated. Here, we found the therapeutic effects of vorinostat to correct both cognitive and non-cognitive symptoms in Fmr1 KO mice. Interestingly, vorinostat only affected behavior in the Fmr1 KO but not WT mice. As the acetylation levels of H2B, H3, and H4 are normal in Fmr1 KO neurons, it is not straightforward to attribute the therapeutic effect of vorinostat to its HDAC inhibition activity. Further, another HDAC inhibitor trichostatin A failed to rescue deficits of either cognitive or non-cognitive functions. However, we cannot exclude that other HDAC targets are hypo-acetylated in Fmr1 KO neurons.
Notably, while vorinostat corrects certain FXS-associated behavior symptoms, it does not normalize the elevated protein synthesis in Fmr1 KO neurons. This is intriguing and suggests that, at least to a certain degree, elevated protein synthesis may not be absolutely linked to all behavioral abnormalities. A recent study found that human FXS samples show various levels of protein synthesis, and fibroblasts derived from some FXS patients display normal translation 30. Another complication is that it is not clear whether the elevated translation is universal or restricted to specific brain regions and cell types. Alternatively, vorinostat may dampen the translation of particular FMRP target mRNAs rather than affecting overall protein synthesis. These possibilities remain to be addressed with future studies.
FXS is a complex disorder. It is unlikely that a single treatment strategy will correct all aspects of symptoms. Regarding drug repurposing, several FDA-approved drugs, including minocycline 31, metformin 32, lovastatin 33, and trifluoperazine 12 have shown certain therapeutic efficacy in the Fmr1 KO mice. However, these drugs are not able to rescue all pathological outcomes. Repurposing new therapeutic reagents such as vorinostat will not only provide a new potential treatment choice but also expand the possibility of combination therapy. Regarding the transcriptome-based approach to identify new drug effects, other FDA-approved similarity drugs of trifluoperazine 12 and vorinostat (Fig. 1 and Supplementary Tables 1, 2, and 3) may be considered and examined in future studies.
In summary, we used unbiased transcriptome analysis to identify the new therapeutic potential of vorinostat as FXS treatment. We provide evidence to support the value of holistic transcriptome signature in drug repurposing. The effects of vorinostat on correcting cognitive and non-cognitive symptoms in FXS mice encourage human trials.
FUNDING AND DISCLOSURE
The authors declare no conflict of interest. The study was supported by NIH grant R01MH119149 (to HW).
AUTHOR CONTRIBUTION
HW initiated the study. XW, XL, and HW identified drug similarity. QD performed empirical investigations and analyzed the data. QD, XL, and HW wrote the manuscript.
Legends for Supplementary Tables
Supplementary Table 1 vorinostat similarity compounds screened by DEGs induced by vorinostat in MCF7 cells. The p-value determines the ranking of the similarity compounds/drugs; compounds/drugs are listed with their p-values in ascending order.
Supplementary Table 2 vorinostat similarity compounds screened by DEGs induced by vorinostat in PC3 cells. The p-value determines the ranking of the similarity compounds/drugs; compounds/drugs are listed with their p-values in ascending order.
Supplementary Table 3 vorinostat similarity compounds screened by DEGs induced by vorinostat in HL60 cells. The p-value determines the ranking of the similarity compounds/drugs; compounds/drugs are listed with their p-values in ascending order.