HDAC4 Reduction: A Novel Therapeutic Strategy to Target Cytoplasmic Huntingtin and Ameliorate Neurodegeneration

1Huntington's disease (HD) is a progressive, inherited neurological disorder characterized by severe motor, cognitive, behavioural, and physiological dysfunction for which there is no effective disease-modifying treatment. The disease is caused by the expansion of a CAG repeat to more than 35 CAGs within exon 1 of the HTT gene. At the molecular level, mutant huntingtin (HTT) containing an expanded polyQ stretch has a propensity to self-aggregate to produce a wide-range of oligomeric species and insoluble aggregates and exerts a gain of toxic function through aberrant protein–protein interactions. Therefore, as with other neurodegenerative diseases such as Alzheimer's disease, Parkinson's disease, and the prion diseases, the polyglutamine (polyQ) disorders including HD are associated with the accumulation of misfolded proteins leading to neuronal dysfunction and cell death.

2Transcriptional dysregulation is part of the complex molecular pathogenesis of HD, to which abnormal histone acetylation and chromatin remodelling may contribute. The imbalance in histone acetylation was proposed to be caused by the inactivation of histone acetyltransferases, which led to the pursuit of histone deacetylases (HDACs) as HD therepeutic targets,. There are 11 mammalian Zn2+-dependent HDACs divided into three groups based on structural and functional similarities: class I (HDACs: 1, 2, 3, 8), class IIa (HDACs: 4, 5, 7, 9), class IIb (HDACs: 6, 10), and HDAC11 as class IV. Initial genetic and pharmacological studies performed in flies, worms, and HD mouse models have suggested that HDAC inhibitors may have a significant therapeutic potential,.

3The preclinical evaluation of the HDAC inhibitor suberoylanilide hydroxamic acid (SAHA) demonstrated a dramatic improvement in the motor impairment that develops in the R6/2 HD mouse model. Initially, SAHA was shown to inhibit class I and II HDACs at nanomolar concentrations, although it is predominantly a class I inhibitor. More recently, SAHA was shown to lead to the degradation of HDACs 4 and 5 via RANBP2-mediated proteasome degradation in cancer cell lines. Following on from this, we demonstrated that in addition to its deacetylase activity and the known effect on decreasing Hdac7 mRNA levels, SAHA treatment results in a reduction in HDAC2 and HDAC4 in brain regions of both WT and R6/2 mice, without affecting their transcript levels in vivo. This was associated with a reduction in aggregate load and the restoration of cortical Bdnf transcript levels in R6/2 mice.

4It is well-established that HDAC4 acts as a transcriptional repressor that shuttles between the nucleus and cytoplasm. Phosphorylated HDAC4 is retained in the cytoplasm through its association with 14-3-3 proteins. The N-terminal region of HDAC4 contains a MEF2 binding site and represses the transcription of MEF2-dependent genes important in the regulation of neuronal cell death. In this context, MEF2 can act as a neuronal survival factor, and its inhibition has been linked to the death of neurons in several cell culture systems. Crystallization of the N-terminal domain of HDAC4 suggested that HDAC4 might have the propensity to self-aggregate through its glutamine-rich domains, consistent with cell-culture studies. Interestingly, regions containing high glutamine content in proteins have been observed to facilitate interactions with other glutamine-rich proteins, leading to the spontaneous assembly of insoluble toxic amyloid-like structures in mammalian cells.

5In this study, we identified a novel mechanism by which HDACs can modify HD pathogenesis in vivo and found that HDAC4 reduction delays the HTT aggregation process. We demonstrated that HDAC4 associates with mutant exon 1 and full-length HTT in vivo in a polyQ-length-dependent manner and co-localizes with cytoplasmic inclusions in the brains of HD mouse models. HDAC4 knock-down inhibited aggregate formation in both the R6/2 (N-terminal fragment) and HdhQ150 (full-length knock-in) HD mouse models. This delay in aggregation occurred in the cytoplasm, consistent with the subcellular localisation of HDAC4 in mouse brain. We found no evidence for HDAC4 translocation to the nucleus during disease progression, and HDAC4 knock-down had no effect on HTT aggregation in the nucleus and no impact on global transcriptional dysregulation. HDAC4 reduction led to a marked restoration of the membrane properties of medium spiny neurons (MSNs) and of corticostriatal synaptic transmission. This was associated with an improvement in neurological phenotypes and extended survival. These data provide a clear demonstration that cytoplasmic pathogenic mechanisms contribute to HD-related neurodegenerative phenotypes and identify HDAC4 as a therapeutic target for HD.

6In order to investigate whether HDAC4 is involved in the molecular pathogenesis of HD, we used a genetic approach to reduce HDAC4 levels in both the R6/2 and HdhQ150 knock-in HD mouse models. R6/2 mice are transgenic for a mutated N-terminal exon 1 HTT fragment. The HdhQ150 mice have an expanded CAG repeat knocked in to the mouse huntingtin gene (Htt),, which is partially mis-spliced with the result that these mice express mutant versions of both an exon 1 HTT and a full-length HTT protein. Because Hdac4 knock-out (Hdac4KO) mice die in early postnatal life, the HD mutation could not be transferred onto an Hdac4 null background. Therefore, we crossed males for each of the HD mouse models to Hdac4HET females (Figure 1A). Analysis of the progeny indicated that Hdac4 mRNA levels were decreased to approximately 50% in both Hdac4HET and double-mutant mice in both crosses (Figure 1B).

7Since HDAC4 functions as a transcriptional corepressor, it was important to check whether Hdac4 knock-down regulated the expression of the R6/2 transgene, as this would be expected to modulate the onset and progression of disease in R6/2 mice. Therefore, we used Taqman qPCR to demonstrate that exon 1 HTT mRNA was not altered in the cortex (Figure 1C), cerebellum, nor striatum (Figure S1A) of Dble::R6/2 mice. Similarly, we showed that endogenous Htt levels were unchanged as a consequence of HDAC4 reduction in both R6/2 and HdhQ150 mice (Figures 1D and S1B). In addition, given that CAG repeat length is linked to aggregation kinetics and disease progression, we ensured that the CAG repeats were maintained at comparable levels throughout the course of this study (Table S1).

8Alteration of HDAC4 levels has been shown to modulate Hdac9 in muscle cells and Hdac5 in primary mouse hepatocytes. Therefore, we used Taqman qPCR to show that Hdac4 knock-down did not affect the levels of the other 10 Hdacs in brain regions of mice that did or did not express the HD mutation (Figures 1E and S1C and S1D).

9Bioinformatic predictions of HDAC4 structure suggested that HDAC4 has an N-terminal coil–coil domain within which it possesses short polyQ tracts that might convey an increased propensity for amyloid formation as was confirmed in cultured cells. Hence, we hypothesized that HDAC4 might exhibit pro-aggregation properties in HD mouse models. In order to investigate the molecular consequences of HDAC4 knock-down in HD mice, we employed the Seprion ligand ELISA to quantify aggregate load and time-resolved Förster resonance energy transfer (TR-FRET) to measure soluble mutant HTT levels. TR-FRET detects a FRET signal between two appropriately labelled antibodies. In this case, 2B7 (epitope: 1–17 amino acids of HTT) is paired with MW1 (epitope: polyQ in nonaggregated HTT).

10In R6/2 mice, the level of soluble exon 1 HTT decreases with disease progression as a consequence of its aggregation. The Seprion ELISA revealed that HDAC4 knock-down reduced the aggregate load in the cortex (Figure 2A), brain stem, hippocampus, and cerebellum (Figure S2A) of Dble::R6/2 mice at 4 and 9 wk but that this effect had diminished by 15 wk of age. Accordingly, TR-FRET demonstrated that reduced HDAC4 levels led to an increase in soluble exon 1 HTT in the cortex (Figure 2B), brain stem, hippocampus, and cerebellum (Figure S2B) of Dble::R6/2 mice at 4 and 9 but not at 15 wk of age. This shift in the ratio between soluble and aggregated exon 1 HTT levels can be visualised on the western blots in Figure 2D. We performed Seprion ELISA to determine whether similar results could be obtained in the HdhQ150 knock-in mice. A significant reduction in the aggregate load was observed in the striatum, cortex and cerebellum of Dble::HdhQ150 mice at 6 and 10 mo of age (Figure 2C). Taken together, these data show that HDAC4 knock-down significantly reduced aggregate load and increased levels of soluble mutant HTT in HD mouse models, reflecting a delay in the aggregation process.

11We used Hdac4KO P3 brain tissue to confirm that the commercially available antibodies Sigma (DM-15), Santa Cruz (H-92), and Cell Signalling (CS2072) detected an HDAC4 specific signal (Figure S2C and unpublished data). On western blotting of nuclear and cytoplasmic fractions of mouse brain, we found that only trace amounts of HDAC4 could be detected in the nuclear fraction (Figure 2E). Therefore, to investigate whether the reduction in aggregation occurred in the cytoplasm, as would be consistent with the presence of HDAC4, we perfomed cellular fractionation on 4 wk and 9 wk brain tissue. We then resolved detergent insoluble high-molecular weight aggregates from the nuclear and cytoplasmic lysates by agarose gel electrophoresis (AGERA), prepared western blots, and performed immunodetection with the S830 antibody (epitope: exon 1 HTT). We found that HDAC4 knock-down reduced the aggregate load in the cytoplasm but not in the nucleus of Dble::R6/2 mice at both 4 (Figure S2D) and 9 (Figure 2F) wk of age. Consistent with this, we found that the cytoplasmic steady-state levels of HDAC4 were significantly reduced in Dble::R6/2 mice as compared to R6/2 at both 4 (Figure S2E) and 9 (Figure 2G) wk of age. The purity of the cellular fractions was validated by immunoblotting with α-tubulin and histone H3 antibodies (Figures 2F and S2D).

12The R6/2 colony was maintained by backrossing to (CBA/Ca×C57BL/6J)F1 mice and the Hdac4 knock-out mice had been bred to the same F1 background for more than six generations. We know from having bred R6/2 mice for 99 generations and from multiple experiments that the differential segregation of CBA/Ca and C57BL/6J alleles has no effect on HD-related phenotypes in R6/2 mice,–. However, the Hdac4 null allele had been created on a 129S mouse strain background, and it is inevitable that even after backcrossing to (CBA/Ca×C57BL/6J)F1 mice for multiple generations, the Hdac4 null allele would be retained in a genomic region of 129S DNA that had not been removed by recombination. Therefore, it was possible that the observed effects might be due to genetic variation in the 129S Hdac4-linked haplotype rather than through a reduction in HDAC4. To rule out this scenario, we identified an Hdac4-linked single nucleotide polymorphism (SNP) that was polymorphic between 129S and both the C57BL/6J and CBA/Ca strain backgrounds. We crossed R6/2 mice to (129S×C57BL/6J)F1s and identified R6/2 progeny that either did not carry or were heterozygous for the 129S SNP (n = 7/genotype). The heterozygous mice contained the 129S haplotype with a wild-type Hdac4 allele. Seprion ELISA on 9-wk-old cortex showed that the 129S Hdac4-linked haplotype did not modify aggregate load in R6/2 mice (Figure S2F), confirming the role for HDAC4 in aggregate reduction.

13To further understand the nature of the reduction in aggregate load by HDAC4 knock-down, we reasoned that HDAC4 might associate with HTT. Hence, we employed an in vitro GST pull-down assay and found that HDAC4 interacted specifically with exon 1 HTT containing a 53 polyQ tract but not with its 20 polyQ counterpart (Figure 3A). To determine whether HDAC4 associates with endogenous HTT, we immunoprecipitated HTT (2B7, epitope 1–17) or HDAC4 (DM-15) from brain lysates of 8-wk-old WT (7Q), HdhQ150 heterozygous, and HdhQ150 homozygous mice and immunoblotted with the MW1 (soluble mutant HTT), MAB2166 (soluble wild type and mutant HTT), and H-92 (HDAC4) antibodies. We found that mutant but not wild-type HTT co-immunoprecipitated with HDAC4 (Figure 3B). To investigate the effect of polyQ length on this interaction, we repeated the experiment with lysates from 8-wk-old heterozygous knock-in mice carrying polyQ tracts of 20 (HdhQ20) or 80 (HdhQ80). We found that HDAC4 could immunoprecipitate HTT containing 80 but not 20 glutamines (Figure 3C), consistent with the in vitro pull-down data. The sequence of HDAC4 is very similar to the class IIa member HDAC5. Therefore, to investigate the specificity of these interactions, we repeated the in vitro and in vivo immunprecipitations with an antibody specific to HDAC5. Although there was a weak interaction between exon 1 HTT and HDAC5 by in vitro pull-down (Figure 3A), this was not specific to mutant HTT, as was the case for HDAC4, and HDAC5 failed to immunoprecipitate HTT from brain lysates (Figure 3D).

14To further explore this association between HDAC4 and mutant HTT, we performed immunohistochemistry to determine whether HDAC4 co-localized with nuclear and/or cytoplasmic inclusions. For this purpose, we validated a number of commercially available antibodies and found CS2072 to be specific for HDAC4 by fluorescent immunolabelling as it gave no signal on HDAC4KO P3 brain sections (Figure S3A). Consistent with our western blot results, HDAC4 was localised to the cytoplasm appearing as a punctate pattern in adult brains (Figures 3E and S3B and S3C). This cytoplasmic localisation of HDAC4 is supported by its co-localisation with the synaptic markers synaptophysin and PSD95 (Figure S3C). Confocal microscopy showed that the S830 HTT antibody detects huntingtin aggregates in R6/2 and HdhQ150 brains and that HDAC4 co-localised with some but not all cytoplasmic inclusions (Figures 3E and S3B) in both cases. Taken together, our data indicate that HDAC4 associates with soluble mutant HTT in vivo and co-localizes with cytoplasmic inclusions in all brain regions studied.

15Transcriptional dysregulation is a well-documented molecular characteristic of HD pathogenesis. A comparative study of the striatal transcription profiles of seven mouse models and HD post mortem brains showed that the dysregulated signature in R6/2 and HdhQ150 models was highy comparable and in both cases more closely replicated that observed in human HD tissue than that of the other mouse models. HDAC inhibitors were initially pursued as a therapy for HD because of their potential for reversing these transcriptional changes. Therefore, we performed Affymetrix microarray profiling of 9- and 15-wk cortex for WT, Hdac4Het, R6/2, and Dble::R6/2 mice to assess whether HDAC4 knock-down might rescue the global transcriptional dysregulation that occurs in R6/2 mice. As expected the cortical expression profile was profoundly changed between WT and R6/2 mice by 9 wk of age and was further dysregulated at 15 wk (Figure 4A). However, comparison of the R6/2 and Dble::R6/2 profiles indicated that only a very small number of probe sets were predicted to be differentially expressed at statistically significant levels (Figure 4A). This suggested that reduction in HDAC4 had not served to rescue transcriptional dyregulation. We used Taqman qPCR to validate the predicted changes in gene expression between R6/2 and Dble::R6/2 cortex. We were only able to detect statistically different expression levels of small effect size for Secis and Casc4, and in both cases this was in the opposite direction of that predicted by the arrays (Figure 4B).

16Dysregulation of Bdnf promoter transcripts is a well-characterised hallmark of disease progression in HD, and restoration of Bdnf levels has been shown to correlate with phenotypic improvements in HD mouse models. As Bdnf probe sets were not represented on the arrays, we used Taqman qPCR to measure the levels of multiple Bdnf promoter trancripts as well as the coding exon (Bdnf-b) in cortex at 15 wk of age. We found that HDAC4 reduction increased Bdnf-b levels in WT cortex and almost restored the R6/2 dysregulated Bdnf transcripts to WT levels in Dble::R6/2 mice (Figure 4C).

17MSNs in symptomatic R6/2 mice show pronounced morphological abnormalities, including dendritic shrinkage and spine loss. Largely consistent with these anatomical changes, at a behaviorally symptomatic age, R6/2 MSNs display a marked increase in membrane resistance, depolarization of the resting membrane potential (RMP), and an increased intrinsic excitability in response to current injection. These phenotypes indicate that the R6/2 MSNs are hyperexcitable relative to MSNs in WT animals. In addition, symptomatic R6/2 mice display a progressive impairment in corticostriatal connectivity. In combination, this reduction of cortical input, coupled with the abnormal excitability of the MSNs within the R6/2 striatum, will have serious consequences for appropriate striatal information processing and resultant basal ganglia output. These features likely contribute and in part underlie the impaired behavioral function exhibited by these mice. We determined the extent of functional improvement in MSNs from Dble::R6/2 as compared to R6/2 mice at 7–8 and 12 wk of age. As previously published, R6/2 MSNs exhibited a higher membrane resistance than those from WT and Hdac4HET mice at both ages, which was restored to WT levels in Dble::R6/2 mice (Figure 5A,E).

18While R6/2 MSNs at 7–8 wk of age were not significantly depolarized (Figure 5B), by 12 wk the RMP was depolarized by approximately 3 mV relative to WT MSNs, and this was rescued in the Dble::R6/2 mice (Figure 5F). Despite previous reports of reduced rheobasic current (minimum current injection required to elicit an action potential) in R6/2 mice, this was a modest phenomenon in our hands and the reduction of HDAC4 in the Dble::R6/2 mice had no effect (Figure 5C,G). Action potential amplitude was, however, significantly reduced in R6/2 compared to WT and Hdac4HET at both ages and was fully restored in the Dble::R6/2 mice (Figure 5D,H). To assess an improvement in corticostriatal transmission, glutamatergic excitatory postsynaptic currents (EPSCs) were evoked by stimulating layer V cortical afferents innervating the striatum. In both 7–8 and 12 wk age groups, R6/2 mice showed significant reduction in EPSC amplitude for any given stimulus intensity applied. There was no difference in evoked EPSCs per genotype between the two age groups, and the data were therefore pooled (n = 28–32 neurons per genotype; Figure 5I). In the Dble::R6/2 mice, a significant restoration of evoked EPSC amplitude compared to R6/2 was noted. To further delineate the locus of this improvement, a paired-pulse stimulation paradigm (20 ms interstimulus interval) was employed to specifically assess glutamate release probability. R6/2 corticostriatal synapses displayed higher paired-pulse ratios than WT synapses, and this was significant in the 12-wk dataset, indicative of impaired glutamate release in this age group. This was fully restored in the Dble::R6/2 mice (Figure 5J). In agreement with the reduction in evoked glutamate release in R6/2 corticostriatal synapses, quantal glutamate release within the striatum, which can arise from release at both thalamostriatal and corticostriatal presynaptic terminals, was severely depressed in the 7–8-wk-old mice [mean frequency of mEPSCs in MSNs from WT = 2.3±0.25 Hz (n = 12), Hdac4HET = 1.87±0.38 Hz (n = 11), R6/2 mice = 0.13±0.03 Hz (n = 11)]. The Dble::R6/2 mice showed a profound rescue in this phenotype, with a mean frequency of mEPSCs of 1.1±0.14 Hz (n = 12) (significance against R6/2 frequency assessed by Kolmogirov–Smirnoff analysis p<0.0001; Figure 5K,L). There was no change in the amplitude of mEPSCs in R6/2 compared to WT mice, suggesting that postsynaptic 2-amino-3-(5-methyl-3-oxo-1,2-oxazole-4-yl) propanoic acid (AMPA) receptor function was not impaired in the HD model, and did not contribute to the impairment in glutamatergic transmission. mEPSC amplitude was unchanged in the Hdac4HET or Dble:R6/2 animals relative to WT or R6/2 mice (Figure 5M).

19In conclusion, the combined restoration of the membrane properties of MSNs from the Dble:R6/2 mice alongside the improvement in glutamatergic cortical input to the striatum would be expected to significantly improve striatal information processing, normalize aberrant basal ganglia output, and result in an improvement in behavioral function.

20In order to evaluate whether the molecular and electrophysiological changes that had been detected in the Dble::R6/2 mice might ameliorate HD-related behavioural and physiological phenotypes, we employed a set of quantitative, well-characterised tests. Mice were well matched for CAG repeat length (Table S1), and phenotypic parameters were measured from 4 to 15 wk of age in WT, Hdac4HET, R6/2, and Dble::R6/2 mice. All analyses were performed blind to genotype, and in all cases, the progression of R6/2 phenotypes was consistent with previous reports.

21Rotarod performance is a sensitive indicator of balance and motor coordination that has been reliably shown to decline in R6/2 mice. In line with previous results, R6/2 rotarod performance was impaired by 8 wk (p<0.001) and deteriorated further with age (Figure 6A). HDAC4 knock-down had no impact on the performance of WT mice. However, it significantly improved R6/2 rotarod performance at all ages and delayed the progression of rotarod impairment by 1 mo to the extent that 12-wk-old Dble::R6/2 mice performed as well as 8-wk-old R6/2 mice (Figure 6A).

22At 14 wk of age, close to end stage disease, the deterioration in the appearance of the Dble::R6/2 mice was considerably less marked than that of R6/2. To document this, we performed a modified SHIRPA analysis and measured 16 phenotypic parameters (Table S2), for which nine gave a positive score in R6/2 as compared to WT mice. In all cases, the appearance of these phenotyes was improved in the Dble::R6/2 mice when compared to R6/2 (Figure 6B). In particular, piloerection and tremor could not be detected in Dble::R6/2 mice, and hunched back and body tone were vastly improved (Figure 6B). The phenotypic improvements are evident in Videos S1 and S2, and in the light of these, it was surprising that we were unabe to detect an any marked decrease in body weight loss (Figure 6D). The dramatic improvement in the appearance of the mice led us to assess whether HDAC4 knock-down had pro-survival effects, and we found that it extended the lifespan of R6/2 mice by approximately 20%, p = 0.0004 (log-rank test) (Figure 6C).

23Brain weight was measured for cohorts of mice that were culled at 8, 12, and 15 wk of age, and as previously described, there was a progressive decrease in the weight of R6/2 brains as compared to WT. HDAC4 knock-down resulted in a very slight but statistically significant increase in brain weight at 9 and 12 but not at 15 wk of age (Figure 6E).

24There is mounting support for the use of HDAC inhibitors in the treatment of a wide range of brain disorders, predominantly aimed at pathological alterations in the brain transcriptome. HDAC4 is a transcriptional repressor that is normally retained in the cytoplasm, but localises to the nucleus upon de-phosphorylation,. In line with previous reports, we found that HDAC4 was located in the cytoplasm in mouse brain and showed that it did not relocate to the nucleus during disease progression. We demonstrated that HDAC4 associates with mutant HTT in a polyQ-dependent manner in vivo, consistent with an association that occurs between the polyQ stretch in HTT and the Q-rich domain of HDAC4. In line with these observations, we found that HDAC4 co-localised with cytoplasmic inclusions in both R6/2 and knock-in HD mouse models. The genetic knock-down of HDAC4 led to a significant delay in the cytoplasmic aggregation process and led to a significant restoration of synaptic function. At a physiological level, knock-down of HDAC4 extended lifespan and partially restored motor coordination and other neurological phenotypes. This suggests that a cytoplasm-based pathophysiological mechanism contributes to key aspects of neurodegenerative phenotype observed in HD.

25A cytoplasmic mechanism of action for HDAC4-mediated beneficial effects was unexpected. In general, HDACs have been pursued as therapeutic targets because of their impact on epigenetics and transcription. HDAC4 is known to repress MEF2 in muscle, and it has been proposed that HDAC4 binds to HDAC3 to activate its deacetylase domain. However, we found no impact on global transcriptional dysregulation upon HDAC4 knock-down in R6/2 mice, consistent with a predominantly cytoplasmic localisation of HDAC4. Surprisingly, the absence of HDAC4 in knock-out postnatal brain tissue had little effect on the brain transcriptome. In fact, it has been shown that HDAC4 does not function as a lysine-deacetylase, and consistent with this, we found that HDAC4 knock-out had no effect on global acetylation in brain in vivo. Our demonstration that the dysregulation of Bdnf promoter transcripts was alleviated in the double mutant mice is consistent with a cytoplasmic-based mechanism of action. Bdnf expression is repressed by RE1 silencing transcription factor (REST), which is retained in the cytoplasm in a complex containing HTT and which translocates to the nucleus in response to HD pathology. The cytoplasmic aggregation process initiates prior to a reduction in Bdnf transcripts in R6/2 and other HD mouse models, and therefore it is not surprising that a delay in this cytoplasmic pathology might in turn result in a delay in Bdnf dysregulation.

26Our demonstration that a delay in the cytoplasmic aggregation process has beneficial consequences is supported by the published correlation between the appearance of neuropil aggregates and disease progression and is consistent with previous predictions. In the human HD post mortem brain, neuropil aggregates are much more common than nuclear inclusions, present in a large numbers, and potentially associated with onset of clinical symptoms. Our data suggest that HDAC4 might modulate HTT aggregation through a direct interaction serving to template or nucleate soluble HTT. Alternatively, given that HDAC4 may self-aggregate, it could influence HTT aggregation indirectly through perturbation of proteostasis networks. The delay in HTT aggregation afforded by a reduction in HDAC4 was most pronounced in presymptomatic and early-stage disease and diminished with disease progression, presumably reflecting changes in aggregation kinetics. This association between phenotypic improvements and a shift from aggregated to soluble mutant exon 1 HTT is consistent with our previously published in vivo studies,. Our data provide no information as to the aggregate species that is toxic, or as to whether all species of aggregates have detrimental consequences, but only indicate that shifting the equilibrium toward soluble HTT is beneficial in vivo.

27The improvement in synaptic function as a consequence of HDAC4 knock-down was not related to a restoration in the expression level of dysregulated synaptic transcripts. Instead, it may act through reducing cytoplasmic aggregation as neuropil aggregates have been shown to inhibit axonal transport, synaptic function, and glutamate release in HD fly models. The restoration of corticostriatal synaptic function demonstrated that the reduction of HDAC4 has functional consequences in the brain. However, in this study, HDAC4 was ubiquitously knocked down, and as HD has a peripheral component to its pathophysiology, it is conceivable that the reduction of HDAC4 also had beneficial consequences in tissues other than the brain. Given that HDAC4 has well-established functions in muscle, that muscle atrophy is a major symptom of HD, and that HDAC4 has been linked to disease progression in an ALS mouse model, we are currently investigating the extent to which HDAC4 reduction in muscle might contribute to the improved HD phenotypes.

28The administration of HDAC inhibitors has been shown to improve disease phenotypes in a wide range of HD models. To better understand which HDACs, when inhibited, are most responsible for these beneficial consequences, we embarked on a series of genetic manipulations in HD mouse models. In this article we show that the genetic knock-down of HDAC4 delayed cytoplasmic aggregation, improved synaptic function, and improved disease phenotypes. In contrast, the genetic knock-down of HDAC3 and the class IIa members HDAC7, HDAC5, and HDAC9 (our unpublished data) had no effect on R6/2 phenotypes. Strikingly, we recently showed that administration of SAHA caused a reduction in HDAC2 and HDAC4 at the protein but not RNA level in some R6/2 brain regions and that this correlated with a reduction in aggregation and a restoration of cortical Bdnf transcripts. Therefore, we speculate that the beneficial effects of SAHA were at least in part transmitted through the down-regulation of HDAC4 via a mechanism not related to its enzyme activity.

29Perhaps the best validated therapeutic target for HD is the HTT protein, and the reduction of HTT through gene silencing is being developed using both antisense oligonucleotides and RNAi –. However, delivery to the brain is a major challenge for these approaches, and as HTT has many essential functions, the potential liability of decreasing HTT to a detrimental level cannot be ignored. We have shown that reduction of HDAC4 shifts the ratio from aggregated to soluble HTT and therefore acts directly on the HD mutation. Our demonstration that HDAC4 levels can be decreased through the administration of a small brain-penetrant molecule (SAHA) is extremely promising as more selective inhibitors (e.g., specific to HDAC4 or class IIa enzymes) may have similar effects, making it possible to target HTT aggregation with a small molecule. Finally, these findings may have wider implications as HDAC4 is a component of Lewy Bodies in Parkinson's disease brains and administration of SAHA improved the synaptic plasticity and learning behaviour in an Alzheimer disease model.

30All animal work was approved by the local ethics committees and was performed in accordance with UK Home Office regulations or the Swiss Law (Kantonales Veterinäramt Basel-Stadt).

31Hemizygous R6/2 mice were bred by backcrossing R6/2 males to (CBA×C57BL/6)F1 females (B6CBAF1/OlaHsd, Harlan Olac, Bicester, UK). Similarly, the Hdac4 knock-out colony was maintained by backcrossing heterozygous males to B6CBAF1/OlaHsd females. HdhQ150 homozygous mice on a (CBA×C57BL/6)F1 background were obtained by intercrossing HdhQ150 heterozygous CBA/Ca and C57BL/6J congenic lines as described previously. The HdhQ20 and HdhQ80 mice were from CHDI colonies at The Jackson Laboratory (Bar Harbor, ME) and maintained on a C57BL/6/J background. 129S2/SvHsd females were from Harlan Olac. The cross between HdhQ150 and Hdac4HET mice, both on a C57BL/6 background, was performed at Novartis. All animals had unlimited access to food and water, were subject to a 12-h light/dark cycle, and housing conditions and environmental enrichment were as previously described. The R6/2 colony was kept on breeding chow (Special Diet Services, Witham, UK).

32Genomic DNA was isolated from an ear-punch. R6/2 and HdhQ150 mice were genotyped by PCR, and the CAG repeat length was measured as previously described. PCR conditions for genotyping Hdac4 knock-out mice were for WT band: Fw: CTTGTTGAGAACAAACTCCTGCAGCT, Rw: AGCCCTACACTAGTGTGTGTTACACA; for Hdac4 mutant band: Fw: AGCCCTACACTAGTGTGTGTTACACA, Neo Rw: CCATGGATCCTGAGACTGGGG.

33Cycling conditions were 4 min at 95°C, 35× (30 s at 95°C; 45 s at 60°C; 2 min at 72°C), 10 min at 72°C using Taq polymerase and buffer from Promega. The HdhQ20, HdhQ80 mice were genotyped as described using the Hotstart polymerase (Thermoscientific).

34Dissected tissues were snap frozen in liquid nitrogen and stored at −80°C until further analysis.

35All behavioural tests were performed as previously described, and the data were analysed by repeated measures general linear model ANOVA with the Greenhouse Geisser post hoc test using SPSS software. Pair-wise statistical comparisons were corrected for multiple comparisons using Bonferroni post hoc test in SPSS. Survival was assessed blind to genotype, and mice were euthanized when they reached end-stage disease. The data are presented as Kaplan–Meier cumulative survival functions and statistically analysed by the log-rank test.

36Total RNA was extracted with the mini-RNA kit accordingly to manufacturer instructions (Qiagen). Reverse transcription (RT) was performed using MMLV superscript reverse transcriptase (Invitrogen) and random hexamers (Operon), and all Taqman-qPCR reactions were performed using the Chromo4 Real-Time PCR Detector (BioRad) as described. Expression level of the gene-of-interest was normalised to the geometric mean of three endogenous housekeeping genes (Primer Design) as described. The primer and probe sequences are detailed in Table S3.

37For the Affymetrix arrays, biotinylated cRNAs were prepared from 200 ng total RNA using the GeneChip 3′ IVT Express Kit (Affymetrix) following the manufacturer's instructions. cRNA (15 µg) was hybridized to GeneChip Mouse Genome 430 version 2.0 Arrays (Affymetrix) and processed, stained, and scanned according to the manufacturer's recommendations. The quality of input RNAs and cRNAs was verified with the Bioanalyzer 2100 (Agilent Technologies) before use. Microarray quality control was performed using the software package provided on RACE. Chips with a median normalized unscaled standard error greater than 1.05 were excluded. Affymetrix annotations (version 3.0) were used for probeset-to-gene assignments. Two-tailed t test was performed to assess the differences in gene expression between groups for each genotype (WT n = 8; R6/2 n = 9; Hdac4HET n = 8; Dble n = 9). Corrections for multiple testing were performed using the false discovery rate (FDR) according to Benjamini and Hochberg with a significance threshold of p<0.05. The array datasets can be found at NCBI GEO accession number GSE38237.

38All primary and secondary antibodies used in this study are presented in Table S4. Preparation of protein lysates and western blotting were as described previously. In general, 20 µg protein lysate was fractionated on 10% SDS-PAGE gels. Aggregates were captured in Seprion ligand-coated plates (Microsens) and detected using the S830 sheep polyclonal or MW8 mouse monoclonal antibodies as described.

39Sample preparation and the TR-FRET assay were performed as previously described.

40Nuclear and cytoplasmic fractions were prepared as previously described, and their purity was determined by immunoblotting with antibodies to anti-histone H3 and α-tubulin. The AGERA assay was performed as described. In general 100 µg of nuclear or cytoplasmic fractions, isolated from whole snap frozen brains, were loaded in nonreducing Laemmli buffer onto 1.5% agarose gels supplemented with 0.1% SDS and run at 3 V/cm followed by western blotting and immunodetection with anti-HTT (S830) antibodies. The high molecular weight protein marker was from Invitrogen.

41Co-immunoprecipitation was performed as previously described. Briefly, protein lysates were prepared from whole brains in HEPES buffer and incubated with protein-G Dynabeads (Invitrogen) overnight at 4°C on a rotating platform.

42The full-length mouse Hdac4 gene in pCMV6 was obtained from Origene. This was amplified by PCR using high fidelity polymerase (Roche) and subsequently cloned into pCR2-Topo (Invitrogen) accordingly to the manufacturer's instructions. The detailed protocol used for GST pull-downs is available in the Text S1.

43For immunohistochemical studies, brains were frozen in isopentane at −50°C and stored at −80°C until further analysis. We cut 10–15 µm sections using a cryostat (Bright instruments), air dried and immersed them in 4% PFA in PBS for 15 min, and washed them for 3×5 min in 0.1% PBS-Triton X-100. Blocking was achieved by incubation with 5% BSA-C (Aurion) in 0.1% PBS-Triton X-100 for at least 30 min at RT. Immunolabelling with primary antibodies in 0.1% PBS-Triton X-100, 1% BSA-C (HDAC4, S830) was completed by overnight incubation in a humidity box at 4°C. Sections were washed 3× in PBS, incubated for 60 min at RT in a dark box with the appropriate combinations of secondary antibodies diluted in PBS, washed 3× in PBS, and counterstained with DAPI (Invitrogen). Sections were mounted in Vectashield mounting medium (Vector Laboratories). Sections were examined using the Leica TCS SP4 laser scanning confocal microscope and analysed with Leica Application Suite (LAS) v5 (Leica Microsystems, Heidelberg, Germany).

44Detailed procedures for acute striatal slice preparation, patch-clamp recordings, and the isolation of miniature excitatory postsynaptic currents (mEPSCs), along with the appropriate statistical analysis, can be found in Text S1.

45All data were analysed with Microsoft Office Excel and two-tailed Student's t test or as otherwise stated.