Brain gene expression correlates with changes in behavior in the R6/1 mouse model of Huntington’s disease


  • A. Hodges,

    1. Department of Psychological Medicine, Wales School of Medicine, Cardiff University, Cardiff, United Kingdom
    2. Present address: MRC Centre for Neurodegeneration Research, Department of Psychological Medicine, Box PO 70, Institute of Psychiatry, King’s College London, De Crespigny Park, London, United Kingdom
    Search for more papers by this author
  • G. Hughes,

    1. Department of Psychological Medicine, Wales School of Medicine, Cardiff University, Cardiff, United Kingdom
    Search for more papers by this author
  • S. Brooks,

    1. Brain Repair Group, School of Biosciences
    Search for more papers by this author
  • L. Elliston,

    1. Department of Psychological Medicine, Wales School of Medicine, Cardiff University, Cardiff, United Kingdom
    Search for more papers by this author
  • P. Holmans,

    1. Department of Psychological Medicine, Wales School of Medicine, Cardiff University, Cardiff, United Kingdom
    2. Bioinformatics and Biostatistics Unit, Wales School of Medicine
    Search for more papers by this author
  • S. B. Dunnett,

    1. Brain Repair Group, School of Biosciences
    Search for more papers by this author
  • L. Jones

    Corresponding author
    1. Department of Psychological Medicine, Wales School of Medicine, Cardiff University, Cardiff, United Kingdom
    2. Institute of Medical Genetics, Wales School of Medicine, Cardiff University, Cardiff, United Kingdom
    Search for more papers by this author

*L. Jones, Department of Psychological Medicine, Henry Wellcome Building, Wales School of Medicine, Cardiff University, Cardiff, CF14 4XN, UK. E-mail:


Huntington’s disease (HD) is an inherited neurodegeneration that causes a severe progressive illness and early death. Several animal models of the disease have been generated carrying the causative mutation and these have shown that one of the earliest molecular signs of the disease process is a substantial transcriptional deficit. We examined the alterations in brain gene expression in the R6/1 mouse line over the course of the development of phenotypic signs from 18 to 27 weeks. Changes in R6/1 mice were similar to those previously reported in R6/2 mice, and gene ontology analysis shows that pathways related to intracellular and electrical signaling are altered among downregulated genes and lipid biosynthesis and RNA processes among upregulated genes. The R6/1 mice showed deficits in rotarod performance, locomotor activity and exploratory behavior over the time–course. We have correlated the alterations in gene expression with changes in behavior seen in the mice and find that few alterations in gene expression correlate with all behavioral changes but rather that different subsets of the changes are uniquely correlated with one behavior only. This indicates that multiple behavioral tasks assessing different behavioral domains are likely to be necessary in therapeutic trials in mouse models of HD.

Huntington’s disease (HD) is a neurodegeneration caused by an expanded CAG repeat in the HD gene. Patients typically have a characteristic movement disorder, decline of cognitive function and variable psychiatric manifestations. Huntington’s disease usually onsets in midlife and is relentlessly progressive with death following inexorably 10–20 years later (Bates et al. 2002). The neurodegeneration starts first and is most severe in the caudate nucleus and putamen, with first loss of the striatal medium spiny projection neurons observed before the manifestation of clinical symptoms (Vonsattel et al. 1985).

A series of mouse models of HD have been generated that recapitulate aspects of the clinical disease, including its progression and the observation of huntingtin-containing neuronal inclusions (Beal & Ferrante 2004; Davies et al. 1997; Mangiarini et al. 1996). Many of the mice have a movement disorder that can be measured by performance on an accelerating rotarod (Bates & Murphy 2002; Carter et al. 1999) and in addition, where it has been examined, most mice also have cognitive deficits (Lione et al. 1999; Spires et al. 2004; Trueman et al. 2007). The timing of onset and progression of the phenotype depends on the mouse line, with longer repeats and the expression of truncated fragments of the huntingtin protein generally giving earlier deficits (Hersch & Ferrante 2004).

One of the earliest observable molecular changes to occur in the brain of HD model mice is an alteration in gene expression (Chan et al. 2002; Luthi-Carter et al. 2000, 2002a, 2002b). In R6/2 mice, changes in gene expression were already present at 6 weeks (Luthi-Carter et al. 2002a). In R6/1 mice, downregulation of caudate-specific genes was first seen in 10-week-old mice and most such genes were further downregulated by 6 months (Desplats et al. 2006). Human HD caudate also shows substantial alterations in gene expression that cannot be accounted for by cell loss (Hodges et al. 2006).

In the present study, we examine the relationship between changes in gene expression and the emergence of the behavioral phenotype in the R6/1 mouse model of HD. The R6 models carry a truncated human HD gene encoded by exon 1 of the human gene, with expanded CAG repeats of >150 CAG in the R6/2 and around 115 CAG in the R6/1 model (Mangiarini et al. 1996). The R6/2 model is the best characterized and has a very rapid disease course; the R6/1 model has been less well characterized but has a slower disease trajectory, with death not occurring until at least 8 months, although the phenotype seems similar to that of R6/2. Motor defects can be detected around 3 months in R6/1 mice (Hansson et al. 2001) and detailed investigation showed that exploratory behaviors were altered as early as 4 weeks (Bolivar et al. 2004). We carried out a battery of behavioral tests longitudinally from 18 weeks of age and we also examined brain gene expression at 18, 22 and 27 weeks of age in these animals.



The mice used in this study came from our colony of R6/1 mice routinely maintained by breeding male R6/1s from our colony with B6CBA F1 (isolator breed: offspring of a cross between the C57BL/6JOlaHsd inbred female and the CBA/CaOlaHsd inbred male) from Harlan. The study was based on 16 R6/1 mice and 16 wild-type litter mate offspring (7 males and 9 females of each genotype) from five closely spaced litters. The transgenic R6/1 mice had CAG repeat lengths of 126.5 ± 3.0 (range 122–131 repeats). The mice were housed in group cages three to four mice per cage, on sawdust flooring, and containing a cardboard tube to provide modest enrichment. Food and water were available ad libitum. The behavioral batteries were applied over 1-week blocks of testing at 18, 22 and 27 weeks of age, following which three mice from each group were killed for gene expression analysis. At the end of the experiment, remaining mice were killed for histological confirmation of huntingtin inclusion pathology.

Gene expression

Three sex-matched mice from each genotype (R6/1 and wild type) were killed at 18, 22 and 27 weeks of age. Brain samples were collected, snap frozen in liquid nitrogen and stored at −80°C. Total RNA was extracted from one hemisphere for microarray analysis, as described previously (Morton et al. 2005a). Microarray analysis of global gene expression was carried out using 10 μg of total RNA and murine genome 430 2.0 Arrays, according to the GeneChip Expression Analysis Technical Manual (Affymetrix UK Ltd, High Wycombe, UK). Quantitative real-time polymerase chain reaction (PCR) was carried out essentially as in Morton et al. (2005a). mRNA expression levels were quantified using QuantiTect SYBR Green RT-PCR Kit (Qiagen, GmbH, Hilden, Germany) for Penk 1, Arpp19, Gria 3, Adnp and Fads 1 according to the manufacturer’s instructions. Gene-specific oligonucleotide primer pairs spanning exons were designed using primer3 software ( and using sequence data from the NCBI database. Primers were as in Morton et al. (2005a) and in addition the following primers were purchased from Invitrogen (Carlsbad, CA, USA): Fads 1 AAGCACATGCCATACAACCA and CAGCGGCATGTAAGTGAAGA; Adnp AGAAAAGCCCGGAAAACTGT and AAGCACTGCAGCAAAAAGGT.

DNAsed total RNA at 20, 50 or 100 ng/μl for each sample, whichever was optimal, was accurately diluted using a NanoDrop® ND 1000A Spectrophotometer (NanoDrop Technologies, Inc., Wilmington, DE, USA) before being amplified in duplicate. The default amplification program using the QuantiTect SYBR Green RT-PCR Kit was 50°C 30.0 min, 95°C 15 min and 94°C 15 seconds, 60°C 30 seconds, 72°C 30 seconds for 35 cycles except for Adnp and Fads 1, which had an annealing temperature of 58°C.



The rotarod apparatus (Ugo Basile, Biological Research Apparatus, Varese, Italy) was used to measure the animals’ balance and motor coordination, as previously described (Brooks et al. 2004b; Carter et al. 1999). The 3-cm-diameter beam was coated in soft rubber (from a bicycle tire inner tube) to prevent the animals from clinging to the beam in the event of them failing to maintain their balance. The mice were trained on the apparatus across 2 days prior to the first test exposure. On the test day, each mouse underwent two test runs, and the mean score of these runs was used for statistical analysis. Mice were initially placed on the stationary beam and then the beam started to rotate at a constantly accelerating rate of 3 r.p.m./30 seconds to a maximum velocity of 44 r.p.m. All mice fell before the maximum velocity was reached, and a trip switch records the latency to fall since the start of the trial. Approximately 45–60 min separated the two trials on each test day, which were spent in the home cage.

Locomotor activity

General levels of 24-h motor activity were measured using an Ethovision automated video recording system (Noldus Information Technology, Wageningen, the Netherlands), as previously described. Levels of motor activity were also assessed after 30 min in the arena as a measure of nonhabituated motor activation. A single open field area (100 × 100 × 29 cm) with white walls and floor (to aid the automated detection of the animals) was divided into four separate and equally sized square arenas (50 × 50 × 29 cm) with the use of wooden partition slides, allowing four mice to be tested at once. To prevent the mice from escaping a wire, meshed roof was placed over the top of the open field. Each of the partition walls had a water bottle mounted on it to allow the mice access to water over the 24-h monitoring period, and pellets of standard laboratory chow (5 g in weight) were randomly placed in each of the arenas. The 24-h monitoring began soon after the 07:00 start of the light period (lights on 07:00 and off at 19:00 daily).

Object recognition test/exploratory behavior

An open field arena (30 × 30 × 15 cm) was constructed of white Perspex with a clear Perspex lid. The floor of the arena was lightly covered with sawdust. The objects that the mice were required to discriminate between consisted of several sets (four identical objects in each set) of identical objects, e.g. two sets of household ornaments, one set of fizzy drink cans and one set of tea cups. The object types were pseudorandomly allocated within and between animal groups so that each object was evenly represented across the experimental conditions. Prior to the onset of each experiment, each mouse was habituated to the test arena, initially as pairs of animals in a 15-min session and then as individuals for a single 10-min session. In both cases, novel objects that were not to be used later in the experiment were placed in the arena to aid habituation.

At the onset of the experiment, the initial object familiarization stage, two identical objects were placed in the arena equidistant from the walls and each other. The mouse was then placed in the arena for a maximum of 5 min or until it had completed 30 seconds of exploration of the objects. Because the objects at this stage were identical, the required 30 seconds of exploration was the sum of the exploration time of both objects. Episodes of object recognition were recorded as the time that the animals’ nose was within approximately 0.5 cm of the object. In some instances, the animal would rear above the object with its nose in the air; this was not deemed to be exploration. After 5 min had elapsed or 30 seconds of exploration had been completed, the animal was removed from the arena and retuned to the home cage.

To assess recognition memory, two time delays were used in the experiment; these were 1 and 4 h after the initial object familiarization stage. For the animals at the youngest age point, these time delays were run as separate experiments with an initial exploration phase for each delay. The results from this initial test suggested that this approach was counterproductive resulting in excessive variance both within and between animals; consequently, at later age points, a single ‘initial exposure’ methodology was adopted. At the 1- and 4-h time-points, the animals were reintroduced to the arena, which again contained two objects, a third (for the 1-h delay) replicate of the original object and a novel object that the animal had never seen before. The animal was permitted to explore the objects for 3 min. Over the course of the 3 min, the time spent exploring each of the objects was recorded separately. After 3 min, the animal was removed from the arena. This procedure was repeated using a 4-h delay.

Startle response

The primary acoustic startle response and the inhibition associated with a prepulse warning stimulus were tested in an automated startle chamber (San Diego Instruments, San Diego, CA, USA) that recorded the startle response by a force-sensitive floor plate to random delivery of 105- and 120-dB 50-msecond bursts of white noise. Prepulse inhibition of the startle response was assessed by delivering brief 20-msecond 85-dB warning stimuli 0–100 mseconds prior to the primary stimulus. Stimulus delivery and data recording was controlled by online connection to a PC microcomputer in a series of counterbalanced trials, as described previously (Brooks et al. 2004a; Carter et al. 1999).

Inverted grid test

The latency of a mouse to fall from an inverted standard laboratory metal runged cage lid was taken as a measure of grip strength. The mouse was placed on the lid, which was then inverted slowly, such that the mouse had to cling to the underside of the lid to prevent itself from falling on to padded landing area below. The time that the animal remained on the inverted cage lid was recorded up to a maximum of 60 seconds when the trial was terminated.

Statistical methods


The data from each dependent behavioral variable was analyzed by three- and four- factor analyses of variance (anovas) (Genstat for Windows v9.1; Lawes Agricultural Trust, Rothamsted, Oxon), with genotype and sex as between-subject variables, and ages plus any test-specific repeated measure as within-subject variables. Data for animals removed for genetic analysis before the second and third tests were entered as missing values, for which the Genstat program corrects, using an unbiased estimator algorithm that corrects for the reduced degrees of freedom that are thereby introduced. Although there were a number of main effects of sex, because of differences in body weight and size, interactions with genotype were small. Consequently, although the factor sex was included in all analyses, the various interactions with this factor are not reported in the Results.

Gene expression

Probeset summary values, normalization and linear modeling were implemented using the Bioconductor software ( Specifically, probeset normalization and summary values were calculated using the Robust multichip average (RMA) from the affy package (Gautier et al. 2004). Tests of differential gene expression by genotype were performed using moderated t-tests in LIMMA (Smyth 2004). Quality control of samples was investigated by examining perfect match density plots, affy QC control probeset summaries, pseudochip RMA weight plots (affyPLM package) and MA plots (Bolstad et al. 2005; Gautier et al. 2004; Irizarry et al. 2003). Quantitative PCR results were analyzed as in Morton et al. (2005a).

Correlation of behavior and expression

To reduce multiple testing, correlation analyses were restricted to gene expression probesets and behaviors exhibiting significant genotype effects (P < 0.001 for probesets, P < 0.05 for behaviors). Behavior/probeset combinations whose P values for correlation satisfied false discovery rate (FDR) < 0.05 (Benjamini & Hochberg 1995) were selected for further analysis (see below).

Variation of behavior/expression correlation with genotype

The effect of genotype on behavior and gene expression for each mouse in the study was tested using linear regression. Two models were fitted, using the behavioral results generated most recently prior to killing as the dependent variable. A flow diagram of the analysis procedure for testing behavior/gene expression correlations is shown in Fig. 2. One model contained genotype main effects only, the second contained main effects of both genotype and gene expression, together with their interaction. The difference in fit between the two models was assessed by anova, providing a test of the effects of expression on behavior after allowing for genotype effects. If the difference in the models was not significant, then the whole of the effect of gene expression on behavior was considered to be as a result of genotype alone. If the difference in the models proved to be significant (P < 0.05), the behavior/gene expression probeset combination was assigned to one of eight bins as follows: first, the significance of the gene expression/genotype interaction term was tested – this measures the difference in correlation between expression and behavior in the two genotype groups. If the interaction term was non-significant (P > 0.05), the behavior/probeset combination was assigned to Bins 1 or 2, depending on the direction of the correlation in the sample as a whole. If the interaction term was significant (P < 0.05), behavior/gene expression correlation was tested in each genotype group separately and the behavior/probeset combination assigned to one of Bins 3–8 depending on the significance and the direction of correlation in each genotype group. Graphs of the correlation between behavior and expression for representative probesets in each bin are shown in Fig. 3.

Figure 2.

Flow diagram of analysis methods used to generate gene expression – behavior correlations. WT, wild type.

Figure 3.

Correlation plots of behavior against brain gene expression in R6/1 mice. These plots are for representative genes from each of the Bins for the rotarod-gene expression correlation. Each point is the data for an individual mouse. The dashed line represents the linear regression of gene expression and behavior for wild-type (WT) mice and the dotted line that for R6/1 mice in each plot.

Overlap in dysregulated probesets between behaviors

The overlap in gene expression associated with each pair of behaviors was tested by comparing the observed number of probes in the overlap to a hypergeometric distribution. The overlap between all three distributions was tested by randomly simulating lists of dysregulated probesets for each behavior equal in length to that observed in the actual data and counting the number of probesets in the overlap. This process was repeated 5000 times to build an expected distribution for the number of probesets in the overlap, and the significance of the observed overlap estimated by comparing it to this distribution.

Significance of overrepresented gene ontology categories

For each gene ontology (GO) category, the total number of probes in that category and the number of probes appearing on a list of differentially expressed probes (P < 0.001) were calculated. A P value for overrepresentation of each category was calculated (Fisher’s exact if either the number of probes on the list or the number not on the list was less than 10, otherwise a Pearson chi square). The number of categories achieving a given P value for overrepresentation was calculated, and significance assessed by permutation (to account for the overlap in categories). The permutation procedure was as follows: a list of differentially expressed probes of equal length to the actual list was generated by sampling probes at random (without replacement). The number of probes on the list for each GO category and hence a P value for overrepresentation was calculated. The number of categories with a P value for overrepresentation less than the specified criterion was counted and compared with that in the actual data. The process was repeated 5000 times. A significantly high number of overrepresented categories were found for both directions of differential expression (up and down).


Behavior is altered in R6/1 mice

R6/1 mice underwent a series of behavioral tests at 18, 22 and 27 weeks of age. On some tests – grip strength, startle/prepulse inhibition and activity over a 24-h period – the R6/1 mice exhibited no differences from the wild-type control group. However, three tests did show significantly different results between wild-type and R6/1 animals: rotarod, exploratory behavior in initial phase of the object recognition task and distance moved in a 30-min period (Fig. 1). Of these, rotarod was the most robustly changed between control and transgenic mice. Specifically, the R6/1 mice were impaired even at the youngest age tested, and the impairment became significantly worse as they aged (Fig. 1a; genotype, F1,28= 25.83, P < 0.001; age × genotype F2,22= 4.16, P < 0.05). In the exploration test, R6/1 animals did not differ from wild-type controls at the 18-week test but spent progressively less time exploring novel objects than did the wild-type mice on the tests at 22 and 27 weeks of age (Fig. 1b; genotype, F1,28= 32.48; F2,27= 30.71; both P = <0.001). Third, R6/1 animals exhibited lower locomotor activity already on the 18-week test than did the wild-type controls (Fig. 1c,d; time spent moving, F1,30= 10.18; distance moved, F1,30= 15.30; both P < 0.001), whereas the time measure remained stable over tests, the total distance moved further declined with age (age × genotype, F1,28= 0.25, NS, and F2,28= 3.68, P < 0.05, respectively).

Figure 1.

Effect of genotype on behavior in the R6/1 mouse.

Gene expression is altered in R6/1 mouse brain

We killed subgroups of mice at the end of each block of behavioral testing at 18, 22 and 27 weeks of age and examined mRNA levels in their brain (Affymetrix MOE430v2 GeneChips). We found substantial differences in gene expression between R6/1 and wild-type mice at all three ages, with no significant effect of time, implying that major changes in gene expression had occurred by 18 weeks. There were 6770 out of 45 101 dysregulated probesets at P < 0.05, adjusted for multiple testing (Benjamini & Hochberg 1995), 15.0% of the total probesets surveyed. For the correlation analysis, we took a more stringent significance threshold of P < 0.001 to reduce the levels of multiple testing, and at this level, 1528 probesets were significantly dysregulated between R6/1 and wild-type brain. Around 40% were downregulated and 60% upregulated. Fold changes in R6/1 mice were significantly correlated with those seen in R6/2 animals at 10 weeks of age (Morton et al. 2005a; r9062= 0.44, P < 10−6). The data are available at the Gene Expression Omnibus accession GSE3621.

We performed quantitative real-time PCR on a selected group of genes altered in expression in these mice. These confirmed the results of the array experiment (Table 1). The actual fold changes observed were rather different between the two techniques but the direction of change was the same. Adnp and Fads were representative genes selected from the behavioral correlation experiment (see Fig. 3). Of the six genes illustrated in Fig. 3, these were the only two expressed at a high enough level for quantitative PCR to be sufficiently sensitive to detect changes in their expression.

Table 1.  Quantitative PCR and MOE430 gene expression change comparisons between R6/1 and wild-type (WT) littermate control mice
GeneQuantitative PCR, R6/1 vs. WTMOE430, R6/1 vs. WT
FCP valueFCP value
  • *

    27-week data only. Across time-points, there is a trend to significance but this only achieves significance at 27 weeks in quantitative PCR data. FC, fold change.

Penk 13.36↓<0.0012.83↓1.6 × 10−9
Arpp191.93↓0.0102.14↓5.4 × 10−9
Gria 3*2.08↓0.0241.41↓1.2 × 10−4
Adnp1.59↑0.0056.96↑1.1 × 10−4
Fads 11.62↓<0.0011.32↓1.9 × 10−5

We examined the function of the dysregulated genes by looking at overrepresentation in the GO database, examining all GO categories (process, function and compartment) for up- and downregulated genes separately. The most significant processes are shown in Table 2 and a full list is available in Table S1. More altered categories are seen than expected by chance for both up- and downregulated genes. More categories and more significant categories were seen among down- than upregulated genes. Among the downregulated genes, there are many functions related to both intracellular and intercellular signaling (Table 2), and notably locomotory behavior is among the processes altered (P = 1.34 × 10−14). The upregulated genes are overrepresented in functions related to RNA biosynthesis and metabolism, and to lipid metabolism, including cholesterol metabolism.

Table 2.  Overrepresentation of GO processes in genes with altered expression in R6/1 mouse brain
GO numberNo. of probesNo. of correlated probesP valueProcess
Downregulated genes
 6816207240.00Calcium ion transport
 7186823620.00G-protein coupled receptor protein signaling pathway
 996875160.00Negative regulation of signal transduction
 7242864585.55 × 10−17Intracellular signaling cascade
 6813283272.55 × 10−15Potassium ion transport
 762687131.34 × 10−14Locomotory behavior
 71651576821.18 × 10−13Signal transduction
 683688123.50 × 10−12Neurotransmitter transport
 7218144165.16 × 10−12Neuropeptide signaling pathway
 7399348263.62 × 10−10Nervous system development
 72161171.44 × 10−9Metabotropic glutamate receptor signaling pathway
 1764155151.89 × 10−9Neuron migration
 7409111111.34 × 10−7Axonogenesis
 6811895451.42 × 10−7Ion transport
 1558147131.89 × 10−7Regulation of cell growth
 64681293562.76 × 10−6Protein amino acid phosphorylation
 469281964.08 × 10−6Regulation of neurotransmitter secretion
 69373772.59 × 10−5Regulation of muscle contraction
 464882663.00 × 10−5Phosphatidylinositol metabolism
 80162965.81 × 10−5Regulation of heart contraction
 425524276.11 × 10−5Myelination
 164861046.35 × 10−5Peptide hormone processing
 350502251.53 × 10−4Embryonic heart tube development
 485142351.92 × 10−4Blood vessel morphogenesis
 427551342.04 × 10−4Eating behavior
 7264476232.84 × 10−4Small GTPase-mediated signal transduction
 76311543.75 × 10−4Feeding behavior
 467777584.59 × 10−4Protein amino acid autophosphorylation
 68879494.63 × 10−4Exocytosis
 66955874.87 × 10−4Cholesterol biosynthesis
 431541644.90 × 10−4Negative regulation of caspase activity
 827785.49 × 10−4G1/S transition of mitotic cell cycle
 161264365.60 × 10−4Sterol biosynthesis
 67252956.00 × 10−4Aromatic compound metabolism
 192221746.29 × 10−4Regulation of metabolism
 71894567.18 × 10−4G-protein signaling, adenylate cyclase activating pathway
 726710499.67 × 10−4Cell–cell signaling
 16621949.86 × 10−4Behavioral fear response
Upregulated genes
 16568288153.54 × 10−11Chromatin modification
 6955443173.96 × 10−8Immune response
 63503110637.45 × 10−7Transcription
 7049852241.67 × 10−6Cell cycle
 63553917741.99 × 10−6Regulation of transcription, DNA dependent
 51301452155.93 × 10−6Cell division
 6281342111.44 × 10−4DNA repair
 303011031.67 × 10−4Cholesterol transport
 86301132.28 × 10−4DNA damage response, signal transduction resulting in induction of apoptosis
 21132.28 × 10−4Mitochondrial genome maintenance
 90481233.01 × 10−4Dosage compensation, by inactivation of X chromosome
 6974363113.34 × 10−4Response to DNA damage stimulus
 7067243105.31 × 10−4Mitosis

Correlation of behavior with gene expression

To identify genes in which changes of expression were associated with the emerging profile of behavioral changes in the R6/1 transgenic mice, we adopted a sequential decision model that classified significantly changed genes into discrete bins dependent upon the nature of the correlation between the change in the gene and the changes in behavior dependent upon mouse genotype and age (see Fig. 2; with further details in Methods). For each individual mouse, data from the behavioral tests that took place immediately before killing of the animal were correlated with gene expression data from brain hemispheres of that mouse.

We used a relatively stringent nominal P value, P = 0.001, as our initial filter for significantly different gene expression, as noted above (Fig. 2). Three behavioral measures were significantly different (P < 0.05) between R6/1 and wild type: rotarod, exploratory behavior and activity. Given the stringent initial cutoff, subsequent criteria for significance were chosen to ensure an overall false discovery rate of 0.05 (Benjamini & Hochberg 1995). Of the 1528 dysregulated probesets, 365 were significantly correlated with performance in one or more of the behavioral tasks: if duplicate probesets for the same gene are removed, this leaves correlations with 331 individual genes. Rotarod correlations were fewest (49 probesets, 36 genes), activity correlated with 158 probesets (131 genes) and exploratory behavior with 158 probesets (133 genes). Three probesets correlated with all three behaviors, more than would be expected by chance (P = 0.013) and 28 correlated with two out of three behaviors (Table S2). The seven genes common to rotarod and exploration and the 21 genes common to exploration and activity do not represent a significant overlap (P = 0.237, P = 0.127, respectively) although the nine genes in common between rotarod and activity show a trend toward significance (P = 0.059).

Although only three gene expressions correlated significantly with all three behaviors, this is more than expected by chance. The genes Cp (encoding ceruloplasmin) and Cdkn1c (encoding cyclin-dependent kinase inhibitor 1C) showed identical patterns of correlation for the rotarod and exploration tasks with better rotarod performance and increased exploratory behavior correlated with reduced expression of these genes in the brains of all animals. The correlation for distance moved also showed this pattern for Cdkn1c. Cdkn1c is imprinted and is likely to encode a tumor suppressor p57Kip2 that operates through binding G1 cyclin/Cdk complexes that negatively regulate cell proliferation (Brakensiek et al. 2005). Ceruloplasmin is a metalloprotein that binds most copper in plasma and is involved in converting toxic ferrous iron to its nontoxic ferric form. Ceruloplasmin is part of the tight regulation system of iron in cells (Harris et al. 2004). Higher Fgfr3 (encoding fibroblast growth factor 3) expression correlated with increased exploration and improved rotarod performance in all animals (Bin 1). For Fgfr3 and Cp, correlations with distance moved were different in that for Fgfr3 increased exploration correlated with decreased R6/1 and increased wild-type gene expression and for Cp increased distance moved correlated with decreased gene expression in both sets of animals but with a larger effect in wild-type animals. Fgfr3 mutations in people cause achondroplasia (Shiang et al. 1994). Fgfr3 appears to be an important growth factor during early embryonic development (Hernandez et al. 2004). Its role in the postnatal human brain is unclear.

The 36 genes correlated with rotarod behavior are shown in Table S3. Bins 1 and 2 are straightforward; in Bin 1, as gene expression increases, the time the animal stays on the rotarod increases, whereas in Bin 2, the opposite effect is observed, increased gene expression correlates with decreased time on the rotarod. These effects are the same in both wild-type and R6/1 animals implying that expression of these genes is related to rotarod performance in all animals. There seem to be no obvious functional links between the genes altered in Bins 1 and 2 with respect to rotarod ability. Ttl participates in microtubule structure, retyrosinating alpha tubulin after depolymerization and may be important in maintaining neural projections (Erck et al. 2003). St3gal5 is an enzyme that makes ganglioside GM3; gangliosides are important in cell recognition and signaling activity, although St3gal5 knockout mice had normal neural function and showed increased insulin sensitivity (Yamashita et al. 2003).

Bin 2 of the rotarod correlation contains more genes than Bin 1 and these include several transcriptionally active genes for instance Zfp28 and Sin3b, although when GO categories are examined, proteins involved in transcriptional processes are significantly overrepresented in the genes upregulated rather than downregulated in R6/1 brain. There is a subunit of the immunoproteasome and a mitochondrial thioesterase as well as ceruloplasmin that may implicate mitochondrial processes in rotarod performance.

Bin 3 genes define a correlation pattern where decreases in gene expression in the wild-type animals correlate with increased time on the rotarod, as in Bin 2. However, unlike Bin 2, this effect is much less in the R6/1 animals, implying that those genes increased in expression in wild-type mice might improve rotarod performance but that the ability to alter the expression of these genes has been lost in R6/1 animals. Consistent with this observation, there are no genes that show a significantly bigger decrease in R6/1 than wild-type animals correlated with increased rotarod performance (Bin 4). Genes in Bin 3 include two that are important in neurodevelopment and maintaining neuronal function, Adnp and Nnat.

Bin 5 contains genes that show decreased expression in R6/1 animals and increased expression in wild-type animals correlated with increased time on the rotarod. The genes normally correlated with improved rotarod performance in mice appear to be downregulated in the R6/1 animals. These genes include Bdnf and two genes involved in ubiquitin-mediated protein degradation. Genes in Bin 6 are those where increased expression in R6/1 brain is correlated with increased time on the rotarod, but the reverse is true for wild-type animals where decreased expression correlates with better performance on the rotarod. One interpretation is that there is increased expression of these genes in R6/1 brain to compensate for other deficiencies and that this is unnecessary in wild-type animals, which are physiologically normal. These genes include some chaperone-type proteins, two proteins involved in lipid metabolism and Ak5, which encodes a protein responsible for regulating adenine-nucleotide phosphate transfer, i.e. the proportions of high-energy intermediates in cells. Bin 8 contains only one known gene and this encodes a subunit of Complex IV of the mitochondrial electron transport chain. Increased gene expression of this gene is correlated with increased time on the rotarod in wild-type mice but no change in R6/1 expression. There are no genes that show increased expression correlated with increased time on the rotarod in R6/1 mice but no change in gene expression in wild-type mice (Bin 7), which could reflect direct compensatory changes.

Relatively few GO functions were significantly overrepresented in the rotarod correlated genes. This is almost certainly because there are only 49 genes correlated with rotarod behavior. Table S4 shows the functions associated with correlation of rotarod behavior including several proteinases, and three genes that are related to energy metabolism and oxidation, iron ion binding, nucleoside triphosphatase activity and oxidoreductase activity.

There are far more genes correlated with performance in the exploration task than with the rotarod in the R6/1 animals (n = 158: Table S5). The majority of these fall into Bins 1 and 2, where increased gene expression correlates with increased exploration in wild-type and R6/1 animals, or where decreased gene expression correlates with increased exploration in all animals (111/145, 77%). There are no genes in Bin 3 where decreased gene expression correlates with increased time exploring with a bigger effect in wild-type animals although there are a small number of genes in Bin 4 where there is a bigger effect in R6/1 animals. These genes are involved with cell polarity and exocytosis. Genes that showed increased expression in R6/1 and decreased expression in wild-type brain with increased exploration include several glycosylated or glycosylating proteins. The opposite pattern in Bin 6 includes Mgll, which is responsible for hydrolysis of triglycerides and may indicate a need for increased energy metabolism in the R6/1 animals. In Bin 7 with a similar pattern except no change in wild-type expression, genes are related to cytoskeletal alterations and include Kalrn, which encodes kalirin, a Hap1 interacting protein (Colomer et al. 1997) that is required for the formation of dendritic spines (Rabiner et al. 2005).

More GO functions were correlated with exploratory behavior than with rotarod activity, partly at least a consequence of having more genes correlated with this behavior (Table S6). There are many categories relevant to intracellular and extracellular signaling activities, especially ion channels. Like the rotarod, this behavior appears to be related to various peptidase activities although these are a different set of degradative pathways.

Like exploration, activity has a relatively large number of correlated genes (n = 158: Table S7). In contrast to exploration, however, Bins 1 and 2 contain proportionately fewer genes (89/157, 56.7%) and there are more in Bins 3, 5, 6 and 7. Bin 3 includes genes that are decreased in expression with increased distance moved with a significantly larger effect in wild-type animals. Some of these genes are transcriptionally active, and some may relate to energy metabolism. Bin 5 contains most genes after Bins 1 and 2 (33 genes) where increased distance moved correlates with decreased gene expression in R6/1 animals and increased gene expression in wild-type animals. These may be genes that are decreased in the large batch of genes downregulated in R6/1 brain but that tend to lead to more activity when increased in the wild-type animals. These include many intracellular signaling molecules, along with some effector molecules. Several transcriptionally active proteins are altered and a couple of genes related to circadian rhythm, known to be altered in many HD model animals (Morton et al. 2005a, 2005b). Genes increased in R6/1 and decreased in wild-type animals fall into Bin 6. Here too there are some transcriptionally active proteins as well as Cbr1 that might be upregulated in response to the downregulation of Cbr3 (Bin 5) as these two genes have similar functions reducing carbonyls to hydroxyl groups and are closely linked genetically (Hattori et al. 2000). These upregulations may be compensatory changes in R6/1 animals. Bin 7 may also contain genes upregulated in R6/1 animals as a compensatory change – these genes are increased in R6/1 animals with increased distance moved but there is no change in wild-type animals.

The GO functions associated with genes correlated with distance moved include the major pathways that transduce external signals in the cell, plus those involved in binding of effector molecules. There is also some evidence for altered energy metabolism at the level of gene expression.


There is a widespread and substantial gene expression dysregulation in human HD (Hodges et al. 2006) and in most of the mouse models of the disease that have been examined (Chan et al. 2002; Luthi-Carter et al. 2000, 2002a, 2002b; Zucker et al. 2005). Such models are being widely used to develop potential therapeutics (Hersch & Ferrante 2004) but it is not clear whether the behavioral tests currently used reflect the complex signs of the human disease. These studies may also identify biomarkers that underlie important behaviors that could be further developed to monitor disease progression. For example, increased ceruloplasmin expression is found in the brains of these mice and correlates with poorer behavioral performance. Ceruloplasmin has been found to be increased in plasma and cerebrospinal fluid of Alzheimer’s disease (AD) patients and may be a biomarker for progression of degenerative processes (Hye et al. 2006).

The R6/1 mouse we used in these experiments shows a defect in rotarod performance by 18 weeks (van Dellen et al. 2000; Hansson et al. 2001). The deficit in R6/1 exploratory activity occurred later consistent with the results of Mazarakis et al. (2005) who saw no differences between 10 and 14 weeks, in contrast to a previous report that showed differences from 4 weeks (Bolivar et al. 2004). We did find changes at 22 and 27 weeks. Differences in the testing protocol may account for these discrepant findings. Mazarakis et al. (2005) examined R6/1 behavioral changes at 10–14 weeks over a number of tasks including open field, rotarod and locomotor activity and saw no differences by 14 weeks. We detected a difference in activity measured by distance moved that was significant at 18 weeks. Several other tasks showed no deficit at all over the time–course of the experiment implying that the deficits in behavior onset at different ages in broad agreement with the previously published work in this line (van Dellen et al. 2000; Mazarakis et al. 2005; Spires et al. 2004) and in R6/2 mice (Carter et al. 1999; Lione et al. 1999).

The gene expression profiles of the R6/1 mouse were similar to those seen in the R6/2 mouse (Luthi-Carter et al. 2002a; Morton et al. 2005a) although occurring later, as expected for this line with delayed appearance of deficits and longer life span (Mangiarini et al. 1996). We used brain hemisphere homogenates to generate gene expression data so we are only seeing the largest alterations: changes that occur in selected neuronal populations or are regionally restricted will not be detected. However, in the R/6 lines, the expanded repeat in the transgene causes a widespread pathology that appears not to be striatum specific (Li et al. 1999; Morton et al. 2000), and gene expression alterations have substantial overlaps in different brain areas assayed (Luthi-Carter et al. 2002a), so changes in gene expression are likely to occur widely throughout the brain. Gene expression differences based on R6/2 whole brain samples (Morton et al. 2005a) correlate well both with those seen in only the striatum (Luthi-Carter et al. 2002a) and with those seen in these R6/1 whole brain homogenates. This consistency corroborates our use of whole brain samples to correlate gene expression with behavioral changes in the present experiment.

One notable feature was that more genes correlated with exploratory behavior and activity than with the rotarod task. This was not simply attributable to greater intrinsic variability or unreliability in the individual animals’ scores on the rotarod test because the proportional standard errors of measurement were within similar ranges to those observed on the other tests, and variance ratios of both main effects and interactions on this test exceeded those calculated on any of the other tests. It is likely therefore that this reflects some neurological differences between the tasks. Whereas the rotarod drives a ‘forced’ motor skill, the exploration and locomotor activity tests have a much higher ‘voluntary’ component, which are correspondingly likely to recruit more diverse brain circuits in the decision and selection of appropriate actions.

Genes correlated with rotarod performance notably include those related to protein degradation and oxidative activity and may reflect compensatory changes to increase protein clearance and energy production, both altered in HD and HD models (Cui et al. 2006; Lin & Beal 2006; Rubinsztein 2006). The association of these genes with rotarod performance may reflect the increased motor control required for successful rotarod balance, on which the transgenic mice show early deficits. The genes correlated with activity and exploration are largely different genes from those correlated with rotarod performance, as well as different from each other, and more likely reflect the different motivational and cognitive processes associated with generalized voluntary activity, and the exploration of familiar environments and novel objects. The GO analysis shows a common alteration in signaling. For activity, these processes relate to the major intracellular pathways that transduce extracellular-ligand-activated signals. In contrast, for the exploration task, similar pathways are overrepresented but several of these are ion channels implying activation of effector pathways in the brain.

One interesting observation is that Bdnf expression changes only correlate with rotarod activity. Bdnf expression is mediated by the wild-type huntingtin protein but this activity is lost from the mutant protein and Bdnf expression reduced (Zuccato et al. 2001, 2003). Bdnf transport is also reduced (Gauthier et al. 2004) and the removal of neurotrophic support contributes to neurodegeneration. In R6/1 mice, Bdnf protein levels are severely decreased in the hippocampus and striatum by 5 months and this can be rescued by enriching the mouse environment (Spires et al. 2004). The mice in our study had some enrichment of their home cages as part of their standard husbandry (cardboard tubes and paper were provided). The rescue of the Bdnf deficit by enrichment in R6/1 animals gave improved performance on the rotarod and this is entirely consistent with our results. Crossing R6/1 animals with Bdnf+/− mice markedly reduced the age at which deficits in rotarod performance appeared (Canals et al. 2004) although no deficits were seen in the heterozygous Bdnf+/− control animals and the performance of the R6/1x Bdnf+/− animals in other tasks was unimpaired. We also find no correlation with the other tasks we used where performance appears to correlate with other factors. Bdnf+/− mice have abnormalities of eating behavior and locomotor activity (Kernie et al. 2000) and a recent report of a subject with haploinsufficiency of the BDNF gene showed a similar pattern with obesity and hyperactivity and some cognitive deficits (Gray et al. 2006). Thus, the phenotype of the HD model mice is unlikely to be accounted for simply by Bdnf deficits.

P57Kip2 has been shown to promote differentiation of dopaminergic neurons, through cofactor binding to the transcription factor Nurr1 (Joseph et al. 2003). This activity is likely to be important for dopamine synthesis and storage. It is possible the link between ceruloplasmin and the behaviors measured in this study could be through its crucial role in binding copper that has previously been shown to affect the dopamine and the GABA systems of the basal ganglia and subsequent control of movement. Higher levels of expression correlated with poorer behavior measures seen in the current study for ceruloplasmin and p57Kip2 may reflect an attempt by the cell to boost their activity to improve motor behavior, which cannot be realized as a result of a deficit in response elsewhere in the molecular cascade. Consistent with the results here, Cp and Cdkn1c were also increased in early-stage postmortem HD caudate (Hodges et al. 2006).

The correlational analysis conducted here establishes associations between changes in gene expression and in behavior. As such, it cannot establish causative relationships, although many of the demonstrated correlations are suggestive. This approach has been successfully applied in drosophila where a correlation analysis of survival after nicotine exposure and altered gene expression showed the gene encoding ornithine aminotransferase to be most highly correlated with nicotine resistance (Passador-Gurgel et al. 2007). Subsequent experiments showed that ornithine and GABA levels modified survival times indicating the functional nature of the observed correlation.

It is also likely that complex compensatory changes are in operation; indeed, our results show some evidence that compensatory changes are occurring. Nevertheless, we have examined a global readout of the transcriptional activity of the brain and have detected a pool of genes whose transcriptional activity may affect that behavior. There are transcription factors and transcriptional modulatory proteins that correlate with individual behaviors but it is difficult to link these directly to the gene expression changes observed. We were surprised how few genes there were in common between the different correlations. This, too, indicates that specific groups of genes may have co-ordinated expression that gives rise to particular responses in animals and suggests a variety of molecular targets for further more detailed analysis. This work strongly supports the use of multiple behavioral readouts in HD models and tasks that examine different aspects of the behavioral phenotype should be included in any behavioral battery that examines the effect of potential therapeutic interventions.


This work was funded by the BBSRC, HighQ foundation and the Hereditary Disease Foundation. We thank Poly Cheng for mouse genotyping. The authors declare that they have no conflicts of interest.