1 Department of Biochemistry, University of Cambridge, Cambridge CB2 1QW
2 Cell and Molecular Biology Section, Biomedical Sciences Division, Imperial College London, London SW7 2AZ
3 Department of Medical and Community Genetics, Imperial College London, Northwick Park Hospital, Harrow HA1 3UJ, United Kingdom
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Machado-Joseph disease; magic angle spinning; metabolomics; multivariate pattern recognition
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The protein produced by this gene, ataxin-3, normally shows cytosolic expression and is expressed throughout the body. However, in affected individuals, ataxin-3 forms ubiquitinated neuronal intranuclear inclusions, particularly in neurones of the main affected brain regions. The function of ataxin-3 is still unknown, and in an attempt to define the role of this protein, a number of animal models of the disease have been produced, with mixed results. For example, Ikeda and colleagues (18) failed to find any neurological disorder for mice expressing the full-length expanded protein, but ataxia was produced in mice expressing a truncated form of the protein with an expanded polyglutamine region.
In the postsequence genomic era, the generation of knockout and transgenic mouse models of disease is set to expand greatly. However, many such models do not demonstrate the human phenotype, displaying a milder pathology which is difficult to investigate. High-resolution 1H-NMR spectroscopy coupled with pattern recognition provides a rapid method for providing a metabolic phenotype (26), or "metabotype" (13), of a tissue or organism (14, 15). In this study a mouse model of SCA3 has been examined which was produced using a human yeast artificial chromosome (YAC) containing an MJD1 gene with a CAG triplet expansion of 84 repeats in length (6). These mice have previously been shown to possess a mild progressive cerebellar deficit, together with neuronal intranuclear inclusion formation in the pontine and dentate nuclei. However, metabolic abnormalities have not been previously reported. Using NMR-based metabolomics/metabonomics, we have identified a metabolic fingerprint associated with triplet repeat expansions in the MJD1 gene for both cerebellum and also the cerebrum, a region of the brain not considered to be principally targeted.
![]() |
METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
NMR data acquisition.
Solution-state and magic angle spinning (MAS) 1H-NMR spectroscopy was performed at 600.2 MHz using a Bruker Avance spectrometer interfaced with a 14.1-Tesla superconducting magnet and high-resolution inverse geometry 1H probes. For solution-state studies, extracts were prepared from cerebrum and cerebellum tissue (100 mg wet weight; n = 5) using a perchloric acid extraction procedure. Following neutralization of the extract with KOH, samples were lyophilized and reconstituted in 2H2O containing 4 mM sodium trimethylsilyl-2H4-propionic acid (TSP). Solution-state NMR spectroscopy was performed using a triple-axis high-resolution inverse geometry 1H-NMR probe. These spectra were then used to quantify the aqueous soluble metabolites in a given tissue.
Intact cerebral tissue (10 mg) was also examined using a high-resolution MAS probe operating at 4°C, interfaced with the magnet described above. Samples were taken from the outer region of the cerebellum and the frontal cortex (n = 7 for SCA3 mice, n = 8 for control strain). Following storage, these samples were placed within a zirconium oxide rotor and spun at 5-kHz spinning speed, to remove the effects of spinning side bands from the spectra acquired. During sample preparation tissue was chilled at 01°C throughout by carrying out sample preparation over a bed of ice. Shimming and NMR preparation time was kept to a minimum, but throughout this and the subsequent NMR analysis the sample was chilled to 4°C by a constant stream of cooled nitrogen gas. Under such conditions no noticeable degradation was observed during acquisition as previously reported for cultured neuronal cells (16), with the resonances from lactate, N-acetylaspartate (NAA), and acetate remaining constant throughout the acquisition times as determined by brief pulse and acquire experiments either side of the main acquisition time.
Solvent-suppressed spectra were acquired into 32 k data points, averaged over 128 (solution state) or 256 acquisitions (MAS NMR spectroscopy; total acquisition time 11 min) using either a solvent suppression sequence based on the start of the nuclear Overhauser effect spectroscopy (NOESY) pulse sequence or a Carr Purcell Meiboom and Gill (CPMG) pulse sequence with continuous wave solvent suppression (MAS NMR spectroscopy only). The NOESY preset pulse sequence employed a mixing time of 150 ms, during which time the effects of B0 and B1 field inhomogeneities were suppressed. The CPMG pulse sequence contained a 40-ms T2 total delay with 500-µs delays between each
pulse. The high-resolution MAS (HRMAS) spectra were acquired within 30 min, and no discernable change in NAA was detected during this period. In prior studies a temperature of 27°C (as opposed to 4°C used in this study) only produced a 5% decrease in the NAA resonance intensity (15).
The resultant spectra were processed using XWINNMR 3.1 software (Bruker). Following Fourier transformation, spectra were integrated, using AMIX (Bruker), across 0.04 ppm spectral regions between 0.4 and either 4.2 or 9.4 ppm for MAS and solution-state NMR, respectively. The output vector representing each spectrum was normalized across the integral regions, excluding the water resonance (1). All ratios are represented as relative to the total detected metabolic resonance intensity unless specified.
Pattern recognition of NMR spectra.
The data reduced format of the high-resolution 1H-NMR spectra data sets were imported into the SIMCA package (Umetrics, Umea, Sweden) and preprocessed with Pareto scaling (10). Each spectral region was scaled to (1/sk)1/2, where sk is the standard deviation for variable k, increasing the contribution of lower concentration metabolites to the models generated compared with models where no scaling is used. Initially each data set was examined using principal components analysis (PCA) to examine trends and clusterings in an unsupervised manner. In this pattern recognition technique the algorithm calculates the most amount of correlated variation in a data set and scores each spectrum according to this variation along principal component 1 (PC1). This procedure is repeated for other components until the majority of the variation in the data set is described. PCA is particularly useful for describing "rectangular data sets" where there are more variables (in this case spectral metabolite regions) than observations (the sample examined).
However, where this was inadequate to define a clustering, partial least squares for discriminate analysis (PLS-DA) was applied to force classification. In this pattern recognition technique, a regression model is formed between the NMR variables and class membership, allowing the selective removal of variables that do not contribute to class distinction.
To determine which metabolic deficits were common to both regions of brain investigated, the spectral filtering routine, orthogonal signal correction (OSC), was applied to the solution-state spectra. In this filter, a regression model is formed between the spectral data and the disease state of the tissue. Variation orthogonal to the vector representing disease state is subtracted to produce a data set that is more influenced by the disease state. OSC was applied by subtracting two components from the data set. The resultant data was examined using PCA.
For each model built, the loading vector for the principal component or PLS-DA component was examined to identify which metabolites contributed to these clusterings. Metabolites are reported if they contributed more than 50% of the maximum contribution of a resonance to the model. However, these reported metabolites only represent the most highly perturbed changes and not the complete profile identified using the pattern recognition process.
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
|
Determining metabolic differences common to both brain regions.
As the perturbations in concentration of glutamine, phosphocholine, and myo-inositol caused by the transgene were common to both regions of the brain, as determined by solution-state spectroscopy, the combined solution-state data set containing spectra from both the cerebellum and cerebrum was examined using a supervised pattern recognition approach. PLS-DA correlates the variation in the data with a vector representing class membership (i.e., a supervised pattern recognition technique which uses information as to whether tissue was from control or SCA3 mice in this case) PLS-DA successfully classified the tissue according to the presence or absence of the transgene with this separation being caused by an increase in concentration of glutamine and a decrease in concentration of lactate, choline, phosphocholine, GABA, NAA, and taurine, despite the tissue being taken from different regions of the brain (Fig. 3, A and B). To further improve this separation, an OSC filter was used to remove variation associated with distinguishing the two brain regions in the solution-state data set. This produced a data set where the remaining variation was caused by the effects of disease state common for both the cerebellum and cerebrum. This variation was investigated by PCA. The combined OSC and PCA analysis separated the tissue according to the presence or absence of the transgene as a result of increased concentrations of glutamine and creatine and decreased concentrations of lactate, choline, phosphocholine and GABA (Fig. 3, C and D). To confirm these differences detected through multivariate analysis, the ratios of glutamine to lactate, glutamine to phosphocholine+choline, and glutamine to GABA were calculated by manual integration of the relevant resonances for all the solution-state spectra. These ratios were found to be significantly altered in SCA3 mice compared with the control strain (combined glutamine/lactate ratio for control mice = 0.32 ± 0.03, SCA3 mice = 0.39 ± 0.05, P = 0.0009; glutamine/phosphocholine+choline for control mice = 0.61 ± 0.02, SCA3 mice = 0.74 ± 0.09, P = 0.003; and glutamine/GABA for control mice = 0.65 ± 0.10, SCA3 mice = 0.76 ± 0.04, P = 0.03 according to Students t-test). The changes detected in aqueous extracts are summarized in Table 1.
|
|
|
|
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The increase in glutamine and decrease in myo-inositol concentrations were also detectable in intact cerebellum tissue using high-resolution MAS 1H-NMR spectroscopy. However, despite the marked perturbations detected in extracts of cerebrum tissue, no differences were detected in the intact cerebrum tissue (frontal cortex) from SCA3 mice. The failure to find differences in intact tissue in this region most likely results from either lipid resonances obscuring key metabolic regions or a lack of reproducibility in terms of the region dissected for use in the MAS spectra. Furthermore, the MAS spectra had reduced resolution compared with those obtained from extracts, in part reflecting the residual line broadening present from susceptibility effects across the sample, and this may have further obscured differences.
A central feature of polyglutamine expansion diseases is the accumulation of the protein to form nuclear inclusions within neurons. These proteins are often ubiquitinated (7, 12, 28, 30), recruiting the proteasome to the nuclear inclusions. Intriguingly, wild-type ataxin-3 is also recruited to Marinesco bodies, ubiquitin-positive nuclear inclusions that are associated with aging, even in the absence of pathologically expanded polyglutamine. This suggests that ataxin-3 may, in part, be responsible for certain age-related neurodegeneration (12). However, there is still no definitive evidence that these nuclear inclusions are important in pathogenesis, and these may indeed even have a protective role.
The increase in glutamine detected in both the cerebrum and the cerebellum may arise from either the proteasome degradation of polyglutamine expanded ataxin-3 or, following neuronal cell death, an increase in the relative contribution of metabolites high in concentration in glia, where the majority of brain glutamine resides (2). This increase in cellular concentration of glutamine appears to be common to a number of other diseases caused by an expansion of CAG trinucleotide repeats. For example, Taylor-Robinson and colleagues (33) reported an increased total glutamine and glutamate concentration relative to creatine in sufferers of Huntington disease using MRS in vivo, while Jenkins and coworkers (19) measured a 100% increase in glutamine in a mouse model of the disorder. However, in human sufferers of Kennedy disease, another disease caused by a CAG triplet repeat expansion, no such increase in the total pool of glutamate and glutamine was detected (22). Hannan (17) has previously proposed that subtle alterations in glutamine, and consequently glutamate levels, in CAG triplet repeat diseases may induce chronic excitotoxicity and induce slow cell death in neuronal populations possessing specific glutamate receptors. However, the concentration of glutamate in the cerebellum of SCA3 transgenic mice was decreased compared with control animals. This suggests that glutamate-mediated excitotoxicity probably does not have a direct part to play in the progression of SCA3-like pathology in this mouse model and did not contribute to the metabolic differences detected in this study. MRS in vivo is unlikely to detect subtle changes in the proportion of glutamate and glutamine, as this technique cannot currently readily resolve the separate glutamate and glutamine resonances.
The consistent detected decrease in concentration of myo-inositol, a glial marker (4), in both brain regions indicates that the metabolic changes, and in particular the increase in concentration of glutamine, are unlikely to result from large-scale gliosis and an increased proportion of glial metabolites in the brain regions. This also suggests the detected decreases in NAA, GABA, and glutamate, three metabolites found in high concentrations in neuronal cells (2, 3), in the cerebrum is probably caused by neuronal cell metabolic dysfunction, as a result of neuronal intranuclear inclusions, prior to large amounts of neuronal cell death and subsequent gliosis. Although histology has shown gliosis, astrocytosis, and loss of Purkinje cells in the cerebellum of these mice (6), these processes primarily affect the cerebellum and would not account for the changes detected in the cerebrum, suggesting the metabolic markers detected in this study relate to initial dysfunction rather than the resulting cell loss. Furthermore, choline and phosphocholine, precursors of the cell membrane component phosphatidylcholine, are decreased in tissue from SCA3 mice, also suggesting the metabolic differences are not a result of cell death, which results in the repartitioning of membrane constituents to the extracellular space, increasing their NMR detectability. Intriguingly, Boesch and colleagues (5) have also described differences in the relaxation behavior and detectability of the combined choline resonance during MRS of SCA types 2 and 6 patients.
The simultaneous analysis of both regions for common metabolic markers of SCA3 did not identify myo-inositol as being changed despite it being detected in the separate models of the cerebellum and cerebrum. This metabolite was found to be increased in the cerebellum compared with the cerebrum when PCA was applied to the total solution-state data set. This indicates that the use of OSC was not able to remove the variation associated with the different brain regions for this metabolite.
Another mechanism by which neuronal intranuclear inclusions produce metabolic abnormalities is the aggregation of other proteins to them (25). Polyglutamine domains provide good targets for transglutaminase, producing cross-links with polypeptides containing lysyl residues. The glycolytic enzyme GAPDH has been shown to tightly bind to proteins containing polyglutamine repeats in vitro, and the depletion of key enzymes from the cytosol may cause some of the metabolic changes detected, as well as providing another mechanism of toxicity.
A number of other metabolic deficits have been reported in sufferers of SCA3, including impaired dopamine transport (34), hypometabolism in the occipital cortex, cerebellar hemispheres, vermis, and brain stem (29) and a raised lactate/pyruvate concentration in the cerebrospinal fluid of sufferers (24). While this last report appears to contradict the decrease in lactate concentration detected in cerebellum and cerebrum tissue from SCA3 transgenic mice in the combined data set for both brain regions, lactate measured in tissue extracts will also represent a significant component associated with the breakdown of glucose and glycogen postmortem during dissection of the brain, prior to storage. Thus the decreased lactate detected in SCA3 mice cerebellar and cerebral tissue may arise from reduced glycolytic capacity in these mice. An increased lactate concentration has also been detected in sufferers of SCA type 2 but not type 6, suggesting that differences in glycolytic capacity may be dependent on the subtype of SCA (5).
This NMR- and pattern recognition-based approach could be further extended to investigate the changing phenotype induced by increasing the polyglutamine repeat length. In keeping with other polyglutamine ataxias, the phenotype associated with SCA3 increases with increase in polyglutamine length. Lodi and colleagues (21) have reported a strong negative correlation with GAA trinucleotide repeat length and mitochondrial ATP production in patients with Friedreich ataxia, confirming that mitochondrial abnormalities are at the center of this disorder. Indeed, a number of neurological disorders including SCA2, SCA6, Kennedy disease, and Huntington disease, all caused by expansions of CAG triplet repeats, demonstrate a strong negative correlation between CAG triplet repeat length and NAA concentration (5, 8, 19, 22, 23). This suggests a similar approach would be profitable for understanding the pathology associated with SCA3. By modeling the changing phenotype with polyglutamine expansion, it may be possible to understand better the toxic events associated with neuronal intranuclear inclusion formation in the SCAs.
![]() |
ACKNOWLEDGMENTS |
---|
J. L. Griffin acknowledges the support of the Royal Society, UK. This work was also supported by Ataxia, UK.
![]() |
FOOTNOTES |
---|
Address for reprint requests and other correspondence: J. L. Griffin, Dept. of Biochemistry, Univ. of Cambridge, Tennis Court Rd., Cambridge CB2 1QW, UK (E-mail: jlg40{at}mole.bio.cam.ac.uk).
10.1152/physiolgenomics.00149.2003.
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|