In the present study we have demonstrated the presence of neurometabolic changes in the cerebellum of patients with ET, which were correlated to ETRS. To our knowledge this is the first study to show a positive correlation between the ratio of GABA+ and Glx in the cerebellum of ET patients, and increasing tremor disability suggesting a pathological correlation. Our interpretation is that this can be attributed to an overall decrease of Glu, rather than an increase in GABA+, which was constant. This suggests that an increasing ETRS score is partly due to a disturbance in cerebellar Glu-concentration, and that there is a greater impact of low Glx concentration on tremor severity than a change of GABA+ concentration. Moreover, a greater ratio between the two main neurotransmitters acting in an inhibitory manner (GABA+) and an excitatory manner (Glx) on the synaptic level respectively, seems to correlate with a more severe tremor. This highlights the importance of the balance between the two excitatory and inhibitory neurotransmitters for proper motor control.
Spatial origin of disease
Recent mechanistic research on ET has focused on anomalies localized in the cerebellum, in particular Purkinje cells and the dentate nucleus [33]; these sites are schematically labeled (1) and (2) in Fig. 4. Treatment of therapy resistant ET can include ‘Deep Brain Stimulation’ (DBS) of the ventral intermediate nucleus of the thalamus or the dentatorubrothalamic tract, and therefore both the cerebellum and thalamus were examined in this study. Moreover, the pathology in ET is not static, and the symptoms often progress and spread to other cerebellar areas [3]. In addition, previous reports have shown a correlation to increasing tremor severity with abnormalities; the latter included decreased cerebellar cortical NAA/tCr [34], CF-PC synaptic pathology [21], both suggesting cerebellar neurodegeneration. Also, increased thalamic Glx concentration and Glx/Cr ratio values [35] have both been shown to correlate to tremor severity. Our findings suggest an additional, previously unreported, perspective on the pathological process in the cerebellum.
We did not find a significant difference in GABA concentration (see Fig. 2), which is in accordance with previous studies, in which a reduction of the number of GABA receptors was observed, but not a reduction in the cerebellar GABA concentration [17]. MEGA-PRESS MRS at 3 T in our application did not allow us to spatially separate signals from the dentate nucleus and cortical Purkinje cells, which is illustrated schematically by the red dotted square in Fig. 4. Thus, we have to interpret the results as a combination of the outcome of cellular events within all of these tissues (see Fig. 1).
Aspects of tissue structure
Tremor is the most prevalent movement disorder, but it represents a wide range of etiologies and it may represent a number of progressive diseases [36]. Moreover, clinical-, pathological and pharmacological heterogeneity may explain why evidence regarding etiology is difficult to conclude. A number of morphological aberrations have been identified in ET, mainly affecting Purkinje cells. In contrast, no pathological morphological changes have been found for other regions involved in the motor network (including the thalamus).
Previous observations include a reduced number of Purkinje cells (PCs), dendritic swelling and increased Purkinje cell torpedoes in the cerebellum [14], and an increased number of heterotopic Purkinje cells, localized in the molecular layer of the cerebellar cortex [37]. In contrast, a number of other studies have failed to show evidence of Purkinje cell loss [38, 39]. A difference in response to pharmacological substances further elucidates the heterogeneity of the disease.
Is the disease symmetric, or not?
Similar to our observations, Louis and co-workers [23] did not find a reduction in GABA+ concentration in the dentate nucleus of ET using MEGA-PRESS, compared to the control subjects. Asymmetries of metabolic changes in ET have previously been reported, and Louis and co-workers did find an asymmetry, as the concentration of GABA in the right dentate nucleus was c. 15% larger than the concentration on the left dentate nucleus. This also agrees with our observations as we did not detect a GABA reduction, but an asymmetric distribution of GABA in the cerebellar region of interest. The latter was, in the patient cohort described here, limited to a trend (representing c. + 10% concentration difference of GABA+, R vs. L), and therefore was not a significant effect.
In the context of symmetry, it is important to consider aspects such as objective asymmetry in clinical expression, asymmetry in perceived disability (cf. dominant hand), as well as tissue asymmetry in terms of tissue structure and biochemical parameters. Contralateral low thalamic NAA/Cr ratio was found in ET patients with predominantly tremor of the right arm [40], and clinical asymmetry of tremor was correlated to a significant degree with cerebellar pathology [41]. All of our patients perceived more symptomatic tremor on the right side, thus displaying an asymmetry from a clinical perspective. This may explain why significant correlation of GABA+/Glx versus ETRS was only detected on the ipsilateral cerebellar side (see Supplementary Figure 2), although one potential limitation of this observation of the side-to-side difference may be the limited sample size.
Inhibitory input
The Purkinje cells convey their input to the deep cerebellar nuclei through GABA, the main inhibitory neurotransmitter, which binds to synaptic GABA receptors. Flumazenil, which specifically binds to the GABAA receptor, has been shown in PET studies to map the binding in the cerebellum (dentate nucleus), but also in the thalamus and the premotor cortex [18]. Other PET studies have also shown that the severity of tremor was correlated to abnormal GABA receptor binding [19]. In addition, a previous study showed a reduction of GABA receptors but not in GABA concentration in the cerebellum [17]. Available medication is very non-specific, but certain vectors that increase GABAergic transmission can be effective in treating ET, although with very variable outcomes [42].
Purkinje cells receive two direct excitatory inputs (Glu), from climbing fibers and from parallel fibers [43]. (See also Fig. 4) Notably, the synaptic distributions of the climbing fibers appear to be important to control Purkinje cell physiology and therefore normal cerebellar function. A particularly interesting finding, which affects the climbing fiber to Purkinje cell synaptic connections, was the observation that synaptic density in the molecular layer of the cerebellar cortex was associated with tremor severity [15]. A lower synaptic climbing fiber-Purkinje cell density would subsequently result in a lower degree of inhibition of the dentate nucleus. Furthermore, it has been shown that the excitatory amino acid transporter 2 (EAAT2) is reduced in the cerebellar cortex and increased in the dentate nucleus in patients with ET, indicating that the glutamatergic synaptic transmission is altered [44]. The combined post-mortem observations of these studies showed that the excitatory input (i.e., Glu-input) was disturbed both in the Purkinje cells and at the dentate nucleus in ET patients. Our interpretation is that the observed lower glutamate concentrations (expressed in terms of reduced Glx concentrations) further advocate the role of altered glutaminergic transmission in ET.
What about the thalamus?
Neurogenic excitatory output from the cerebellum is propagated to the thalamus, which is suggested to have a role in the pathogenesis of ET. Interestingly, the effects of a focal stroke can ameliorate the symptoms, as has been reported by Duncan [45]. Metabolic abnormalities with low NAA/Cr ratio of the contralateral thalamus in ET patients with predominantly tremor of the right arm have been reported [40]. In a recent study, the Glu-concentration was increased in the thalamus of ET patients and correlated to tremor severity, suggesting that thalamic glutamatergic transmission is involved in the pathophysiology [35]. No significant differences in thalamic GABA+, Glx or GABA+/Glx between the ET and HC groups were observed in our study. Furthermore no correlation with GABA+/Glx and tremor severity was seen in the thalamus. The voxel placed in our series is larger and extends beyond the thalamic anatomical borders, thus involving the brain areas in the vicinity of the thalamus (see Fig. 2). Our thalamic results are therefore not easily comparable with previous reports and this may be the reason why our results differ.
Limitations
Certain experimental aspects of this work were restricted by safety, hardware, partial volume and/or techniques. The patient cohort examined here were all diagnosed by a movement disorder specialist, and then referred to the neurosurgical department, thus they constituted a limited, but very narrow and well-defined group (bilateral, but right-handed tremor). Most of them had severely debilitating hand tremor and they were therefore all selected as being eligible for surgery. Due to the severely disabling disease, surgery was performed immediately after these measurements, with an excellent therapeutic outcome; all patients were also significantly better after surgery (data not shown), indicating the uniquely well-defined cohort.
Unfortunately, the measurements could not be repeated after the surgery, both for ethical and for safety reasons, as the acceptable exposure to electromagnetic fields of the DBS-implants was not compatible with the pre-surgical examination conditions (i.e., 3 T and the SAR level required), and this was a limitation of the present work. At a later time (likely within a few years?) when a sufficiently acceptable set of MR safety conditions have been developed, this may nevertheless be possible to perform.
When performing GABA detection, there are several technical aspects to be considered, one being the complex nature of the spectral scalar coupling pattern of the GABA resonances, and also due to the low signal intensity compared to the other metabolite signals in the spectra. Additionally, another major issue is spectral overlapping by the more intense creatine signal at 3.00 ppm, which obscures the GABA signal at 3.01 ppm severely. In addition, spectral editing is needed to reveal the GABA+ signal, and single voxel MEGA-editing is currently the standard technique. Moreover, as we are quantifying difference spectra (using water scaling in LCModel), a high signal-to-noise ratio (SNR) is necessary; thus, we need a sufficiently large number of signal averages and also a large spectral voxel to achieve a high enough SNR for reliable concentration quantification of the resulting spectra. In comparison with conventional MRS, MEGA-PRESS requires a substantially larger voxel volume, which results in tissue heterogeneity between Purkinje cells and the dentate nucleus etc.
Compared to Louis et al. [23] we used a larger voxel in our cerebellar measurements (35 × 25 × 25 mm3, which is equivalent to 21.9 mL compared to 25 × 25 × 25 mm3, 15.6 mL). Thus, in our measurements of cerebellar tissue we analyzed part of the cerebellar cortex, in combination with the deep cerebellar nuclei and white matter tracts in the cerebellum. Furthermore, to compensate for the large SNR reduction when using a smaller voxel size, Louis et al. collected more signal averages (196 ON/OFF compared to our 160 ON/OFF). In addition, there might also be other systematic differences between our results since we used different MR systems (Philips compared to Siemens), and different LCModel basis sets were used in the quantification. Our experience is that there are small, but noticeable, systematic differences in spectral analysis when different basis sets are used. Furthermore, Louis et al. used segmentation to determine the tissue composition of their voxel, a procedure which we did not apply. However, our interpretation is that most of these effects are removed when concentration ratios are used, and thus we found a significant correlation between cerebellar GABA+/Glx and ETRS. Although these differences in acquisition and analytical approach exist, the results should in our view still be comparable.
Our present data does not let us to come to a conclusion on whether ET is characterized by such morphological abnormalities in specific cerebellar regions, or whether the biochemical alterations are limited to certain parts, or widespread, as the procedures used here were limited with respect to the spatial resolution of the MRS-technique utilized. However, with increasing advances in MRS technology the voxel volume can be reduced, allowing different regions to be separately analyzed. One such procedure is 3D spiral-MEGA-PRESS GABA Chemical Shift Imaging devised by Bogner et al. [46], for whole brain detection of GABA+ at a spatial resolution of about 1 cm (compared to 3 cm which was used here); the technique requires a uniquely stable spiral acquisition. Another procedure is the Glu-CEST procedure championed by Reddy et al. for detecting Glu-weighted images of the excitatory neurotransmitter with the spatial resolution of conventional imaging [47]. However, both techniques are extremely demanding on subjects, hardware (Glu-CEST requires 7 T) as well as operator skills, and at this time they are only available to selected laboratories.
Since we base our argument on the ratio between GABA+ and Glx, it is of interest to consider what we have measured. The GABA+ measurements include a proportion of macromolecules, and probably also homocarnosine, which is often considered to be a constant fraction (about 40–50%); thus, any short-term changes in GABA+ mainly reflect changes in GABA. Glx is also a composite but of Glu and Gln, although a majority of the signal typically represents Glu [48]. The largest Glu to Gln ratios at the cellular level have been observed in excitatory terminals of parallel fibers in the molecular layer, and also in mossy and climbing fibers, corresponding to a concentration ratio of about 80% Glu (of Glx); and the lowest ratios have been observed in glial cells [49]. Another parameter to consider is that most Gln in the brain is less metabolically active than Glu. Please also note that MRS is typically acquired at a much lower spatial resolution, and the partial volume effects of a wide range of different cell populations then make it harder to predict a neurotransmitter ratio accurately based on cellular concentrations. However, our interpretation is that the majority of the Glx-resonances in the cerebellum are indeed Glu, and this agrees with a report by Goryawala which determined a concentration ratio of between 2 and 4 in gray matter, corresponding to c. 70% Glu (of Glx) in the cerebellum and close 80% in the brain [50].
Determination of absolute concentrations is typically recommended as this will facilitate comparison between different tissues and subjects [22]. Usually, this is achieved by the use of LCModel analysis in combination with water scaling, which is much preferred to creatine-ratios. In the cerebellum, in consideration of the relative tissue heterogeneity, water scaling appears to be less effective due to the spatial (including intra vs. extra-cellular) tissue variation of water content. This is the likely cause of the relatively large variation of the GABA+ and Glx concentrations in Fig. 3a and b. Instead, the uniquely functionally coupled ratio between the related neurotransmitters GABA and Glu will remove the effects of such spatial variation (see Fig. 3c). It is also worth noting that another difference between our study and some others, was the limited number of included ET patients and healthy control subjects (here 10 ET patients and six HCs). However, our patients belonged to a highly selective cohort of ET patients intended for DBS therapy, which may show stronger disease-related parameters.