OUP user menu

Decreased carbon-11-flumazenil binding in early Alzheimer’s disease

Belen Pascual, Elena Prieto, Javier Arbizu, Josep M. Marti-Climent, Ivan Peñuelas, Gemma Quincoces, Rosina Zarauza, Sabina Pappatà, Joseph C. Masdeu
DOI: http://dx.doi.org/10.1093/brain/aws210 2817-2825 First published online: 8 September 2012


Neuronal loss in Alzheimer’s disease, a better correlate of cognitive impairment than amyloid deposition, is currently gauged by the degree of regional atrophy. However, functional markers, such as GABAA receptor density, a marker of neuronal integrity, could be more sensitive. In post-mortem hippocampus, GABAA messenger RNA expression is reduced even in mild cognitive impairment. We measured whole-brain GABAA binding potential in vivo using [11C]-flumazenil positron emission tomography and compared GABAA binding with metabolic and volumetric measurements. For this purpose, we studied 12 subjects, six patients with early Alzheimer’s disease and six healthy controls, with [11C]-flumazenil and [18F]-fluorodeoxyglucose positron emission tomography, as well as with high-resolution magnetic resonance imaging. Data were evaluated with both voxel-based parametric methods and volume of interest methods. We found that in early Alzheimer’s disease, with voxel-based analysis, [11C]-flumazenil binding was decreased in infero-medial temporal cortex, retrosplenial cortex and posterior perisylvian regions. Inter-group differences reached corrected significance when using an arterial input function. Metabolism measured with positron emission tomography and volumetric measurements obtained with magnetic resonance imaging showed changes in regions affected in early Alzheimer’s disease, but, unlike with [11C]-flumazenil binding and probably due to sample size, the voxel-based findings failed to reach corrected significance in any region of the brain. With volume of interest analysis, hippocampi and posterior cingulate gyrus showed decreased [11C]-flumazenil binding. In addition, [11C]-flumazenil hippocampal binding correlated with memory performance. Remarkably, [11C]-flumazenil binding was decreased precisely in the regions showing the greatest degree of neuronal loss in post-mortem studies of early Alzheimer’s disease. From these data, we conclude that [11C]-flumazenil binding could be a useful marker of neuronal loss in early Alzheimer’s disease.

  • Alzheimer’s disease
  • flumazenil PET
  • MRI
  • GABAA receptors


In brains of patients with Alzheimer’s disease with the mildest clinically detectable dementia (Clinical Dementia Rating scale = 0.5), a 32% loss of neurons has already occurred in the entorhinal cortex (Gomez-Isla et al., 1996). Neuronal loss spreads as the disease progresses and correlates better than amyloid deposition with the degree of clinical worsening (Crystal et al., 1988; Gomez-Isla et al., 1996; Giannakopoulos et al., 2003). Regional neuronal loss can be gauged by the degree of atrophy, currently used as a neuroimaging marker of neuronal loss (Killiany et al., 2000; Chetelat et al., 2010). It is possible, however, that imaging of receptors ubiquitous in neurons, such as the GABAA receptor (Kisvarday et al., 1990), could be more sensitive than atrophy to neuronal loss in Alzheimer’s disease. In post-mortem hippocampus, a decreased expression of the GABAA messenger RNA has been described even in mild cognitive impairment (Rissman et al., 2004). Regional GABAA receptor availability can be quantified in vivo with [11C]-flumazenil PET, which has been reported to be more specific than diffusion-weighted MRI for the detection of neuronal loss in stroke (Heiss et al., 2004). In contrast, a single patient study and two previous series using [11C]-flumazenil in Alzheimer’s disease have been negative (Sedvall et al., 1987; Meyer et al., 1995; Ohyama et al., 1999). The study by Meyer et al. (1995) included five patients with mild or early Alzheimer’s disease (mean Mini-Mental State Examination = 22) and six healthy controls. Ohyama et al. (1999) studied five patients with probable Alzheimer’s disease and five healthy controls. Partly because lateral temporoparietal association cortex shows the greatest metabolic loss in Alzheimer’s disease, these PET studies focused on a comparison of metabolism or regional cerebral blood flow with [11C]-flumazenil binding in these regions. However, these are not the regions where the greatest neuronal loss has been detected in post-mortem studies (Brun and Englund, 1981). Additionally, the relationship between atrophy and metabolism is complex, with areas, such as the hippocampus, where there is atrophy but relatively preserved metabolism and others with marked metabolic decline but little atrophy (Chetelat et al., 2008). This is not surprising because metabolism reflects synaptic activity (Rocher et al., 2003); metabolic loss in a region may simply reflect neuronal loss in the region home to the neurons projecting to the metabolically impaired region. Lateral parietotemporal association cortex receives a heavy projection from the medial temporal region (Lavenex et al., 2002; Munoz and Insausti, 2005), with marked neuronal loss in Alzheimer’s disease (Gomez-Isla et al., 1996, 1997). None of the previous [11C]-flumazenil studies in Alzheimer’s disease were performed using voxel-wise comparisons, partial volume correction or normalization starting with PET co-registration to native MRI. We postulated that more recent imaging methods could facilitate detecting GABAA-binding abnormalities in early Alzheimer’s disease. We chose a small sample size to determine whether [11C]-flumazenil PET could separate patients with early Alzheimer’s disease from healthy controls when the power of the voxel-based analysis was too low to separate these two groups based on two Alzheimer’s disease neuroimaging markers, namely grey matter atrophy on MRI or reduced metabolism on [18F]-fluorodeoxyglucose PET (Chen et al., 2011).

Materials and methods


Twelve participants, divided into two groups (Alzheimer’s disease and healthy controls) of six participants each, were studied with neuropsychological testing, structural MRI and PET with [18F]-fluorodeoxyglucose PET and [11C]-flumazenil, a GABAA receptor antagonist. Participants in the two groups had similar age, sex and educational level (Table 1). Patients with Alzheimer’s disease met National Institute of Neurological and Communication Disorders and Stroke - Alzheimer Disease and Related Disorders Association (NINCDS-ADRDA) criteria (McKhann et al., 1984) for the diagnosis of probable Alzheimer’s disease but with a mild degree of cognitive impairment (mean Mini-Mental State Examination score was 21.2). Other neurological or psychiatric disorders were excluded in both patients with Alzheimer’s disease and controls by a thorough medical and neurological evaluation, including standard biochemical and haematological studies. Because both groups had served as controls for another study (Pascual et al., 2010), comparing them to patients with microvascular disease, the presence of vascular lesions, including white matter changes visible on T1-weighted images or more than punctate hyperintensities on T2-weighted images, was carefully excluded by MRI. To facilitate sleep, one patient and one healthy control were taking a similar dose of lormetazepam, which in both cases was discontinued 1 week before the PET procedure. Two patients were taking donepezil, also discontinued 1 week before undergoing PET. The study had been approved by the local Ethics Committee following Declaration of Helsinki guidelines for the study of human subjects. All subjects were fully instructed on the experimental procedures and provided written, informed consent.

View this table:
Table 1

Demographics and results of dementia screening tests

VariableAlzheimer’s diseaseHealthy controls
Number of subjects66
Gender (female)34
Age (years)78.0 (6.1)78.2 (3.2)
Years of education10.0 (3.4)10.0 (3.4)
Dementia screening
Clinical dementia rating
    Memory (range: 0–3)1.1 (0.5)*0
    Sum of boxes (range: 0–18)4.2 (3.6)*0
Mini-Mental State Examination (range: 0–30)21.2 (5.9)*28.8 (0.7)
DemTect (range 0–19) (Kalbe et al., 2004)8.0 (2.3)*18.2 (1.2)
  • Data are presented as means, with standard deviations in parentheses.

  • *Worse performance than healthy controls, P < 0.05.

Neuropsychological testing

The tests performed are listed in Table 2. They were obtained according to standard procedures and were analysed using SPSS Version 17 (2006). Due to the small sample size, comparisons between patients with Alzheimer’s disease and healthy controls were performed using the non-parametric Mann–Whitney U test. Alpha was set at 0.05 for all analyses. Spearman simple correlations were conducted across the entire sample to determine the relationship between neuropsychological test performance and hippocampal [11C]-flumazenil binding.

View this table:
Table 2

Neuropsychological testing performance and its correlation with [11C]-flumazenil hippocampal binding

TestPerformance Alzheimer disease patientsPerformance Healthy controlsCorrelation (r) with [11C]-flumazenil hippocampal binding
Cognitive domains
    First five trials22.6 (7.2)*40.3 (9.1)0.834*0.569
    Immediate recall2.4 (1.7)8.7 (5.1)0.731*0.460
    Delayed recall1.2 (1.8)*8.3 (5.1)0.763**0.558
    Recognition8.8 (5.8)13.3 (1.5)0.4970.534
FCSRT44.0 (25.0)*83.3 (4.5)0.620*0.524
Logical memory
    Immediate recall10.0 (6.0)*28.5 (9.0)0.825*0.620*
    Delayed recall2.4 (2.8)15.3 (9.9)0.752**0.679*
    Recognition10.4 (6.5)19.0 (7.0)0.764**0.682*
    Rey complex figure, recall4.2 (4.0)*18.4 (5.7)0.682*0.491
Boston naming test39.2 (9.6)50.2 (5.6)0.6390.694*
Verbal fluency (named in 1’)
    Animals11.0 (3.6)*18.0 (4.0)0.627*0.471
    Words starting with letter ‘P’12.8 (6.3)16.8 (6.2)0.2230.355
Token test17.2 (3.6)19.5 (0.5)0.616*0.665*
Visuospatial function
Rey complex figure, copy22.7 (13.0)*33.7 (2.1)0.673*0.426
Visual span backwards4.0 (1.9)6.8 (2.6)0.4720.306
Executive function, attention and speed of processing
Trail making test
    Part A107.4 (108.6)70.7 (26.2)−0.210−0.260
    Part B265.8 (76.5)147.0 (89.5)−0.582−0.458
Stroop test: WC23.8 (15.8)29.3 (6.0)−0.0780.164
Mental control WAIS-R17.0 (7.6)24.5 (5.2)0.607*0.457
Digit span forward, backwards10.0 (3.4)14.7 (3.4)0.4250.315
Fluctuation inventory scale0000
Other domains
IADL19.2 (6.8)*8.3 (0.8)−0.849**−0.658*
Geriatric depression scale12.4 (4.1)5.2 (4.4)−0.687*−0.407
Neuropsychiatric inventory3.6 (2.9)1.7 (1.9)−0.479−0.074
  • Performance data are presented as means, with standard deviations in parentheses.

  • FCSRT = free and cued selective reminding test, pictures version; IADL = instrumental activities of daily living scale; RAVLT = rey auditory verbal learning test; WAIS-R = Wechsler adult intelligence scale-revised; WC = word-color.

  • *P < 0.05; **P < 0.01.

Magnetic resonance imaging data acquisition and processing

MRI was obtained for all participants on the same 3 T Siemens Trio scanner (Siemens Medical Solutions), including T1-weighted structural images acquired using a 3D MP-RAGE sequence, with imaging parameters: resolution = 1 × 1 × 1.1 mm3, field of view = 240 × 256 mm2, 160 sagittal slices, repetition time/echo time/nversion time = 2250/900/2.96 ms, flip angle = 9°, bandwidth = 230 Hz/pixel. Whole-brain cortical thickness and hippocampal volume were determined using FreeSurfer software (Athinoula Martinos Centre for Biomedical Imaging). Voxel-based morphometry was performed with SPM8 software (Wellcome Department of Imaging Neuroscience, University College of London), using a fast diffeomorphic registration algorithm (DARTEL; Ashburner 2007). Details of the MRI methods are provided in the online Supplementary material.

[18F]-Fluorodeoxyglucose positron emission tomography data acquisition and processing

Brain metabolism was measured with [18F]-fluorodeoxyglucose PET using standard acquisition and processing methods (Supplementary material).

[11C]-Flumazenil positron emission tomography data acquisition and processing

GABAA binding was measured with [11C]-flumazenil PET using an ECAT EXACT HR+ (Siemens/CTI) scanner with a nominal resolution of 4.5 mm but a measured resolution of 7.3 mm (Prieto et al., 2010). Subject motion was minimized with individually fitted thermoplastic masks. Following a transmission scan obtained using three 68Ge sources and after the bolus injection of 659.71 (±136.16) MBq (17.83 ± 3.68 mCi) of [11C]-flumazenil, a set of 19 sequential 3D PET frames of the entire brain were obtained over a period of 60 min, according to the following protocol: 4 × 30 s, 3 × 60 s, 2 × 150 s and 10 × 300 s. Arterial blood samples were drawn from the radial artery during the dynamic emission scans, for a total of 21 blood samples (intervals: 8 × 15 s, 3 × 60, 2 × 150 and 8 × 300 s). The sampled blood was processed to obtain arterial plasma input function for modelling. Using a procedure that provides reliable findings for [11C]-flumazenil PET (Okazawa et al., 2004), metabolites were not measured but were calculated as described by Okazawa et al. (2004), and plasma input function corrected accordingly. The radioactivity measured in the PET scans and blood samples was corrected for decay to the starting point of each scan.

Voxel-based parametric images of [11C]-flumazenil volume of distribution (VT) were computed with PMOD software (PMOD Technologies) using the graphic plot method of Logan. The slope of the graphical plot was calculated from 15 to 60 min, representing the volume of distribution (Innis et al., 2007), which is here simply called ‘binding’. For the voxel-based analysis, PET studies were spatially normalized with SPM8 after individual PET and structural MRI studies had been co-registered in native space. Using Statistical Parametric Mapping 8 (SPM8), we analysed [11C]-flumazenil binding, both for the whole brain and for a volume of interest including the entire temporal lobes of both hemispheres as the medial temporal regions have previously been identified as bearing marked neuronal loss in early Alzheimer’s disease (Gomez-Isla et al., 1996).

For volume of interest analysis, [11C]-flumazenil parametric images of each subject were processed and corrected for partial volume effects using the method of Rousset et al. (1998) implemented in PVElab (pvelab{at}nru.dk) (Quarantelli et al., 2004). The point-spread function to perform partial volume correction was calculated by determining the true resolution of the PET camera (Prieto et al., 2010). Thus, a full-width at half-maximum of 7.3 mm was used for this purpose. Individual PET images were co-registered to high-resolution native-space T1-weighted MRI images using the mutual information method of SPM8. Then, the T1-weighted images were segmented into grey matter, white matter and CSF. The segmented grey matter images were spatially normalized into the Montreal Neurological Institute (MNI) grey matter provided by SPM8, and the inverse transformation matrix was used to label the segmented grey matter voxels in the patient’s MRI-PET space according to a template of volumes of interest defined in the MNI space (Supplementary Fig. 1) (Berkouk et al., 2006). For each volume of interest, partial volume effects-corrected, mean tracer concentrations were calculated for each participant. Briefly, segmented grey matter volumes of interest were corrected for the percentage of activity spilling from/to white matter whereas the CSF was constrained to zero because CSF is supposed to not have significant activity (Quarantelli et al., 2004). Because the size of the sample was small, comparisons between Alzheimer’s disease and healthy controls were performed using the Mann–Whitney U test. Alpha was set at 0.05 for all analyses.

In addition to using the arterial input function, we analysed the [11C]-flumazenil PET data and compared the two groups using the pons as a reference region (Supplementary material).


Neuropsychological performance

Patients with Alzheimer’s disease had impaired memory. In addition, they performed significantly worse than healthy controls in some language tasks, but not on executive function tasks (Table 2).

Volumetric magnetic resonance imaging

On volume of interest analysis, hippocampal volumes were smaller in the Alzheimer’s disease group (Fig. 2, Supplementary Table 1 and Supplementary Fig. 2). Cortical thickness and voxel-based morphometry yielded several regions, including the left medial temporal area, with decreased volume in Alzheimer’s disease at uncorrected P < 0.005, but none which survived corrected significance (Supplementary Table 2 and Supplementary Figs 3 and 4). All the voxel-based, parametric contrasts, are displayed at the same P-value, to facilitate comparison across imaging modalities.

[18F]-Fluorodeoxyglucose positron emission tomography

Left parietal association cortex and both precunei showed lower metabolism in Alzheimer’s disease (P < 0.005 uncorrected), but none of these comparisons survived corrected significance (Supplementary Table 3 and Supplementary Fig. 5).

[11C]-Flumazenil positron emission tomography

With voxel-based parametric analysis using arterial input function, flumazenil binding in patients with Alzheimer’s disease was decreased in the left medial temporal region, including entorhinal cortex (whole brain analysis, cluster-level FWE-corrected P = 0.048; volume of interest, voxel-level FWE-corrected P = 0.023); and with uncorrected significance (voxel level P < 0.005), there was decreased binding in the right inferomedial temporal region, left precuneus and supramarginal gyri of both hemispheres (Fig. 1 and Table 3). With volume of interest analysis, [11C]-flumazenil binding was decreased (P < 0.05) bilaterally in hippocampi and in the posterior cingulate gyrus (Supplementary Table 4 and Supplementary Fig. 6). Unlike with volumetric MRI, no false positives were recorded with [11C]-flumazenil binding (Fig. 2). Hippocampal [11C]-flumazenil binding correlated positively with measures of memory and other cognitive functions (Table 2 and Supplementary Fig. 7).

Figure 1

Regions where [11C]-flumazenil binding was decreased in subjects with Alzheimer’s disease compared to healthy controls (SPM8 voxel-based comparison, displayed at P < 0.005 uncorrected). Regions are displayed in red on rendered images of the brain surface (A) and in yellow-red on a standard axial template (B) using the neurological orientation (right hemisphere on the right side of each section). Note that in the medial temporal lobe, the findings reached corrected significance. The regions with decreased [11C]-flumazenil binding are similar to the regions described as bearing the greatest degree of neuronal loss in classical neuropathological studies of early Alzheimer’s disease (Supplementary Fig. 9).

Figure 2

Box-and-whisker diagrams of partial volume effects-corrected hippocampal [11C]-flumazenil binding (volume of distribution) and hippocampal volume (*ratio of individual hippocampal volume to individual whole-brain volume) in healthy controls (HC) and patients with Alzheimer’s disease (AD). As customary, box boundaries represent quartiles, the thick line in the box is the median and the whiskers represent the samples minimum and maximum or, in the case of open circles, outliers. Statistical comparison was performed using the non-parametric Mann–Whitney U test.

View this table:
Table 3

Regions of lower [11C]-flumazenil binding in patients with Alzheimer’s disease compared with healthy controls on the voxel-based SPM analysis (sorted by cluster-level FWE-corrected significance)

Area of between-group differenceBrodmann area(s)SidektZP-valueMaxima voxel
Cluster-levelVoxel-levelMNI coordinates
FWE-correctedUncorrected*(x, y, z)
Medial temporal lobe28, 35, 36L19806.604.01P = 0.048P < 0.000 (P = 0.023** FWE-corrected)−42,−64,−4
Supramarginal gyrus39, 40L3524.503.25P = 0.758P < 0.001−46,−28,20
Infero-medial temporal lobe28, 36R2424.243.14P = 0.865P < 0.00140,−26,18
Precuneus cortex31L1904.453.23P = 0.909P < 0.001−14,−72,30
Supramarginal gyrus39, 40R1814.213.12P = 0.916P < 0.00148,−34,22
Posterior parahippocampal gyrus35, 36R603.862.95P = 0.985P < 0.00212,−40,6
Anterior insular cortex44L514.503.25P = 0.988P < 0.001−34,26,8
Angular gyrus39R273.682.86P = 0.994P < 0.00246,−68,48
  • *Voxel-level uncorrected P-values are provided as is customary and to facilitate comparison with other techniques (Supplementary material) as most of the intergroup comparisons do not reach corrected significance.

  • **Corrected for a volume encompassing both temporal lobes, known to be the cortical regions with earliest neuronal loss.

  • FWE = family-wise error; SPM = Statistical Parametric Mapping.

With the analysis using the pons as reference region, no regions were identified reaching corrected significance on voxel-based testing (Supplementary Table 5) or in volume of interest analysis (Supplementary Table 6). However, in voxel-based testing, regions that differed across groups were detected at the uncorrected P < 0.005 voxel level, largest in the left medial temporal region (Supplementary Table 5 and Supplementary Fig. 8) that were very similar to the ones detected with the arterial input method.


We found [11C]-flumazenil binding to be decreased in regions which, on extensive histological sampling, have shown the greatest degree of neuronal loss in early Alzheimer’s disease. In a classical study, Brun and Englund (1981) found most marked neuronal loss in the medial temporal region, where [11C]-flumazenil binding was most decreased in our study, but also in the posterior cingulate gyrus and in the supramarginal gyri of both hemispheres, regions with a strong trend for decreased [11C]-flumazenil binding on voxel-based imaging (Supplementary Fig. 9), and in the case of the posterior cingulate, significantly lower for Alzheimer’s disease in volume of interest analysis (Supplementary Table 4 and Supplementary Fig. 6).

In a study comparing [11C]-flumazenil binding in older participants, two potential confounders were vascular disease and depression, both of which have been reported to lower [11C]-flumazenil binding (Heiss et al., 2004; Klumpers et al., 2010). Vascular disease was carefully excluded from the Alzheimer’s disease group, which served as a control for a different [18F]-fluorodeoxyglucose PET study (Pascual et al., 2010), where Alzheimer’s disease was compared with microvascular disease, the type of vascular disease most likely to be a confounder for this study. The depression rating scale score was higher in the group with Alzheimer’s disease, although not significantly so and [11C]-flumazenil binding correlated with depression scores (Table 2). However, we used the standard Geriatric Depression Scale to measure depression. Some of the items in this scale will be checked as positive not only in patients with depression, but also in those with cognitive impairment. For instance, the item that reads: ‘Is your mind as clear as it used to be?’ None of the patients with Alzheimer’s disease had clinical depression.

We did not find significantly decreased binding in the postero-lateral parietotemporal association cortex, where decreased metabolism is regularly seen in Alzheimer’s disease (Chetelat et al., 2008) and where an intermediate degree of neuronal loss was found by Brun and Englund (1981). Preserved [11C]-flumazenil binding in this region may result from a delicate balance of neuronal loss versus increased expression of the GABAA receptor in the remaining neurons. Receptor plasticity has been documented after lesions of projection pathways. In rodents, lesions of the perforant hippocampal pathway resulted in an acute loss of GABA receptors, but similar levels to controls were recorded 30 days after the acute lesion (Mizukami et al., 1997). A similar compensatory plasticity has been shown in Alzheimer’s disease using post-mortem tissue. In the hippocampal formation, immunohistochemical, biochemical and in situ hybridization methods have shown an increased expression of GABAA receptors in the surviving neurons (Armstrong et al., 2003; Iwakiri et al., 2009). Similar data are not available for the posterior parietotemporal association cortex. While marked neuronal loss in medial temporal regions (Brun and Englund, 1981; Gomez-Isla et al., 1996, 1997) would dictate that the total GABAA activity be reduced here, a milder loss in posterior parieto-temporal association cortex (Brun and Englund, 1981) may be compensated by an increase of GABAA receptors in the remaining neurons, leaving regional [11C]-flumazenil binding unaltered, as we found in this region.

Several methodological differences may explain the positive finding in our study while previous studies of [11C]-flumazenil in Alzheimer’s disease, with a number of patients similar to ours, were negative (Meyer et al., 1995; Ohyama et al., 1999). First, we used voxel-based comparisons. By comparing all the voxels in the brain of patients with Alzheimer’s disease and healthy controls, decreased binding can be found in areas that, when averaged with preserved areas in a volume of interest methodology, may pass undetected. In our study, apart from extensive changes in the medial temporal regions, other areas of decreased binding were small and likely to disappear when averaged with tissue containing normal binding values. For instance, despite decreased binding in both supramarginal gyri and the left precuneus (Fig. 1), overall binding in parietal lobes was not decreased in our volume of interest analysis (Supplementary Table 4 and Supplementary Fig. 6). Secondly, we used native space MRI to co-register it with native PET data and bring PET data to standard space. This technique not only allowed us to correct for partial volume effects, but also improved the quality of PET data in standard space and therefore the yield of all the comparisons performed across subjects and groups of subjects (Shan et al., 2011). In addition, our subject samples were better matched by sex (Meyer et al., 1995) and by age (Meyer et al., 1995; Ohyama et al., 1999).

Our findings agree with binding studies on Alzheimer’s disease post-mortem tissue showing a 40% loss of benzodiazepine binding in the entorhinal cortex (Jansen et al., 1990) and variable losses of benzodiazepine or GABAA binding in the hippocampus and temporal cortex (Chu et al., 1987; Shimohama et al., 1988). As densitometric measurements on post-mortem tissue are confounded by artefactual volume loss, researchers working with tissue have often placed their findings in the context of in vivo receptor studies performed with PET and were cautious given the previously negative PET studies (Meyer et al., 1995; Ohyama et al., 1999). Our study, on the contrary, suggests that GABAA changes can be detected in early Alzheimer’s disease using current imaging techniques. We studied early Alzheimer’s disease because up to 40% of patients with mild cognitive impairment will not have prodromal Alzheimer’s disease (Petersen, 2004). However, a quantification of hippocampal GABAA messenger RNA in post-mortem Alzheimer’s disease and prodromal Alzheimer’s disease–mild cognitive impairment tissue found a similar reduction of GABAA receptors at both stages of the disease (Rissman et al., 2004), suggesting that it is worthwhile to look for changes in [11C]-flumazenil binding in mild cognitive impairment or even preclinical stages.

Other than the hippocampal volumes on the volume of interest analysis, the findings on structural MRI did not reach corrected statistical significance. Given that structural MRI has proven useful in Alzheimer’s disease and mild cognitive impairment (Misra et al., 2009), the lack of significance in our study probably reflects the small sample size. Yet, the regions showing a trend for decreased volume were smaller but anatomically similar to areas showing decreased [11C]-flumazenil binding (Supplementary Fig. 4). This finding supports the notion that decreased GABAA binding reflects neuronal loss (la Fougere et al., 2011), which will also cause atrophy. However, the effect size for volume loss, both in the voxel-based analysis and the volume of interest analysis of volume (compare r values in Supplementary Tables 1 and 4), was smaller than for [11C]-flumazenil binding loss.

[18F]-Fluorodeoxyglucose PET showed a trend for decreased metabolism in the regions where this finding has been classically described in Alzheimer’s disease (Chetelat et al., 2008). A greater magnitude of neuronal loss, measured as decreased [11C]-flumazenil binding, in the left medial temporal region (Fig. 1 and Table 3), corresponded to a greater loss of metabolism in the ipsilateral angular gyrus and retrosplenial region (Supplementary Fig. 5). This finding supports the concept that loss of metabolic activity in parietal association cortex may result, at least in part, from a reduction in the rich projections reaching this region from inferomedial temporal cortex (Lavenex et al., 2002; Munoz and Insausti, 2005). It also highlights the dissociation between neuronal loss and loss of metabolism that has been reflected in other studies by a regional dissociation between atrophy and metabolic loss (Chetelat et al., 2008).

The applicability of [11C]-flumazenil PET imaging to the study of large samples of patients with Alzheimer’s disease and controls would be greatly limited by the need to obtain arterial sampling, which is an invasive procedure. Fortunately, reference region methods have been described that obviate the need for arterial sampling in order to obtain an input function (Hammers et al., 2008; Mourik et al., 2008; Klumpers et al., 2008). In our study, using the pons as a reference region yielded intergroup differences in medial temporal regions and precuneus that were significant at the uncorrected P < 0.005 level, which compares favourably to studies of [11C]-flumazenil PET binding in epilepsy (Hammers et al., 2008) or depression (Klumpers et al., 2012) using the pons as reference region. However, at the same threshold, the arterial input method provided larger and more significant regions on voxel-based analyses and a greater intergroup effect on volume of interest analyses (Fig. 1 and Table 3; Supplementary Fig. 8 and Supplementary Tables 46). This finding agrees with the dictum that the ideal input function should not contain specific binding. Although in smaller amounts than in the cortex, GABAA receptors are present in the pons, and therefore the presence of some specific binding in this region makes it less than ideal as a reference region (Delforge et al., 1995). It is possible that using carotid pixel density as an input function may yield findings even more similar to the arterial-input-function method, yet without the need for arterial sampling (Mourik et al., 2008). Another potential drawback of using [11C]-flumazenil is the short half-life of 11C, which dictates that a cyclotron be available at the imaging site. To address this issue, flumazenil labelled with 18F, an isotope with a longer half-life, is being readied for clinical use (Massaweh et al., 2009; la Fougere et al., 2011).

[11C]-Flumazenil is not the only agent available to image GABAA receptors in vivo. [123I]-Iomazenil, a partial agonist, has also been used to image GABAA receptors in Alzheimer’s disease by means of single photon emission computed tomography. The two [123I]-iomazenil studies in Alzheimer’s disease were positive, but data on medial temporal binding, the region with the greatest neuronal loss, were not available (Soricelli et al., 1996; Fukuchi et al., 1997). The patients in both studies had a lower Mini-Mental State Examination score than in our study, 14.7 ± 2.3 in Soricelli et al. (1996), suggesting more advanced Alzheimer’s disease than in our sample (Student’s t-test, two-tailed, P = 0.02). A study of [123I]-iomazenil in mild cognitive impairment has been negative (Pappata et al., 2010). Our findings may encourage more research with this compound. However, because the methodologies differ substantially, it is outside the scope of this paper to discuss investigations performed with [123I]-iomazenil.

Using a sample size similar to those in previous studies of [11C]-flumazenil binding, but updated methodology, we were able to separate patients with early Alzheimer’s disease from healthy controls. As the findings on [11C]-flumazenil PET reached statistical significance in precisely the areas of the brain known to have greatest neuronal loss, this report may encourage larger studies, which could confirm the usefulness of [11C]-flumazenil PET for the detection of neuronal loss in early clinical or even preclinical stages.


The EC-FP6-project DiMI of the European Community [LSHB-CT-2005-512146], the ‘Unión Temporal de Empresas (UTE): Centro de Investigación Médica Aplicada,’ University of Navarra, the ‘Fundación Mutua Madrileña’, the Government of Navarra [13255308-53-GN], and the ‘Centro de Investigación Biomédica en Red de Enfermedades Neurodegenerativas (CIBERNED) of the Spanish Ministry of Health.

Supplementary material

Supplementary material is available at Brain online.


The patients and healthy controls who participated in this study are gratefully acknowledged. We thank Dr Karl Herholz for leadership and many important suggestions, Dr Bradford Dickerson for helpful suggestions, Ms Jaione Irigoyen for providing nursing care and Mr Shane Kippenhan for help with the MRI morphometric procedure.


  • *Present address: Frontotemporal Dementia Unit, Athinoula A. Martinos Centre for Biomedical Imaging, Department of Neurology, Massachusetts General Hospital; and Department of Neurology, Harvard Medical School, Charlestown, Massachusetts 02129, USA

  • Present address: Section on Integrative Neuroimaging, Clinical Brain Disorders Branch, Intramural Research Program, National Institutes of Health, Bethesda, MD 20892, USA


View Abstract