Original Research

Decreased g-Aminobutyric Acid Levels in theParietal Region of Patients With Alzheimer’sdisease

Xue Bai, MD,1 Richard A.E. Edden, PhD,2,3 Fei Gao, MD,1 Guangbin Wang, MD,1*

Lebin Wu, MD,1 Bin Zhao, MD,1 Minzhong Wang, MD,4 Queenie Chan, PhD,5

Weibo Chen, MS,6 and Peter B. Barker, DPhil2,3

Purpose: To determine whether there are in vivo differ-ences of g-aminobutyric acid (GABA) levels in frontal andparietal regions of Alzheimer’s disease (AD) patients, com-pared with healthy controls using magnetic resonancespectroscopy (1H-MRS).

Materials and Methods: Fifteen AD patients and fifteenage- and gender-matched healthy controls underwent 1H-MRS of the frontal and parietal lobes using the “MEGA-Point Resolved Spectroscopy Sequence” (MEGA-PRESS)technique, and cognitive levels of subjects were evaluatedusing Mini-Mental State Examination (MMSE) tests. MRSdata were processed using the Gannet program. Becausethe signal detected by MEGA-PRESS includes contribu-tions from GABA, macromolecules and homocarnosine, itis labeled as “GABAþ” rather than GABA. Differences ofGABAþ/Cr ratios between AD patients and controls weretested using covariance analysis, adjusting for gray mat-ter fraction. The relationship between GABAþ/Cr andMMSE scores was also analyzed.

Results: Significant lower GABAþ/Cr ratios were foundin the parietal region of AD patients compared with con-trols (P¼0.041). In AD patients, no significant correla-tions between GABAþ/Cr and MMSE scores were foundin either the frontal (r¼�0.164; P¼0.558) or parietalregions (r¼0.025; P¼0.929).

Conclusion: Decreased GABAþ/Cr levels were presentin the parietal region of patients with AD in vivo, suggest-ing that abnormalities of the GABAergic system may bepresent in the pathogenesis of AD.

Key Words: Alzheimer’s disease; GABA; 1H-MRS; MMSE;MEGA-PRESSJ. Magn. Reson. Imaging 2014;00:000–000.VC 2014 Wiley Periodicals, Inc.

ALZHEIMER’S DISEASE (AD) is a progressive neuro-degenerative disorder, characterized by behavioralabnormalities and cognitive deficits. Increasing evi-dence suggests that inhibitory neurotransmitter sys-tem dysfunction may be involved in AD (1). g-Aminobutyric acid (GABA) is the predominant inhibi-tory neurotransmitter in the human brain (2), and isconsidered to be associated with cognitive function(1). A variety of studies, including postmortem (3),neuroimaging (4), and animal model systems (5),have suggested that GABAergic dysfunction isinvolved in AD.

MRS enables the noninvasive in vivo measurementof major neurometabolite levels in various neurodege-nerative diseases (6). However, the accuracy of con-ventional in vivo 1H-MR spectroscopy for themeasurement of GABA measurement is limited by itsrelatively low concentration, and spectral overlap withthe signals of glutamate (Glu), N-acetylaspartate(NAA), total creatine (Cr), and macromolecules (MM).With the development of edited MRS methods, how-ever, such as MEGA-Point Resolved Spectroscopymethod (MEGA-PRESS), brain GABA levels may bemore reliably estimated (7), and has been measuredboth in normal human brain (8) and in various psy-chiatric diseases (9).

Moreover, the relationship between brain metabo-lites levels and cognitive impairment remains poorlyexplored. Although the correlation between NAA andMini-Mental State Examination (MMSE) scores hasbeen reported (10), the relationship between GABAlevel and cognitive impairments remains unknown.

1Shandong Medical Imaging Research Institute, Shandong University,Jinan, Shandong, China.2Russell H. Morgan Department of Radiology and Radiological Sci-ence, The Johns Hopkins University School of Medicine, Baltimore,Maryland, USA.3F.M. Kirby Center for Functional Brain Imaging, Kennedy KriegerInstitute, Baltimore, Maryland, USA.4Shandong Provincial Hospital, Shandong University, Jinan, Shan-dong, China.5Philips Healthcare, 6/F, Core Building 1, Hong Kong Science Park,Shatin, New Territories, Hong Kong.6Philips Healthcare, Philips Innovation Campus Shanghai Building 1No. 10, Shanghai, China.

Contract grant sponsor: The National Natural Science Foundation ofChina; Contract grant number: 81171380/H1807.

*Address reprint requests to: G.W., Shandong Medical ImagingResearch Institute, Shandong University, 324#, Jingwu Road, Jinan250021, Shandong, PR China. E-mail: wgb7932596@hotmail.com

Received February 12, 2014; Accepted May 6, 2014.

DOI 10.1002/jmri.24665View this article online at wileyonlinelibrary.com.

JOURNAL OF MAGNETIC RESONANCE IMAGING 00:00–00 (2014)

CME

VC 2014 Wiley Periodicals, Inc. 1

In this initial study, the edited MRS techniqueMEGA-PRESS was used to explore in vivo brain GABAlevels in patients with AD, and age- and gender-matched healthy control subjects. The assessment ofthe relationship between GABA levels and cognitiveimpairments in AD patients were also performed.

MATERIALS AND METHODS:

Subjects

Fifteen patients with clinical diagnosis of AD (9 womenand 6 men aged 55–79 years, mean¼65.736 8.53years) and 15 age- and gender-matched healthy controlsubjects (HC) (8 women and 7 men aged 61–76 years,mean¼66.336 4.62 years) were recruited in this study(Table 1). AD patients were diagnosed according to theDiagnostic and Statistical Manual of Mental DisordersDSM-IV-TR Fourth Edition (DSM-IV) (11), the NationalInstitute of Neurological and Communicative Disordersand the Alzheimer’s Disease and Related DisordersAssociation criteria (NINDS-ADRDA) (12). To obtain anoverall assessment of cognitive function, all subjectswere administered the MMSE (13) by a trained neuro-psychologist, blinded to the MRI data, on the same dayof the MR examination. Exclusion criteria included: (i)a history of a severe brain trauma or psychiatric dis-ease; (ii) intake of medication that may affect cognitivefunction, e.g., donepezil, memantine; (iii) major ische-mic vascular damage, e.g., cerebral ischemic stroke; or(iv) subcortical arteriosclerotic encephalopathy, leu-koaraiosis, as diagnosed by conventional brain MRI.The age- and gender-matched control subjects had nohistory of cognitive decline, previous neurologic or psy-chiatric disorder. Local Ethical Committee approval andwritten informed consent from all the subjects andtheir guardians were obtained before study initiation.

MRI/MRS Study Protocol

All MRI/MRS experiments were performed on a 3 Tesla(T) scanner (Philips Achieva TX, Best, The Netherlands)

equipped with an eight-channel phased-array headcoil. Before the MRS examination, T1-weighted three-dimensional (3D) turbo field echo (TFE) images wereused for localization, with major parameters as follows:TR¼8.2 ms; TE¼3.7 ms; slice thickness¼1 mm; reso-lution¼256 � 256; FOV¼24�24 cm2; and flipangle¼8

�. The MRS voxels (volume size 3� 3 �3 cm3)

were set on medial frontal region and the parietalregion (Fig. 1). These regions were chosen because theywere large enough to accommodate the 27 cm3 voxel,and prior postmortem studies have suggested GABAer-gic dysfunction in frontal, parietal (and temporal)regions (1). The median sagittal plane was selected as areference slice for voxel localization: the voxel in thefrontal region was prescribed superior to the genu ofthe corpus callosum, aligned with the shape of the cor-pus callosum, and positioned on the medial aspect ofthe axial plane; the voxel in the parietal region was pre-scribed posterior and superior to the splenium of thecorpus callosum, aligned with the shape of the corpuscallosum, and positioned again on the medial aspect ofthe axial plane. All of the voxels were positioned in amanner to avoid the lateral ventricle and skull.

1H-MRS

The MEGA-PRESS sequence (7) was used for GABAediting, with the parameters as follows: TR¼2000 ms;TE¼68 ms; 320 signal averages; acquisitionbandwidth¼1000 Hz; scan duration 13 min. J-evolution for GABA was refocused during odd-numbered acquisitions (ON) but not during even-numbered acquisitions (OFF) by applying Gaussianinversion pulse to the 3CH2 resonance of GABA at 1.9ppm (ON) and symmetrically about the water peak at7.5 ppm (OFF), respectively. Water suppression wascarried out using chemical shift-selective (CHESS)pulses after automatic optimization. FASTMAP shim-ming of the VOI was conducted automatically beforeeach acquisition.

The difference of the “ON” and “OFF” spectra pro-vided an edited spectrum of GABA. The signal

Table 1

Demographic, MRS, and Segmentation Dataa of AD Patients and Control Subjects

AD HC P valueb

Number 15 15 –

Age 65.7368.53y 66.3364.62y 0.81

Female, no. (%) 9 (60.00%) 8 (57.33%) 0.724

Age of onset 64.0769.38y – –

Duration of illness 2.2761.01y – –

MMSE 15.8765.03 29.2060.86 <0.001

Frontal region:

GABAþ/Cr level 0.09860.008 0.10460.022 0.345

GABAþ FitError 7.04%61.61% 5.60%61.20% 0.151

GM/(GMþWM) 55.30%65.29% 56.94%61.34% 0.26

Parietal region:

GABAþ/Cr level 0.08660.016 0.09960.012 0.041

GABAþ FitError 6.11%61.84% 5.51%62.42% 0.900

GM/(GMþWM) 54.69%65.70% 57.61%62.00% 0.07

AD¼Alzheimer’s disease; HC¼healthy controls; MMSE¼Mini-Mental State Examination; GM¼gray matter; WM¼white matter.aData are given as mean 6standard deviation (SD).bSignificant differences between groups are tested by analysis of covariance with P<0.05 accepted as significant.

2 Bai et al.

detected at 3.02 ppm with these parameters and theMEGA-PRESS technique is known to contain contribu-tions from both macromolecules (MM) and homocarno-sine (14), hence, the detected signal is referred to asGABAþ rather than GABA. Quantification was per-formed using the Gannet 2.0 toolkit, a Matlab-basedquantitative batch analysis tool for analyzing GABAMEGA-PRESS spectra (15). Gannet contains two mod-ules: GannetLoad and GannetFit. The GannetLoadmodule is used to parse certain variables from thedata headers, apply a line broadening of 3 Hz, and fre-quency and phase correct the individual spectra usingSpectral Registration (16). GannetFit uses a single-Gaussian model to fit the edited GABAþ signal andevaluates GABAþ relative to creatine (Cr), which hasbeen reported to be unchanged in AD in previous stud-ies (17,18). In this manuscript, the GABAþ/Cr ratio isreferred to as GABAþ level. GannetFit also providesthe standard deviation of the fitting residual divided bythe amplitude of the fitted peaks, generating the overallFitError which reflects the signal-to-noise ratio. Onlyspectra with a relative FitError of GABAþ below 10%were used for the statistical analysis.

VOI Segmentation

Different voxel proportions of gray matter (GM), whitematter (WM), and cerebrospinal fluid (CSF) in 1H-MRSmay confound group differences in metabolite mea-surement. To determine if tissue composition differen-ces, between AD subjects and healthy controls, couldaccount for differences in GABAþ level, each MRSvoxel was segmented as GM, WM, or CSF using the 3DT1-weighted brain images and the automatic brain seg-mentation program (Fig. 1), FAST (FMRIB’s automatedsegmentation tool) in the FSL package (Oxford Univer-

sity, Oxford, UK) (19). The VOIs were co-registered tothe anatomical images using the “Re-creation of VOI”Matlab tool (20). Tissue GM fractions were obtained bycalculating the ratio of GM volume to the GMþWM vol-umes in the VOIs. The concentrations of GABA and Crin CSF were considered to be negligible.

Statistical Analysis

Statistical analyses were conducted using the Statisti-cal Package for Social Sciences software (SPSS 17.0,Chicago, IL). MRS data were presented as mean6 stan-dard deviation (SD) values. The differences of GABAþlevels between patients and controls were tested usinganalysis of covariance (ANCOVA), adjusting for GM/(GMþWM). Spearman correlation coefficients wereused to assess the presence of linear associationsbetween GABAþ levels and MMSE scores. P-values ofless than 0.05 were considered as significant.

RESULTS

The Demographic Characteristics of StudySubjects

The demographic information and the results of mem-ory test of the subjects are presented in Table 1.There was no difference in age (P¼0.812) and genderdistribution (P¼0.724) between the two groups. TheAD patients (range, 10–23; mean¼15.9 6 5.0) per-formed significantly worse on the MMSE than controls(range, 28–30; mean¼29.2 6 0.9) (P<0.001).

VOI Segmental Results

The mean GM tissue fraction GM/(GMþWM) was55.3% and 54.7% in the frontal and parietal regions,

Figure 1. The position of voxels, corresponding segmentation data and representative spectra. The (a,d) and (b,e) panelsshow the position of volumes of interest in frontal (upper) and parietal regions (lower) on sagittal and coronal T1-weightedTFE images in an AD patient. The corresponding results of brain segmentation are shown in the frontal (c) and parietalregions (f). (g, h) Show the representative MEGA-PRESS spectra processed using the “Gannet 2.0” toolkit in frontal and parie-tal regions of AD. (g) Shows the curve-fitting of the GABA peak using Gannet, the red lines in the panels are the results of theGannetFit curve-fitting, the blue lines show the postphase and frequency aligned GABA data, and the black line is the resid-ual difference between the experimental data and the curve-fit. (h) is the MEGA-PRESS difference spectrum. The Glx, GABAand MM peaks resonate at 3.7, 3.0 ppm and 0.9 ppm respectively.

In Vivo GABA Levels Measurement in AD 3

respectively, in AD patients, and 56.9% and 57.6%,respectively, in HC (Table 1). There were no significantdifferences in GM fraction in the VOIs between ADpatients and HC (frontal region, p¼0.254; parietalregion, p¼0.072).

Comparisons of GABA1 Level Between Groups(AD Patients versus Control Subjects)

Edited spectra were successfully obtained from thefrontal and parietal regions in all 15 AD patients and15 controls (Fig. 2), with mean GABAþ FitError offrontal and parietal voxels 7.0% and 5.6%, respec-tively, for AD patients, 6.1% and 5.5% for HC (Table1). A significantly lower GABAþ level was found inparietal region of AD patients, relative to controls(P¼0.041). Lower mean GABAþ levels were alsofound in the frontal region (0.098 6 0.008), comparedwith HC (0.104 6 0.022), but this was not statisticallysignificant (P¼0.345). (Table 1; Fig. 2).

Correlation of GABA1 Levels With CognitiveFunction

MMSE scores were evaluated in all subjects. No sig-nificant correlations were found between MMSE andGABAþ levels, either in frontal (r¼�0.164, P¼0.558)or parietal voxels (r¼�0.025, P¼0.929). Also, withregard to the correlation of MMSE scores and GABAþlevels in healthy controls, no statistical correlationwas detected (frontal region: r¼0.178, P¼0.526; pari-etal region: r¼�0.197, P¼0.481).

DISCUSSION

The main result of this study is that significantlydecreased GABAþ levels were present in the parietalregion of AD patients, as compared to age- andgender-matched healthy controls. The mean GABAþlevels in the frontal region also showed a trend (notstatistically significant) to be lower in AD than HC. Tothe best of our knowledge, this is the first in vivo dem-

onstration of reduced GABAþ levels in AD patients.Because GABA is primarily localized to inhibitory neu-rons (21), GABAþ level deficits could indicate eitherloss or dysfunction of GABAergic neurons in AD.These findings are consistent with previous postmor-tem studies, which observed decreased GABA levels intemporal, parietal, occipital and frontal cortex in ADsamples (3,22–24). Similar findings have also beenreported in animal model studies: a transgenic AD ratmodel study found lower GABA levels in dorsal hippo-campus and frontal cortex at age 9 months (25), andan MRS study of mouse AD model (5xFAD) also founddecreased GABA levels at age 40 months (26).

As stated above, the GABA deficit may be due toeither GABAergic neuronal loss or dysfunction; if dueto dysfunction, either GABA synthesis itself or enzy-matic deficiency in glutamate-glutamine cycling maybe implicated (27). A study of the TgCRND8 AD mousemodel found lower glutamate decarboxylase (GAD67 -the enzyme that catalyses the conversion of glutamateto GABA) in the hippocampus, and suggested thatamyloid-b toxicity contribute to the GABA neuron dys-function, followed at later stages by cell death (28).Single photon emission computed tomography (SPECT)studies have also demonstrated reduced GABAA/ben-zodiazepine (BZD) receptor density in the frontal andparietotemporal lobes of patients with AD, and in theparietal lobe of patients with mild cognitive impairment(MCI) (29,30). GABA-A subunit studies found that, inthe hippocampus, mRNA for GABA-A subunits a1, a5,b3 were decreased in AD patients with severe neuropa-thology (31) and a1, a2 GABA-A subunit expressionwas reduced in early and late AD stages (32).

Because the results of this and prior studies sug-gest that the GABAergic system is involved in thepathogenesis of AD, it is, therefore, also a potential atarget for AD therapies. Of interest, it was found thatthe administration of a GABA-chloride ionophoreantagonist to aging rats could protect the animalsfrom age-related decline in cognitive functions, andlosses of hippocampal neurons (5), suggesting thatthe GABAergic systems does contribute to cognitivedysfunction in AD and may be amenable to prophy-lactic treatments. However, the current study and aprevious SPECT study on GABAA/ BZD receptors inAD found no significant correlation between MMSEscores and GABAA/BZD receptor availability in anybrain regions tested (30). The lack of correlationbetween MMSE and regional GABAþ levels may bedue in part to the limited sample size, and a follow-upstudy in a larger cohort of AD patients will be requiredto answer this question definitively.

Limitations

The data acquisition strategy for this study has sev-eral limitations. First, as in most GABA MRS studies,the voxel volume used was relatively large (3 � 3 �3 cm) due to the low intensity of the GABAþ signal. Asmaller-VOI MRS approach may prove more efficientfor region-specific analyses of brain GABA levels. Sec-ond, the MEGA-PRESS technique as implementedhere at 3T, in addition to editing GABA, also contains

Figure 2. The GABAþ/Cr ratios of AD patients and HC infrontal (a) and parietal (b) regions. The mean and standarddeviation of GABAþ/Cr ratios are displayed. a: Shows thatthe mean GABAþ/Cr ratios are lower in AD than in HC infrontal region (P¼0.345), but no statistical difference isdetected. There does appear to be one “outlier” value in thecontrol group, however, no significant differences are foundin this region between AD and HC, whether this point isincluded or not. b: Shows GABAþ/Cr ratios of AD group aresignificantly decreased in the parietal region (P¼0.041).

4 Bai et al.

contributions from co-edited macromolecules (MM),as well as a smaller contribution from the GABA-dipeptide homocarnosine. New methods for macro-molecule suppression are being developed (33) whichcan detect “pure GABA,” and these methods shouldbe used in future studies of AD. On the other hand,the concentration of homocarnosine in vivo is muchlower than that of GABA (34), so isolated changes inhomocarnosine are less likely to be driving theobserved changes. Third, the study acquisition proto-col did not include acquisition of nonsuppressedwater scans, which precluded the calculation ofGABAþ concentrations relative to the brain water sig-nal. The results presented here are given as ratios toCr, meaning that we were only able to present metab-olite ratios. However, quantification relative to Cr islikely to be more robust (than water referencing) topossible group differences in brain water content andCSF due to atrophy (17,18). Fourth, the sample sizeof 15 in each group is rather small for a clinical studyof this kind, and, therefore, these results should beviewed as preliminary, and will require confirmationin larger numbers of subjects. It is possible that thelack of an observed difference in the frontal region ismore due to insufficient power (from a combination ofsample size and high variance among controls) thanany true regional difference.

Finally, although many prior MRS studies on ADhave focused on the hippocampal region because ofits important role in AD pathology, in this study thehippocampus was not studied. Because of the lowconcentration of GABA in the brain, the VOI ofMEGA-PRESS technique is relatively large (3 � 3 �3 cm3), much larger than the hippocampus itself, andwould, therefore, have a considerable amount of par-tial volume with surrounding tissues. Additionally,hippocampal atrophy is a nearly ubiquitous finding inAD (35–37). The proportions of gray matter and whitematter differ between AD patients and healthy con-trols, which may confound the results of GABA levelalterations. As mentioned above, the MEGA-PRESSvoxels were placed on the medial frontal and the pari-etal regions instead of the hippocampus region.

In conclusion, decreased GABAþ levels, as meas-ured by edited MRS, were found in the parietal regionof AD patients compared with age-matched controls.These in vivo results add to previous in vitro and ani-mal model studies which suggest GABAergic systemdysfunction in AD pathogenesis, as well as a potentialtreatment target for AD. More studies in larger num-bers of patients, preferably using MM-suppressiontechniques, are required to probe the relation betweenregional GABA levels and cognitive impairment.

ACKNOWLEDGMENT

The authors thank the patients and healthy partici-pants who volunteered for this study.

REFERENCES

1. Lanctot KL, Herrmann N, Mazzotta P, Khan LR, Ingber N.GABAergic function in Alzheimer’s disease: evidence for dysfunc-

tion and potential as a therapeutic target for the treatment ofbehavioural and psychological symptoms of dementia. Can JPsychiatry 2004;49:439–453.

2. Obata K. Synaptic inhibition and gamma-aminobutyric acid inthe mammalian central nervous system. Proc Jpn Acad Ser BPhys Biol Sci 2013;89:139–156.

3. Seidl R, Cairns N, Singewald N, Kaehler ST, Lubec G. Differencesbetween GABA levels in Alzheimer’s disease and Down syndromewith Alzheimer-like neuropathology. Naunyn Schmiedebergs ArchPharmacol 2001;363:139–145.

4. Ball S, Busatto GF, David AS, et al. Cognitive functioningand GABAA/benzodiazepine receptor binding in schizophrenia:a 123I-iomazenil SPET study. Biol Psychiatry 1998;43:107–117.

5. Marczynski TJ. GABAergic deafferentation hypothesis of brainaging and Alzheimer’s disease revisited. Brain Res Bull 1998;45:341–379.

6. Bartha R, Smith M, Rupsingh R, Rylett J, Wells JL, Borrie MJ.High field (1)H MRS of the hippocampus after donepezil treatmentin Alzheimer disease. Prog Neuropsychopharmacol Biol Psychiatry2008;32:786–793.

7. Mescher M, Merkle H, Kirsch J, Garwood M, Gruetter R. Simulta-neous in vivo spectral editing and water suppression. NMRBiomed 1998;11:266–272.

8. Choi IY, Lee SP, Merkle H, Shen J. In vivo detection of gray andwhite matter differences in GABA concentration in the humanbrain. Neuroimage 2006;33:85–93.

9. Yoon JH, Maddock RJ, Rokem A, et al. GABA concentration isreduced in visual cortex in schizophrenia and correlates withorientation-specific surround suppression. J Neurosci 2010;30:3777–3781.

10. Huang W, Alexander GE, Chang L, et al. Brain metabolite con-centration and dementia severity in Alzheimer’s disease: a (1)HMRS study. Neurology 2001;57:626–632.

11. Gmitrowicz A, Kucharska A. [Developmental disorders in thefourth edition of the American classification: diagnostic and sta-tistical manual of mental disorders (DSM IV – optional book)].Psychiatr Pol 1994;28:509–521.

12. McKhann G, Drachman D, Folstein M, Katzman R, Price D,Stadlan EM. Clinical diagnosis of Alzheimer’s disease: report ofthe NINCDS-ADRDA Work Group under the auspices of Depart-ment of Health and Human Services Task Force on Alzheimer’sDisease. Neurology 1984;34:939–944.

13. Folstein MF, Folstein SE, McHugh PR. “Mini-mental state”. Apractical method for grading the cognitive state of patients for theclinician. J Psychiatr Res 1975;12:189–198.

14. Rothman DL, Behar KL, Prichard JW, Petroff OA. Homocarnosineand the measurement of neuronal pH in patients with epilepsy.Magn Reson Med 1997;38:924–929.

15. Edden RAE, Puts NA, Harris AD, Barker PB, Evans CJ. Gannet: abatch-processing tool for the quantitative analysis of GABA-edited MRS spectra. J Magn Reson Imaging 2013 (in press). doi:10.1002/jmri.24478.

16. Near J, Edden R, Evans CJ, Paquin R, Harris A, Jezzard P. Fre-quency and phase drift correction of magnetic resonance spec-troscopy data by spectral registration in the time domain. MagnReson Med 2014 (in press). doi: 10.1002/mrm.25094.

17. Graff-Radford J, Kantarci K. Magnetic resonance spectroscopy inAlzheimer’s disease. Neuropsychiatr Dis Treat 2013;9:687–696.

18. Valenzuela MJ, Sachdev P. Magnetic resonance spectroscopy inAD. Neurology 2001;56:592–598.

19. Zhang Y, Brady M, Smith S. Segmentation of brain MR imagesthrough a hidden Markov random field model and theexpectation-maximization algorithm. IEEE Trans Med Imaging2001;20:45–57.

20. Roebuck JR, Windham JP, Hearshen DO. Segmentation of MRSsignals using ASPECT (analysis of SPectra using EigenvectorDecomposition of Targets). Med Phys 1994;21:277–285.

21. Kelsom C, Lu W. Development and specification of GABAergiccortical interneurons. Cell Biosci 2013;3:19.

22. Lowe SL, Francis PT, Procter AW, Palmer AM, Davison AN, BowenDM. Gamma-aminobutyric acid concentration in brain tissue attwo stages of Alzheimer’s disease. Brain 1988;111:785–799.

23. Sasaki H, Muramoto O, Kanazawa I, Arai H, Kosaka K, Iizuka R.Regional distribution of amino acid transmitters in postmortembrains of presenile and senile dementia of Alzheimer type. AnnNeurol 1986;19:263–269.

In Vivo GABA Levels Measurement in AD 5

24. Ellison DW, Beal MF, Mazurek MF, Bird ED, Martin JB. A post-mortem study of amino acid neurotransmitters in Alzheimer’s dis-ease. Ann Neurol 1986;20:616–621.

25. Nilsen LH, Melo TM, Saether O, Witter MP, Sonnewald U. Alteredneurochemical profile in the McGill-R-Thy1-APP rat model ofAlzheimer’s disease: a longitudinal in vivo 1 H MRS study.J Neurochem 2012;123:532–541.

26. Mlynarik V, Cacquevel M, Sun-Reimer L, et al. Proton and phos-phorus magnetic resonance spectroscopy of a mouse model ofAlzheimer’s disease. J Alzheimers Dis 2012;31:S87–S99.

27. Kugler P. Enzymes involved in glutamatergic and GABAergic neu-rotransmission. Int Rev Cytol 1993;147:285–336.

28. Krantic S, Isorce N, Mechawar N, et al. Hippocampal GABAergicneurons are susceptible to amyloid-beta toxicity in vitro and aredecreased in number in the Alzheimer’s disease TgCRND8 mousemodel. J Alzheimers Dis 2012;29:293–308.

29. Hanyu H, Kume K, Sato T, et al. Regional differences in corticalbenzodiazepine receptors of Alzheimer, vascular, and mixeddementia patients. J Neurol Sci 2012;323:71–76.

30. Pappata S, Varrone A, Vicidomini C, et al. SPECT imaging ofGABA(A)/benzodiazepine receptors and cerebral perfusion inmild cognitive impairment. Eur J Nucl Med Mol Imaging 2010;37:1156–1163.

31. Rissman RA, De Blas AL, Armstrong DM. GABA(A) receptors inaging and Alzheimer’s disease. J Neurochem 2007;103:1285–1292.

32. Luchetti S, Bossers K, Van de Bilt S, et al. Neurosteroid biosyn-thetic pathways changes in prefrontal cortex in Alzheimer’s dis-ease. Neurobiol Aging 2011;32:1964–1976.

33. Edden RA, Puts NA, Barker PB. Macromolecule-suppressedGABA-edited magnetic resonance spectroscopy at 3T. MagnReson Med 2012;68:657–661.

34. Govindaraju V, Young K, Maudsley AA. Proton NMR chemicalshifts and coupling constants for brain metabolites. NMR Biomed2000;13:129–153.

35. Clerx L, van Rossum IA, Burns L, et al. Measurements of medialtemporal lobe atrophy for prediction of Alzheimer’s disease insubjects with mild cognitive impairment. Neurobiol Aging 2013;34:2003–2013.

36. Oosterman JM, Oosterveld S, Rikkert MG, Claassen JA, KesselsRP. Medial temporal lobe atrophy relates to executive dysfunctionin Alzheimer’s disease. Int Psychogeriatr 2012;24:1474–1482.

37. Visser PJ, Verhey FR, Hofman PA, Scheltens P, Jolles J. Medialtemporal lobe atrophy predicts Alzheimer’s disease in patientswith minor cognitive impairment. J Neurol Neurosurg Psychiatry2002;72:491–497.

6 Bai et al.