Age- and sex-related effects on the neuroanatomy of healthy elderly pdf

Actually, as the majority of these studies were based on large age range cohorts, little is actually known about the effect of sex on age-related changes in brain structure of healthy el

Age- and sex-related effects on the neuroanatomy of healthy elderly Herve´ Lemaıˆtre,a Fabrice Crivello,a Blandine Grassiot,a Annick Alpe´rovitch,b

Christophe Tzourio,b and Bernard Mazoyera,c,d,T

a Groupe d’Imagerie Neurofonctionnelle, UMR 6194, CNRS, CEA, Universite´s de Caen et Paris 5, GIP Cyceron, BP5229, F-14074 Caen, France

b INSERM U360, Hoˆpital Pitie´-Salpeˆtrie`re, 75013 Paris, France

c Unite´ IRM, CHU de Caen, 14000 Caen, France

d Institut Universitaire de France, 75005 Paris, France

Received 16 December 2004; revised 4 February 2005; accepted 24 February 2005

Available online 13 April 2005

Effects of age and sex, and their interaction on the structural brain

anatomy of healthy elderly were assessed thanks to a cross-sectional

study of a cohort of 662 subjects aged from 63 to 75 years T1- and

T2-weighted MRI scans were acquired in each subject and further

processed using a voxel-based approach that was optimized for the

identification of the cerebrospinal fluid (CSF) compartment Analysis

of covariance revealed a classical neuroanatomy sexual dimorphism,

men exhibiting larger gray matter (GM), white matter (WM), and CSF

compartment volumes, together with larger WM and CSF fractions,

whereas women showed larger GM fraction GM and WM were found

to significantly decrease with age, while CSF volume significantly

increased Tissue probability map analysis showed that the highest

rates of GM atrophy in this age range were localized in primary

cortices, the angular and superior parietal gyri, the orbital part of the

prefrontal cortex, and in the hippocampal region There was no

significant interaction between bSexQ and bAgeQ for any of the tissue

volumes, as well as for any of the tissue probability maps These

findings indicate that brain atrophy during the seventh and eighth

decades of life is ubiquitous and proceeds at a rate that is not

modulated by bSexQ.

D 2005 Elsevier Inc All rights reserved.

Keywords: Brain; Aging; Sex; Voxel-based morphometry; MRI

Introduction

The increase of life expectancy during the last century has led

to a growing number of dementia cases in the aging population

Prevalence studies suggested that, in 2000, the number of persons

with Alzheimer’s disease in the United States was 4.5 million and

predicted to rise to 13.2 million by 2050 (Hebert et al., 2003) This

dementia incidence upsurge has reinforced the importance of

characterizing the mechanisms of the human brain aging during the seventh and eighth decades of life Indeed, a better understanding

of the normal neuroanatomical aging could be of high interest for dissociating processes specifically associated with pathologic brain changes from those associated to normal changes

During the past two decades, several studies have investigated the effect of aging on the human brain More often than not, these studies investigated cerebral changes over life span (from 20 up to

80 years) Their findings have led to a large consensus regarding the global morphological changes due to aging First, postmortem studies have described, starting at the fourth decade, a decrease of the brain weight and an increase of the cerebrospinal fluid volume (CSF) (Dekaban, 1978) Then, studies using Magnetic Resonance Imaging (MRI) have confirmed and refined these findings by showing that the gray matter (GM) volume starts to decrease earlier

in the life (at the end of the first decade), whereas the white matter (WM) volume starts to decrease at the fourth decade (Courchesne

et al., 2000; Pfefferbaum et al., 1994)

There seems to exist, however, a large variability in the way the different brain areas are reacting to aging These selective age-related neuroanatomical changes could be explained by several aging theories One of them is based on brain ontogeny and phylogeny and states that the age-related changes of the various cerebral regions follow a time pattern that is the reverse sequence

of their maturation during development (Braak et al., 1999; Raz et al., 1997) According to this model, late maturating unimodal or high-order heteromodal associative cortices are the first and the most age-sensitive, while early maturating primary areas are subject to later and smaller age-related changes In agreement with this model, several studies have specifically focused on associative cortices and have shown a preferential atrophy of the regions belonging to the prefrontal cortex (Coffey et al., 1992; Jernigan et al., 2001; Salat et al., 2001) Other studies have reported focal atrophy localized into the temporal lobe (Bigler et al., 2002) including the hippocampus (Raz et al., 2004b; Tisserand

et al., 2000) However, other aging hypotheses based on the dysfunction of the principal neurotransmitter systems could also explain the affliction of these cerebral regions in healthy elderly

1053-8119/$ - see front matter D 2005 Elsevier Inc All rights reserved.

doi:10.1016/j.neuroimage.2005.02.042

T Corresponding author Groupe d’ Imagerie Neurofonctionnelle

UMR6194, CNRS, CEA, Universite´s de Caen et Paris 5, GIP Cyceron,

BP5229, F-14074 Caen, France Fax: +33 231 470 271.

E-mail address: mazoyer@cyceron.fr (B Mazoyer).

Available online on ScienceDirect (www.sciencedirect.com).

NeuroImage 26 (2005) 900 – 911

subjects Indeed, the age-related decline of dopaminergic (Volkow

et al., 2000) and cholinergic (Podruchny et al., 2003) systems,

which project on the frontal and limbic structures, respectively,

could be associated to this cerebral pattern of atrophy

Meanwhile, using whole brain exploratory approaches, several

other studies were aimed at identifying other potential targets of

normal aging These studies have found an age-related atrophy of

associative cortices but, more surprisingly, an implication of

several primary cortices normally considered as spared by aging

(Good et al., 2001; Sowell et al., 2003; Van Laere and Dierckx,

2001) For example, Salat et al (2004) found that regional

cortical thinning with age (which has been found highly

correlated with regional GM density, Narr et al., in press) is

widespread over large parts of the cortex including motor,

auditory, and visual primary areas, as well as association cortices

such as the inferior lateral prefrontal cortex Interestingly, a few

recent studies have specifically focused on the seventh and

subsequent decades, a period of life where maturation processes

no longer interfere with aging, and have reported a similar pattern

of regional age-related atrophy (Resnick et al., 2003; Tisserand et

al., 2004)

Beside age, sex is another major player of the inter-individual

brain morphology variability and several studies have been

interested in the potential impact of sex on age-related brain

changes As a rule, these studies concluded that men exhibited

larger age-related brain atrophy and CSF increase than women over

the entire life span (Coffey et al., 1998; Gur et al., 1999; Yue et al.,

1997), this effect being enhanced in the frontal and temporal lobes

(Gur et al., 2002; Murphy et al., 1996; Raz et al., 1997, 2004a)

Conversely, reports of regional age-related atrophy higher in

women than in men are rare, although larger reduction of gray

matter in women have been reported in the visual cortex (Raz et al.,

1993), the parietal lobes and the hippocampus (Murphy et al.,

1996)

Actually, as the majority of these studies were based on large

age range cohorts, little is actually known about the effect of sex on

age-related changes in brain structure of healthy elderly subjects In

the present study, we have investigated this issue by taking

advantage of a large epidemiology study dealing with vascular

aging for which a large cohort of subjects in their seventh or eighth

decades were recruited and examined with MRI

Methods

Subjects

The sample of subjects who participated to the present protocol

is a sub-sample of the EVA (Epidemiology of Vascular Aging)

cohort (n = 1389), a longitudinal study on vascular aging and

cognitive decline in healthy elderly subjects, the characteristics of

which have been described elsewhere (Dufouil et al., 2001)

Subjects, born between 1922 and 1932, were recruited from

electoral rolls in Nantes (West of France) from June 1991 to June

1993 All participants gave their written informed consent to the

EVA study protocol, which was approved by the Ethic committee

of the Kremlin-Biceˆtre hospital A number of biological and

sociological parameters were collected from each subject including

age, sex, hypertension, education level (number of schooling

years), and handedness Subject’s global cognitive performances

were assessed using the Mini-Mental State Examination (MMSE)

(Folstein et al., 1985)

At 4-year follow-up, MRI examination was proposed to all subjects and 88% of them agreed to participate This sub-sample did not differ from the rest of the cohort in terms of age, sex ratio, hypertension, and cognitive performances Due to financial limitations, MRI could be performed in 845 subjects only, among whom 32 had to be excluded because of the poor technical quality

of their scans, and 11 others because of previous history of stroke

as confirmed by a neurologist Left with a sample of 802 subjects (471 women, 331 men), we randomly selected 331 women in order

to obtain groups of men and women with identical size Basic demographic statistics are presented inTable 1 At the time of their MRI, the 331 men and 331 women did not differ for age However, the men group had a higher mean level of education, a larger proportion of hypertensive subjects, and a smaller proportion of right-handed subjects, than women ANCOVA reveals no effect of Sex (P = 0.14) or Age (P = 0.29) on the cohort MMSE scores Rather, we found a significant bSex by AgeQ interaction (P = 0.0012), the age-related decrease of MMSE being larger in women than in men

MRI imaging MRI acquisition

MR images were acquired between November 1995 and September 1997, using the same machine (1.0 T Magnetom Expert, Siemens, Erlangen) and a standardized acquisition proto-col Exclusion criteria were conventional: (1) carrying a cardiac pacemaker, valvular prosthesis, or other internal electrical/mag-netic device; (2) history of neurosurgery or aneurysm; (3) presence

of metal fragments in the eyes, brain, or spinal cord; (4) claustrophobia MRI acquisition was performed after the bio-logical/psychological testing

The MRI acquisition which consisted of a three-dimensional (3D) high-resolution T1-weighted brain volume was first acquired using a 3D inversion recovery spoiled-gradient echo sequence (3D IR-SPGR; TR = 97 ms; TE = 4 ms; TI = 300 ms; sagittal acquisition) The 3D volume matrix size was 128
256
256, with a 1.4
0.89
0.89 mm3voxel size T2- and PD- (proton density) weighted brain volumes were also acquired during the same sequence using a 2D axial turbo spin-echo sequence with two echo times (TR = 3500 ms; TE1 = 15 ms; TE2 = 85 ms; 23 cm field of view) T2 and PD acquisitions consisted of 26 contiguous

Table 1 Sample characteristics

[63.77, 75.60]

69.56 (2.95) [63.69, 75.47]

0.87# Education

level (years)

11.3 (3.8) [4, 20] 10.2 (3.0) [4, 20] 0.00051y Hypertensive

subjects (%)

Right-handed subjects (%)

MMSE score (max = 30)

Mean (standard deviation); [range]; MMSE: Mini-mental state examination.

# Student’s t test.

y Wilcoxon rank sum and signed rank test.

z Pearson’s chi-squared test.

5-mm-thick axial slices (13.0 cm axial field of view), having a

256
256 matrix size, and a 0.89
0.89 mm2in-plane resolution

Positioning in the magnet was based on a common landmark for all

subjects, namely, the orbito-meatal line, so that the entire brain,

including cerebellum and mid-brain, was contained within the field

of view of both T1 and T2/PD acquisitions Data sets (T1, T2, and

PD) were readily reconstructed, visually checked for major

artifacts, before further analysis Finally, only the T1- and

T2-weighted images were used in the framework of our study

Image processing

The T1- and T2-weighted images of each subject were first

aligned to each other (Woods et al., 1992) and then analyzed with

SPM99 (http://www.fil.ion.ucl.ac.uk/spm/) We used the so-called

optimized Voxel-Based Morphometry (VBM) protocol (Good et

al., 2001) that we slightly modified in two ways in order to account

for the structural characteristics of aged brains (seeFig 1) First,

GM, WM, and CSF templates specific to our database (EVA

priors) were used for tissue segmentation Second, segmentation of

the CSF class was refined using T2 images

Creating EVA priors

Tissue templates specific to our database (EVA priors) were

created using a sub-sample of 120 randomly selected subjects (60

men and 60 women) matched for age, hypertension frequency,

handedness, and education level with the entire group Each of the

120 subject T1 volumes was segmented using the MNI priors

available in SPM, providing 120 individual GM, WM, and CSF

tissue maps The 120 GM images were then non-linearly spatially

normalized to the GM MNI template (7
8
7 non-linear basis

functions in the three orthogonal directions) The normalization

parameters (deformation fields) obtained from the GM warping

step were then reapplied to the WM and CSF partition images, the

resulting images being further interpolated as 1 mm3 isotropic

voxel volume Individual GM, WM, and CSF image volumes were

further smoothed with an 8-mm full-width at half-maximum

isotropic Gaussian kernel Finally, EVA priors were obtained by

computing GM, WM, and CSF probability maps based on the set

of 120 GM, WM, and CSF partition volumes, respectively

Processing individual images of the 662 subjects cohort

Each subject T1 volume image was first segmented using

MNI priors in order to obtain a GM partition image in his

native space This GM volume was then non-linearly spatially

normalized to the EVA GM template using 7
8
7

non-linear basis functions in the three orthogonal directions

Corresponding normalization parameters (deformation fields)

were reapplied to the subject original brain T1 and T2 images,

the resulting images being further interpolated (1 mm3 isotropic

voxel) The resulting normalized T1 volume was then

segmented using the EVA priors thereby providing GM, WM,

and CSF partition images (see Fig 1, left side)

Optimizing the CSF partition image

Obtaining a good segmentation of the CSF compartment

requires an accurate definition of its borders Accordingly, we

proceeded to a multi-spectral segmentation of both the T1 and T2

volumes, again using the EVA priors An optimized CSF partition

image was obtained by subtracting the GM and WM partition

images provided by the first mono-spectral T1 segmentation from

the sum of the GM, WM, and CSF partition images provided by

this second segmentation (seeFig 1, right side) In summary, the final CSF partition images were derived from a multi-spectral segmentation combining T1 and T2 volumes, while the final GM and WM partition images were derived from the segmentation of the T1 volumes only (see Fig 1for a detailed description of the pipeline procedure) The improvement provided by this modified CSF segmentation scheme was quantified by comparing the absolute CSF and total intracranial volumes (see below for tissue volumes estimation) obtained either without or with T2 image inclusion in the segmentation process

Image modulation Finally, we applied a so-called bmodulationQ to each cerebral partition image, adjusting their voxel intensities for the strength

of the deformation they were submitted to during the spatial

Fig 1 Flow chart of the image processing protocol The blue part is equivalent to the optimized VBM protocol proposed by Good et al (2001) , whereas the red part describes how T2-weighted MR images were incorporated in order to optimize the CSF tissue segmentation MRI: whole brain images in their native space GM, WM, and CSF: gray matter, white matter, and cerebrospinal fluid tissue images, respectively The prefixes bnQ and bmQ denote images in the stereotactic space after normalization and modulation, respectively T1 and T1T2 indices refer to mono-spectral (T1) and multi-spectral (T1 and T2) segmentations, respectively The figure shows four images corresponding to the same axial slice of the same subject: nT1 and nT2 (gray-scaled) are the normalized T1- and T2-weighted images, respectively, whereas nCSF T1 and nCSF Opt (color-scaled) are the CSF tissue images without and with optimization, respectively The skull inner and outer limits were derived

as iso-intensity contours in the normalized T2 image (nT2) and super-imposed on both CSF tissue images.

normalization process (Good et al., 2001) Modulation preserves

the subject’s original tissue quantity after its transfer to the

reference space Finally, all cerebral partition images were

smoothed with a 12-mm full-width at half-maximum isotropic

Gaussian kernel

Volume estimation

For each subject, GM, WM, and CSF volumes were computed

as the integral of the voxel intensities over the corresponding

modulated tissue partition image Total Intracranial Volume (TIV)

was computed as the sum of the GM, WM, and CSF volumes, and

fractional cerebral compartment volumes as the ratios of tissue

absolute volumes to TIV

Statistical analysis

Volumetry

TIV and GM, WM, CSF absolute and fractional volumes were

analyzed using the same ANCOVA design, with bSexQ as the main

factor, bAgeQ as the covariate, including a bSex by AgeQ

interaction Significance level set at P b 0.05 for each tissue

volume analysis Slopes of the linear regressions of cerebral

compartment volumes with age were estimated separately for men

and for women

Tissue partition maps ANCOVA was applied to modulated and smoothed GM, WM, and CSF probability maps as implemented in SPM (Friston et al.,

1995), using two different intensity normalizations: voxels of each tissue partition map were scaled to either TIV value, adjusting for head size, or to absolute cerebral compartment volume, searching for local variations within each cerebral compartment A map-wise threshold of P b 0.05 corrected for multiple comparisons was used for each tissue map analysis

Results

A brain atlas for healthy elderly Fig 2 shows selected slices through the average T1 volume, and the GM, WM, and CSF probability maps computed over the sample of 662 subjects Such maps constitute a probabilistic brain atlas in healthy elderly human subjects aged between 63 and 75 years GM and WM atrophy, and CSF enlargement, are the most prominent features of these maps when compared with their counterparts in young healthy adults As such maps could be of value for others working with anatomical/functional brain images

of aged subjects, they will be made available to the neuroimaging community on the Internet

Evaluation of the optimized CSF tissue segmentation Using a multi-spectral rather than a mono-spectral segmentation led to smaller average volumes both for the CSF (357 F 58 cm3vs

494 F 68 cm3, mean F SD, n = 662) and for TIV (1371 F 132

cm3vs 1515 F 134 cm3) It also gave a larger age-related CSF volume increase (3.6 cm3/year vs 2.3 cm3/year) and a smaller age-related TIV decrease (0.4 cm3/year vs 1.7 cm3/year) This last finding constitutes a clear indication that including T2 images improved the CSF segmentation since one cannot expect TIV to significantly decrease over such a short age range In the subsequent results, we will thus only consider the CSF volume obtained with the multi-spectral segmentation only

Fig 2 Selected slices through the average (n = 662) normalized T1

volumes and corresponding gray matter (GM), white matter (WM), and

cerebrospinal fluid (CSF) probability maps The gray scale applies to GM,

WM, and CSF tissue images and gives the probability for a voxel to belong

to the considered tissue The location of the five axial slices is shown on a

three-dimensional rendering of the average T1 volume (z = 49, 31, 15, 1,

and 17 mm from the bicommissural plane, respectively).

Table 2 Sex and age effects and bSex by AgeQ interaction on absolute cerebral compartment volumes

( P value)

Age effect ( P value)

Sex by Age ( P value)

Slope 0.059 ns

0.28 ns

Mean (standard deviation) of absolute cerebral compartment volumes (in

cm 3 ) in men and women (upper line) and slopes of their regression on age (in cm 3 /year) with their significance levels (lower line): ns: non-significant,

*P b 0.05, **P b 0.001.

The last three columns give the P values of the Sex and Age effects as well

as the bSex by AgeQ interaction of the ANCOVA analysis TIV: total intracranial volume; GM: gray matter; WM: white matter; CSF: cerebro-spinal fluid.

Volumetric data

Results regarding absolute and fractional brain tissue volumes

are shown inTables 2 and 3, respectively As expected, TIV, GM,

WM, and CSF absolute volumes were larger in men than in

women There was no bSex by AgeQ interaction for any of the

absolute cerebral compartment volumes For the 662 subjects, TIV

was found to be unaffected by age, while GM (2.2 cm3/year) and

WM (1.7 cm3/year) volume significantly decreased with age, their

decreases being compensated by an equivalent increase of CSF

volume (3.6 cm3/year) Note, however, that the rate of GM loss

was somewhat smaller in men than in women whereas the rate of

WM loss was identical for both sexes Nevertheless, the GM to

WM volume ratio did not vary with age and stayed higher in

women (1.25) than in men (1.18)

The GM fraction was found higher in women than in men,

whereas both the WM and CSF fractions were higher in men than

in women There was a significant effect of age on all cerebral

compartment fractions, with no bSex by AgeQ interaction for any of

them but, again, the GM fraction decrease was somewhat larger in

women (0.20% per year) than in men (0.12% per year) For WM,

men and women exhibited the same rate of fractional volume

decrease (0.11% per year) The GM and WM fraction losses were

compensated by a rate of CSF fraction increase of 0.23% per year

for men and of 0.32% per year for women

Voxel-based morphometry

Adjusted either by TIV or by cerebral compartment volumes,

the regional regression coefficients with age for the GM, WM, and

CSF compartments were not statistically different between men

and women (P b 0.05 corrected for multiple comparisons) As no

bSex by AgeQ interaction was found in any of the three

compart-ment maps, age-related effects on tissue distribution are presented

for the entire sample of 662 subjects Note that a trend for a larger

(albeit not significant) age effect in women was observed in the

GM and CSF TIV-adjusted maps, similar to what was reported above for cerebral compartment volumes when expressed as TIV fractions However, this trend vanished when the tissue maps were adjusted for tissue volumes rather than for TIV

Age-related changes in tissue probability maps corrected for TIV The age-related variations of GM, WM, and CSF probability maps corrected for TIV are depicted in the Fig 3 The rate of

GM loss was highest in primary cortices, including the Heschl’s gyrus, the cortex surrounding the Calcarine fissure and the pre-and postcentral gyri Rates of GM losses were also very high in the angular and superior parietal gyri, in the orbital part of the prefrontal cortex, and in the hippocampal region By contrast, the rate of GM losses appeared marginal in areas such as the lateral and medial surfaces of the superior frontal gyri, the median cingulate gyrus, and the inferior temporal gyrus Interestingly, we found positive regression slopes with age in the subcortical gray nuclei bordering the third and lateral ventricles, namely, the caudate nuclei, putamen, pallidum, and thalami

For the white matter, the general pattern brought out high WM losses in the corpus callosum and in the major pathways surrounding the lateral ventricles such as the anterior and posterior callosal fibers, the optical tracts, and the posterior limb of the internal capsule By contrast, smaller WM fasciculi, close to the cortical surface, did not show any significant variations with age

Finally, increase of CSF with age was highest in the third and lateral ventricles, and in the interhemispheric and Sylvian fissures Age-related changes in tissue probability maps corrected for absolute cerebral compartment volumes

The effect of age on GM, WM, or CSF maps corrected for their absolute tissue volumes is summarized in Fig 4 and Table 4 Variability of cranial vault was implicitly accounted for in these analyses since each global cerebral compartment volume was highly correlated with TIV (r2= 0.81, 0.91, and 0.77 for GM, WM, and CSF, respectively, P b 0.001 in all three cases) The results show, for each cerebral compartment, the regions in which the age-related rate of local volume variation exceeds that of the global tissue volume Significantly higher reductions of GM with age were found in the Heschl’s, precentral, postcentral, middle frontal (orbital part), and superior parietal gyri, as well as in the hippocampus Meanwhile, the rate of WM losses was significantly higher in the bundle of fibers running alongside the lateral ventricles and in the genu of the corpus callosum By contrast, the increase of CSF was homogeneous over the entire compartment

as no significant regional age-related increase was found in the CSF map of subjects when adjusted for their CSF global volume Discussion

Enhanced CSF compartment using multi-spectral segmentation in the elderly

Including T2 images in the tissue segmentation procedure resulted in a better characterization of the outer border of the CSF compartment and a more realistic CSF probability values in the ventricles and major sulci This was expected since T2 images exhibit a good contrast between the subarachnoidal CSF and the dura mater adhering to the inner skull surface However, the larger

Table 3

Sex and age effects and bSex by AgeQ interaction on fractional cerebral

compartment volumes

( P value)

Age effect ( P value)

Sex by Age ( P value)

GM fraction 0.396

(0.021)

0.414 (0.021)

WM fraction 0.337

(0.016)

0.332 (0.017)

CSF fraction 0.266

(0.027)

0.253 (0.028)

(0.08)

1.25 (0.09)

Mean (standard deviation) of fractional cerebral compartment volumes

(relative to TIV) and of the gray to white matter ratio in men and women

(upper line) and slopes of their regression on age (in %/year) with their

significance levels (lower line): ns: non-significant, *P b 0.01, **P b 0.001.

The last three columns give the P values of the Sex and Age effects as well

as the bSex by AgeQ interaction of the ANCOVA analysis GM: gray matter;

WM: white matter; CSF: cerebrospinal fluid.

slice thickness of the original T2 images (5 mm) as compared to

the original T1 images (1.4 mm) induced an important partial

volume effect, which affected the quality of the multi-spectral

segmentation For this reason, multi-spectral segmentation was

only used to classify the voxels belonging to the CSF

compart-ments, while the GM and WM compartments were obtained with a

mono-spectral segmentation of T1 images Note that the CSF

volumes so estimated are consistent both with another in vivo

study that also used a multi-spectral segmentation (Courchesne et

al., 2000) and with postmortem data (Blinkov and Glezer, 1968)

Actually, mono-spectral segmentation leads to an underestimation

of the CSF volume in the oldest subjects (i.e., those who present

the largest atrophy) Consequently, when estimated using a

mono-spectral segmentation, TIV appears to decrease with age in the

elderly while it stays roughly constant when estimated with a

multi-spectral segmentation Note that a previous study using the

same optimized VBM approach and T1-weighted image

segmen-tation only, also reported a linear decline of TIV with age for men

but not for women (Good et al., 2001) As the age of the subjects of

this latter study spread over seven decades, these authors

interpreted the TIV decrease as a secular trend of increasing

cranial vault over the last century Obviously, such an explanation

does not hold for our findings since they were observed over a

single decade (cranial perimeter and height of our subjects did not vary with age) The fact that TIV decrease with age could be corrected by including T2-weighted images in the segmentation leads us to conclude that it was an artifact of the mono-spectral segmentation

Age effects in cross-sectional versus longitudinal studies Before discussing our results in details, it is also worthwhile discussing the intrinsic limitations of cross-sectional studies, such

as ours, where age effects on neuroanatomy are measured at a single time across a sample of subjects having different ages The limited age range of our cohort does limit potential secular effects

on brain volumes that could severely bias cross-sectional studies performed on the entire span of life (such as the increase in the height, and as a result, the TIV, of subjects born between 1920 and

1990, for example) A short age range does not, however, reduce the between-subject variability and statistical power loss that characterize cross-sectional studies and make longitudinal studies preferable Conversely, very large samples are more manageable in cross-sectional than in longitudinal studies, which can compensate the statistical power difference between the two designs For instance,Davatzikos and Resnick (2002)found that age effects on

Fig 3 Age-related gray matter, white matter, and cerebrospinal fluid volume regression maps (after correction for total intracranial volume) Regression maps are superimposed onto their corresponding tissue probability maps and displayed without statistical threshold The hot (green to red) and cold (green to blue) color scales represent the negative and positive slopes with age, respectively The location of the axial slices is shown on a three-dimensional rendering of the average T1 volume [z = 59, 49, 39, 31, 23, 15, 7, 0, 9, and 17 mm from the bi-commissural plane (pink box), respectively] L: Left; R: Right.

white matter connectivity in elderly were significant both in

cross-sectional and longitudinal studies, but that longitudinal findings

were more pronounced than cross-sectional ones Amazingly, the

same authors performed a longitudinal study of 116 healthy elderly

subjects aged from 59 to 85 years, and did not find any detectable

changes in global or regional brain volumes over 1 year, while they

found rates of tissue loss of 1.4 cm3/year and 1.9 cm3/year for the

GM and WM, respectively, using a cross-sectional analysis on the

same sample (Resnick et al., 2000) These authors invoked, here,

the limits of their image processing accuracy when only subtle

cerebral changes are expected over a short period of time Note,

however, that very short longitudinal investigation can be sufficient

to highlight neuroanatomical differences in pathological processes

such Alzheimer’s disease (Fox et al., 2001) Interestingly,

re-analyzing 92 subjects among their initial 116 ones over a 4-year

period,Resnick et al (2003)found a 71% and 63% increase of the

GM and WM rate of atrophy as compared to the rates they

estimated in their previous cross-sectional analysis, showing that

when a larger period of time (3 to 4 years) separates two MRI

examinations of a longitudinal study, higher age-related effects on

brain atrophy rates are found in longitudinal analysis as compared

to cross-sectional ones

Global age-related cerebral volume changes in healthy elderly

We observed a loss of 3.9 cm3/year of brain tissue (GM plus

WM), in agreement with previous studies dealing with elderly

subjects (Liu et al., 2003; Resnick et al., 2000, 2003) In fact, the

latter studies reported a loss of 4.4 cm3/year on average (range

from 3.2 to 5.4 cm3/year), a value very close to ours However, the

rate of brain tissue loss we found was somewhat different from that

of studies based over the entire life span Postmortem studies have reported an age-related decrease of brain volume close to 2 cm3/ year between the third and eighth decades (Dekaban, 1978; Pakkenberg and Gundersen, 1997) In addition, the average of atrophy rates reported by MRI studies performed over the entire life span sets at 2.5 cm3/year (range from 1.5 to 4.2 cm3/year) (Blatter et al., 1995; Good et al., 2001; Guttmann et al., 1998; Jernigan et al., 2001; Liu et al., 2003; Van Laere and Dierckx,

2001) Actually, according to some authors, the GM volume linearly decreases starting from the second decade, whereas the

Fig 4 Areas of age-related reductions in gray matter and white matter after correction for global tissue volume Student’s t maps are superimposed onto their corresponding tissue probability maps and displayed at a P b 0.05 significance level corrected for multiple comparisons The x and z coordinates (in mm) give the slice locations in the stereotactic space L: left; R: right.

Table 4 Regional gray matter reduction with age Anatomical

label

L Middle frontal gyrus, orbital part

R Middle frontal gyrus, orbital part

L Superior parietal gyrus

t value: Student’s t value (P b 0.05 corrected for multiple comparisons); x y z: MNI space stereotactic coordinates in mm; L: left; R: right.

WM volume increases until the fourth decade and, then, decreases

in the following decades (Courchesne et al., 2000; Jernigan et al.,

2001) Thus, one should expect the annual rate of brain tissue loss

to increase in elderly Our findings are consistent with this

hypothesis and confirm that brain shrinkage is a non-linear

phenomenon over the life span that accelerates after the sixth

decade

We found that GM and WM almost equally contributed to brain

shrinkage, no significant difference being observed between the

annual atrophy rates of these two brain compartments (P = 0.58)

This is in agreement with the findings of two previous studies in

elderly (Resnick et al., 2000, 2003), and with those of another

study dealing with a larger age range sample (Good et al., 2001)

However, the regression slopes we found for GM (2.2 cm3/year)

and WM (1.7 cm3/year) do not fit with the supposed larger WM

loss rate proposed by other authors (Guttmann et al., 1998; Liu et

al., 2003) Difference in study designs (i.e., cross-sectional vs

longitudinal) is an unlikely explanation given the short age range

of the samples of the Liu study and of ours Rather, even though

we could not find in the two above reports whether or not the

atrophy rates of GM and WM were significantly different (both

reports state that the rate of atrophy is significant for WM only), we

believe that the use of different segmentation procedure could be at

the origin of these discrepant findings First, note thatGuttmann et

al (1998) used T2- and PD-weighted images only for the

segmentation step which renders the GM/WM limit hard to define

Second, in elderly subjects aged from 57 to 77 years, Liu et al

(2003)reported an annual loss of brain tissue (GM plus WM) that

did not match the corresponding annual increase of CSF in the

same sample, the unexplained 1.4 cm3/year difference being

possibly the consequence of an inaccurate tissue segmentation It

seems thus reasonable to assume that GM and WM contributions to

brain shrinkage are similar during the seventh and eighth decades,

but additional studies focusing on the following decades are

needed to check whether this holds later in life

Voxel-wise age-related changes in healthy elderly

The regional distribution of age-related reduction of GM

volume was found to be very heterogeneous, some areas seeming

particularly vulnerable, others being relatively spared

Interest-ingly, the largest rates of atrophy were found in the primary

auditory, somatosensory, and motor cortices (seeFig 4) Highly

negative regression slopes of GM density with age were also

observed in the primary visual cortex but failed to reach

significance after adjustment for the global GM rate of atrophy

We believe this lack of significance to be the consequence of

higher residual standard errors of the regression slope estimated in

this region (about twofold the average residual standard error

computed over the whole GM map as indicated by analysis of the

residual variance image) This is likely to be due to the high

residual anatomical variance given both the large spatial variability

of the Calcarine fissure (Thompson et al., 1996) and the relative

small cortical thickness (Von Economo, 1929) observed in the

primary visual cortex as compared to other regions (see also the

GM probability map inFig 2) Thus, notwithstanding the lack of

significant findings, we believe that the primary visual cortex

should be considered as a focus of age-related GM reduction, as

well as others primary cortices

More generally, it should be stressed that VBM findings are

influenced by the amount of residual anatomical variability between

subjects after spatial normalization (Crivello et al., 2002; Good et al., 2001) since this procedure does not perfectly align cerebral structures between subjects However, we believe this bias source to have a weak impact on our findings First, the smoothing applied to our images (FWHM = 12 mm) dramatically reduces the inter-individual misalignment of cerebral structures after spatial normal-ization Second, the very large number of subjects included in our study, as opposed to studies performed on relative small samples, acts as a supplementary image smoothing process, compensating in part the anatomical residual variability As a matter of fact, inspection of the residual variance image, that partly reflects the spatial distribution of the inter-individual anatomical variability, revealed that the occipital cortex was the only region presenting a high negative regression coefficient associated with a high residual variance Meanwhile, the same image also revealed that many associative regions presented small residual variances, a pattern also shared by the primary cortices (except the primary visual one) These findings allow to refute the idea that the strong age effect found on primary cortices could be explained by a weaker inter-individual anatomical variability in these regions

Note that primary cortices have been previously reported as spared by the aging processes (Jernigan et al., 2001; Raz et al.,

1997), as predicted with the classical blast in, first outQ brain area aging theory (Raz, 2001) Actually, a close look at the most recent literature reveals that several studies, using a voxel-based approach similar to ours, have mentioned primary cortices as the seat of large rates of atrophy (Good et al., 2001; Resnick et al., 2003; Salat et al., 2004; Tisserand et al., 2004) Concerning the age-related decline in perisylvian regions such as insula and Heschl’s gyrus, Tisserand et al (2004) suggested that cerebral regions with complex anatomical boundaries for manual tracing have been largely ignored in aging studies using classical ROI approach This assumption may partially explain why we found in the present study some new cerebral regions vulnerable to aging Such converging results require reconsidering the status of the primary cortices in normal aging One could postulate that, whereas the associative cortices are particularly affected in pathological aging such as Alzheimer’s disease, the same associative cortices would

be distinctly less affected and primary cortices more vulnerable in normal aging This hypothesis is consistent with reports of cognitive decline of the lowest echelons of sensory and motor systems in healthy elderly subjects (Kaye et al., 1994) Moreover, several studies have shown that, in absence of peripheral sensor age-related changes, hearing loss, visual decline, as well as motor slowness during aging could be associated to an affliction of their respective primary cortices (Mendelson and Ricketts, 2001; Schmolesky et al., 2000; Yordanova et al., 2004)

The other areas where preferential age-related GM reduction was observed, included the hippocampus and the orbital part of the middle frontal gyri and are more classically found in studies dealing with normal and/or pathological brain aging (Petersen et al., 2000; Salat et al., 2001)

The prefrontal cortex is usually considered as the structure most affected during normal aging, all age ranges taken together (Jernigan et al., 2001; Raz et al., 1997), and therefore is a key region of the frontal aging theory (relating that the major part of cognitive aging is related to a structural deficit of the prefrontal cortex,West, 1996) In recent whole brain exploratory studies, GM reduction with age was also found in the left middle frontal gyrus (Good et al., 2001), the orbital and inferior frontal cortex (Resnick

et al., 2003), the frontal pole and dorsolateral prefrontal cortex

(Tisserand et al., 2004), or the inferior lateral prefrontal cortex

(Salat et al., 2004) Therefore, taking into account nomenclature

differences, the orbital part of the middle frontal gyrus appears to be

a preferential target for age-related decrease of GM in healthy

elderly Note that this area has been reported in functional

neuroimaging studies as mainly involved in maintaining

informa-tion in working memory (seeTisserand and Jolles, 2003for review)

In this context, increased atrophy rates in this area in healthy elderly

may constitute an early neural correlate of future diminished

performances in executive functions Nevertheless, because of the

importance of the prefrontal cortex in cognitive aging, future

imaging studies are clearly needed to better differentiate the specific

functions of the different frontal regions in relation to aging

Regarding the hippocampus, although it is a key target of

age-related memory changes, previous studies have experienced

difficulty to demonstrate significant hippocampal atrophy with

age in absence of Alzheimer’s disease (Jack et al., 2002)

Interestingly, Raz et al (2004a) recently showed a non-linear

relationship between the hippocampus volume and age, the rate of

atrophy in this region being small until the sixth decade, while

larger atrophy rate occurs afterwards This model fits with our

findings, observed in a sample of subjects aged between 63 and 75

years, as well as with those of two other studies dealing with

subjects over 50 years (Resnick et al., 2003; Tisserand et al., 2004)

The biological mechanisms driving the differential age

vulner-ability of the various cortical regions remain unclear Age-related

impairment of specific neurotransmitter systems, such as the

dopaminergic or cholinergic systems (Kaasinen and Rinne, 2002;

Mesulam, 1995), may be put forward As a matter of fact, key

structures of these two systems (the substantia nigra and the nucleus

basalis of Meynert, respectively) show a loss of dopaminergic/

cholinergic neurons with age (Rehman and Masson, 2001) This

could in turn trigger atrophy in the cortical structures on which these

subcortical nuclei mainly project, such as the prefrontal cortex and

the hippocampus (Goldman-Rakic and Brown, 1981; Volkow et al.,

2000; Wenk et al., 1989) However, further investigations are

clearly needed to determine the exact link between regional atrophy

and the impairment of the neurotransmitter systems

Surprisingly, regional GM analysis also revealed some foci of

age-related increase which were localized bilaterally in the caudate,

putamen and pallidum, and thalami, a phenomenon previously

reported by others (Good et al., 2001) Although these areas may

be less affected than others by aging, we agree with others that they

must also be the seat of a normal age-related shrinkage (

Gunning-Dixon et al., 1998) Thus, we believe that what we observed in

these areas could be an artifact due to the presence of particular

GM/CSF and GM/WM interfaces First, the age-related ventricle

enlargement due to brain atrophy could lead to a displacement of

adjacent gray nuclei simulating an artificial increase of GM with

age in voxel-based approaches Secondly, the volume left by the

loss of myelin in the WM fibers of the internal capsule (Abe et al.,

2002) could be replaced by putamen and pallidum neuron cell

bodies, producing an apparent spatial expansion of GM

Alter-nately,Ylikoski et al (1995) reported in healthy elderly an

age-related increase of white matter hyperintensities (WMH) in the

periventricular areas This type of lesion, observed with a

hyposignal in T1-weighted images, could be potentially

misclassi-fied as GM and imitate an increase of GM with age This remark is

all the more right as several subjects were hypertensive and as

hypertension has been significantly associated with an increased

severity of WMH in our cohort of subjects (Dufouil et al., 2001)

Finally, as opposed to what was found for the GM, there were only few areas of accelerated WM atrophy with age after removal

of the global age-related WM volume reduction In fact, accelerated WM atrophy rates were observed almost exclusively

in the corpus callosum, in agreement with the findings of a previous study in healthy subjects aged between 70 and 82 years (Sullivan et al., 2002) Such age-related WM reduction could be attributed to the micro-structural deterioration of the WM identified

in diffusion imaging studies (Pfefferbaum et al., 2000), which was interpreted as a demyelination of WM fibers during aging ( Meier-Ruge et al., 1992) Otherwise, the ventricular enlargement in aging could determine partly the age-related changes in WM fibers surrounding ventricles by a simple mechanical force (Peterson et al., 2001)

Global versus voxel-wise age-related brain changes The results obtained in the TIV-adjusted VBM analysis show, at the voxel level, the same age-related trends that those observed at the cerebral volumetric level Such concordance is explained by the fact that the TIV-adjusted VBM analysis did not take into account the age effect on the cerebral volumes Therefore, the age-related changes estimated in the fractional cerebral volumes reflect the global outcome of all age-related variations identified at the voxel level By contrast, adjusting VBM analysis for absolute cerebral volumes rather than TIV provided quasi-identical age-related regression maps of GM, WM, and CSF compartments between men and women This means that the regional pattern of age-related changes were similar in men and women for each tissue taken separately More generally, the age effects on global cerebral volumes and on tissue maps do not necessarily match since VBM findings are highly dependent on the kind of adjustment used (TIV

or cerebral volumes for instance) Thus, several scenarios can be envisaged On the one hand, if a VBM analysis is not adjusted for a global effect, this global effect naturally spreads over regionally, and as a consequence, the volumetric and VBM findings are well related On the other hand, if the global effect is modeled and adjusted for in a VBM analysis, regional changes due to this effect (i.e., regional changes greater that the global one) could be highlighted or not, leading to related or discrepant findings between volumetric and VBM findings

Sex effect on structural brain aging The neuroanatomical sexual dimorphism we observed in healthy elderly is in close agreement with previous observation

in younger adults (Gur et al., 1999) In addition, we did not find any significant bSex by AgeQ interaction either on global cerebral compartment volumes (either absolute or fractional) or in tissue probability maps, although a trend for larger rate of GM loss and CSF increase was present in women (associated with a larger age-related decline of MMSE score in women) These findings are in contradiction with the common idea that men brains are more vulnerable to aging (Coffey et al., 1998) In a sample of elderly aged from 66 to 96 years, these authors reported an increase of sulcal CSF volume in men only Taking a sub-sample of subjects aged from 65 to 75 years, the same authors highlighted an annual rate of sulcal CSF increase for men and women of 2.1 and 0.06

cm3/year, respectively By contrast, we estimated an annual rate of CSF increase (including sulcal and ventricular CSF compartments) for men and women of 3.3 and 4.0 cm3/year, respectively

A possible explanation of this discrepancy could come from

differences in hypertensive subject proportion or education level

between men and women in our cohort Indeed, some studies have

reported the effect of these two factors on the neuroanatomical

aging For example, concerning the hypertension,Strassburger et

al (1997) reported a greater cerebral atrophy in occipital and

temporal regions for hypertensive elderly subjects as compared to

normotensive elderly subjects Concerning the education level,

Coffey et al (1999)highlighted a positive correlation between the

number of years of education and the peripheral CSF volume in

healthy elderly subjects However, including these variables, as

confounding factors in the analysis, did not modify our results One

could also raise the issue of using a common normalization

template, including both men and women, with the possible

ensuing bias of reproducing a similar atrophy scheme in men and

women However, using a specific template for each sex did not

significantly modify our results

Rather,Coffey et al (1998)also reported no sex effect on brain

atrophy on the same sample, what seems contradictory with their

findings concerning the CSF and may indicate a problem in

volume estimation that could possibly originate from the manual

tissue segmentation performed in this study

As other recent studies (Resnick et al., 2000, 2003), based on

automated image segmentation rather than manual tracing, also

reported no bSex by AgeQ interaction in healthy elderly, one is led to

admit that in their seventh and eighth decades, men brain are not

more, if not less, vulnerable to aging than that of women Arguments

in favor of this hypothesis may be found in several studies of white

matter lesions that have shown a larger prevalence of this type of

lesions in women compared to men aged over 60 years (Sijens et al.,

2001; Wen and Sachdev, 2004), which may be due to a larger

age-related decrease of the brain choline level in women (Sijens et al.,

2003) The drastic changes in circulating hormone concentrations

due to menopause in women around age 50 years could be one cause

of such phenomenon (Lamberts, 2002; Raz et al., 2004c), but this

assertion requires further investigations to be validated

Conclusion

Modifications of brain anatomy in the seventh and eighth

decades appear to be characterized by (1) a shrinkage due to

approximate equal loss of gray and white matter, (2) an

inhomogeneous cortical pattern of atrophy rates, larger rates being

observed in primary cortices as well as in associative and limbic

areas These modifications seem to be sex independent

Acknowledgments

This study has been conducted within the framework of the

ICBM project (http://www.loni.ucla.edu/ICBM/) The authors are

grateful to N Tzourio-Mazoyer for her thoughtful comments on

the manuscript H Lemaıˆtre and B Grassiot are supported by

grants from the Commissariat a` l’Energie Atomique and the

Basse-Normandie Regional Council

References

Abe, O., Aoki, S., Hayashi, N., Yamada, H., Kunimatsu, A., Mori, H.,

Yoshikawa, T., Okubo, T., Ohtomo, K., 2002 Normal aging in the

central nervous system: quantitative MR diffusion-tensor analysis.

Neurobiol Aging 23, 433 – 441.

Bigler, E.D., Anderson, C.V., Blatter, D.D., Andersob, C.V., 2002 Temporal lobe morphology in normal aging and traumatic brain injury.

Am J Neuroradiol 23, 255 – 266.

Blatter, D.D., Bigler, E.D., Gale, S.D., Johnson, S.C., Anderson, C.V., Burnett, B.M., Parker, N., Kurth, S., Horn, S.D., 1995 Quantitative volumetric analysis of brain MR: normative database spanning 5 decades of life Am J Neuroradiol 16, 241 – 251.

Blinkov, S.M., Glezer, I.I., 1968 The Human Brain in Figures and Tables,

A Quantitative Handbook Plenum Press, New York.

Braak, E., Griffing, K., Arai, K., Bohl, J., Bratzke, H., Braak, H., 1999 Neuropathology of Alzheimer’s disease: what is new since A Alzheimer? Eur Arch Psychiatry Clin Neurosci 249 (Suppl 3),

14 – 22.

Coffey, C.E., Wilkinson, W.E., Parashos, I.A., Soady, S.A., Sullivan, R.J., Patterson, L.J., Figiel, G.S., Webb, M.C., Spritzer, C.E., Djang, W.T.,

1992 Quantitative cerebral anatomy of the aging human brain: a cross-sectional study using magnetic resonance imaging Neurology

42, 527 – 536.

Coffey, C.E., Lucke, J.F., Saxton, J.A., Ratcliff, G., Unitas, L.J., Billig, B., Bryan, R.N., 1998 Sex differences in brain aging: a quantitative magnetic resonance imaging study (published erratum appears in Arch Neurol 1998 May;55(5):627) Arch Neurol 55, 169 – 179.

Coffey, C.E., Saxton, J.A., Ratcliff, G., Bryan, R.N., Lucke, J.F., 1999 Relation of education to brain size in normal aging: implications for the reserve hypothesis Neurology 53, 189 – 196.

Courchesne, E., Chisum, H.J., Townsend, J., Cowles, A., Covington, J., Egaas, B., Harwood, M., Hinds, S., Press, G.A., 2000 Normal brain development and aging: quantitative analysis at in vivo MR imaging in healthy volunteers Radiology 216, 672 – 682.

Crivello, F., Schormann, T., Tzourio-Mazoyer, N., Roland, P.E., Zilles, K., Mazoyer, B.M., 2002 Comparison of spatial normalization procedures and their impact on functional maps Hum Brain Mapp 16, 228 – 250 Davatzikos, C., Resnick, S.M., 2002 Degenerative age changes in white matter connectivity visualized in vivo using magnetic resonance imaging Cereb Cortex 12, 767 – 771.

Dekaban, A.S., 1978 Changes in brain weights during the span of human life: relation of brain weights to body heights and body weights Ann Neurol 4, 345 – 356.

Dufouil, C., Kersaint-Gilly, A., Besancon, V., Levy, C., Auffray, E., Brunnereau, L., Alperovitch, A., Tzourio, C., 2001 Longitudinal study

of blood pressure and white matter hyperintensities: the EVA MRI cohort Neurology 56, 921 – 926.

Folstein, M., Anthony, J.C., Parhad, I., Duffy, B., Gruenberg, E.M., 1985 The meaning of cognitive impairment in the elderly J Am Geriatr Soc 33, 228 – 235.

Fox, N.C., Crum, W.R., Scahill, R.I., Stevens, J.M., Janssen, J.C., Rossor, M.N., 2001 Imaging of onset and progression of Alzheimer’s disease with voxel-compression mapping of serial magnetic resonance images Lancet 358, 201 – 205.

Friston, K.J., Holmes, A.P., Worsley, K.J., Poline, J.B., Frith, C.D., Frackowiak, R.S.J., 1995 Statistical parametric maps in functional imaging: a general approach Hum Brain Mapp 2, 189 – 210 Goldman-Rakic, P.S., Brown, R.M., 1981 Regional changes of mono-amines in cerebral cortex and subcortical structures of aging rhesus monkeys Neuroscience 6, 177 – 187.

Good, C.D., Johnsrude, I.S., Ashburner, J., Henson, R.N., Friston, K.J., Frackowiak, R.S., 2001 A voxel-based morphometric study of ageing

in 465 normal adult human brains NeuroImage 14, 21 – 36.

Gunning-Dixon, F.M., Head, D., McQuain, J., Acker, J.D., Raz, N., 1998 Differential aging of the human striatum: a prospective MR imaging study Am J Neuroradiol 19, 1501 – 1507.

Gur, R.C., Turetsky, B.I., Matsui, M., Yan, M., Bilker, W., Hughett, P., Gur, R.E., 1999 Sex differences in brain gray and white matter in healthy young adults: correlations with cognitive performance J Neurosci 19,

4065 – 4072.

Gur, R.C., Gunning-Dixon, F.M., Turetsky, B.I., Bilker, W.B., Gur, R.E.,

2002 Brain region and sex differences in age association with brain