A

aa(ia1nt©

Kd

1

gpd2tsc(

WT

0d

Neurobiology of Aging 31 (2010) 1077–1088

Subregional hippocampal atrophy predicts Alzheimer’s dementiain the cognitively normal

Liana G. Apostolova a,b,∗, Lisa Mosconi c, Paul M. Thompson a,b, Amity E. Green a,b,Kristy S. Hwang a, Anthony Ramirez a, Rachel Mistur c, Wai H. Tsui c,d, Mony J. de Leon c,d

a Department of Neurology, David Geffen School of Medicine, UCLA, CA, United Statesb Laboratory of NeuroImaging, David Geffen School of Medicine, UCLA, CA, United States

c Center for Brain Health, Department of Psychiatry, New York University School of Medicine, NY, United Statesd Nathan Kline Institute, Orangeburg, NY, United States

Received 3 April 2008; received in revised form 29 June 2008; accepted 8 August 2008Available online 24 September 2008

bstract

Atrophic changes of the hippocampus are typically regarded as an early sign of Alzheimer’s dementia (AD). Using the radial distancetrophy mapping approach, we compared the longitudinal MRI data of 10 cognitively normal elderly subjects who remained normal at 3-yearnd 6-year follow-up (NL-NL) and 7 cognitively normal elderly subjects who were diagnosed with mild cognitive impairment (MCI) 2.8range 2.0–3.9) and with AD 6.8 years (range 6.1–8.2) after baseline (NL-MCIAD). 3D statistical maps revealed greater hippocampal atrophyn the NL-MCIAD relative to the NL-NL group at baseline (left p = 0.05; right p = 0.06) corresponding to 10–15% CA1, and 10–25% subiculartrophy, and bilateral differences at 3-year follow-up (left p = 0.001, right p < 0.02) corresponding to 10–30% subicular, 10–20% CA1, and0–20% newly developed CA2-3 atrophy. This preliminary study suggests that excess CA1 and subicular atrophy is present in cognitively

ormal individuals predestined to decline to amnestic MCI, while progressive involvement of the CA1 and subiculum, and atrophy spreadingo the CA2-3 subfield in amnestic MCI, suggests future diagnosis of AD.

2008 Elsevier Inc. All rights reserved.

eywords: Alzheimer’s disease (AD); Mild cognitive impairment (MCI); Magnetic resonance imaging (MRI); Memory; Hippocampus; Predicting cognitive

hopp(tde

ecline

. Introduction

Alzheimer’s disease (AD) is the most common neurode-enerative disorder. In 2003, there were 28 million ADatients worldwide, leading to an estimated 156 billion USollars in direct and indirect healthcare costs (Wimo et al.,006). Given the increased life expectancy and the graying ofhe baby boomer generation, we will soon be faced with sub-

tantially increased numbers of elderly persons with cognitiveomplaints in need of evaluation and anti-dementia treatmentHebert et al., 2003). Recently the AD research spotlight

∗ Corresponding author at: Alzheimer’s Disease Research Center, 10911eyburn Avenue, 2nd floor, Los Angeles, CA 90095, United States.

el.: +1 310 794 2551; fax: +1 310 794 3148.E-mail address: lapostolova@mednet.ucla.edu (L.G. Apostolova).

ssceamaa

197-4580/$ – see front matter © 2008 Elsevier Inc. All rights reserved.oi:10.1016/j.neurobiolaging.2008.08.008

as turned to the pre-symptomatic (also called latent) stagef AD, characterized by silent yet relentless spread of ADathology through the brain. Reports of profound AD-typeathologic changes in the brains of cognitively normal elderlyHaroutunian et al., 1998; Price and Morris, 1999) emphasizehe pressing need for a robust and reliable pre-symptomaticiagnostic marker. In addition to cerebrospinal fluid mark-rs, noninvasive or mildly invasive neuroimaging techniquesuch as magnetic resonance imaging (MRI), positron emis-ion tomography (PET) and amyloid imaging have receivedonsiderable attention (Dubois et al., 2007). Some mark-rs of hippocampal damage have been shown to be useful

s predictors of future conversion of mild cognitive impair-ent (MCI) to AD (Apostolova et al., 2006b; de Leon et

l., 1993; Mosconi, 2005; Mosconi et al., 2007; Silvermannd Thompson, 2006). However, the potential usefulness of

1 iology o

hnM

brimocstrp(arrAaadgsfhit

p3nrm3dajdWsiC

2

2

oakt2y

SyajawUiiapamfie(eofapasaarcdelI

2

aseucwarpsbsrtoy

078 L.G. Apostolova et al. / Neurob

ippocampal measures for predicting cognitive decline fromormal aging to MCI is emerging (de Leon et al., 2001;osconi et al., 2007; Rusinek et al., 2003).The hippocampus, one of the first brain regions affected

y AD pathology, has been extensively studied with neu-oimaging. The most commonly used technique for analyzingmages of the hippocampus is the region of interest (ROI)

ethod. It has been applied to most neurodegenerative dis-rders and especially to AD. ROI studies rely on a simpleomparison of the average hippocampal volume betweentudy groups. Using the ROI method, Jack et al. reportedhe annual hippocampal atrophy rate in normal elderly whoemain cognitively intact to be 1.4% per year, but 3.3%er year in normal elderly who decline to MCI or ADJack et al., 2004). Newer analysis techniques can modelnd detect regional hippocampal changes in 3D; these haveecently been used to study normal aging and AD and haveeported subtle yet statistically significant shape changes inD (Frisoni et al., 2006; Thompson et al., 2004; Wang et

l., 2003), and MCI (Apostolova et al., 2006a,b; Becker etl., 2006). Some of these studies (Apostolova et al., 2006a,b)emonstrated a subregion-specific hippocampal atrophy pro-ression that corresponds to the reported neuropathologicalpread of neurofibrillary tangles through the hippocampalormation (Bobinski et al., 1998, 2000). Even so, no studiesave yet examined the subregional hippocampal changes dur-ng decline from normal cognition to MCI and subsequentlyo AD.

In the present study we apply a method known as hip-ocampal radial distance mapping to compare baseline and-year follow-up 3D hippocampal maps of 17 cognitivelyormal subjects, retrospectively divided into those whoemained cognitively stable on all annual cognitive assess-ents and those who carried a diagnosis of MCI at their

-year follow-up. All subjects from the second group metiagnostic criteria for probable AD on average 6.75 yearsfter their initial evaluation (range 6–8.2 years) while all sub-ects from the first group continued to be cognitively normaluring an average follow-up of 9.2 years (range 6–13.6 years).e hypothesized that subjects who declined to MCI and sub-

equently AD (NL-MCIAD) would show atrophic changesn hippocampal regions corresponding to the subiculum andA1 subfield as compared to the non-decliners (NL-NL).

. Methods

.1. Subjects

We retrospectively selected 17 subjects from a larger poolf participants, all high-functioning healthy elderly individu-ls at the time of the first examination, of whom 7 (41%) were

nown to have declined to MCI within 3(±1) years (meanime to MCI diagnosis 2.8 years; S.D. = 0.8 years; range.0–3.9 years) and received a diagnosis of probable AD 6ears or longer after their baseline evaluation (mean 6.8 years;

tDA1

f Aging 31 (2010) 1077–1088

.D. = 1.0 year; range 6.1–8.2 years). All subjects receivedearly cognitive and neuropsychological examinations. Theyll had a baseline and a 3-year structural MRI scan. Our sub-ects were community-residing volunteers who responded todvertisements or were spouses and/or caregivers of patientsith dementia who were being evaluated at the New Yorkniversity (NYU) AD Research Center. All subjects signed

nformed consent for research participation and were admin-stered an extensive battery of screening and diagnostic testst baseline and follow-up inclusive of medical, neurological,sychiatric, and family history assessments, MR imaging,nd a comprehensive battery of neuropsychometric tests. Theeasures used to assess cognition and memory included the

ollowing: the Global Deterioration Scale (GDS), with scor-ng based on results of extensive clinical interviews (Reisbergt al., 1988); the Mini-Mental State Examination (MMSE)Folstein et al., 1975); and the Guild Memory Test (Gilbertt al., 1968), which included immediate and delayed recallf paragraphs, verbal and visual paired associates, digits-orward-and-backward testing, digit symbol substitution, andvocabulary test. Also used were tests of recall of a shop-

ing list, visual recognition memory, confrontational naming,nd delayed spatial recall (Flicker et al., 1993). The baselineelection criteria were a GDS score of 1 or 2, MMSE ≥ 28,ge of 65 years or older, no contraindications to MR imaging,bsence of gross brain abnormalities, and no evidence of neu-ologic, medical, or psychiatric conditions that could affectognition. At baseline, all subjects were free of clinicallyetected cognitive impairment in memory, concentration, ori-ntation, language, executive function, and activities of dailyife. The analyses reported here were approved by the NYUnstitutional Review Board.

.2. Cognitive outcomes

The subjects received annual cognitive assessments fort least 6 years (range 6.0–13.6 years) to ascertain whichubjects remained cognitively healthy and which subjectsxperienced cognitive decline. Decline to MCI was definedsing guidelines similar to those advocated by Petersen andolleagues at the Mayo Clinic (Petersen et al., 2001). MCIas defined by the following: (a) a GDS score of 3, indicatingdecline in functioning (including memory complaints) cor-

oborated by an informant and determined by the examininghysician; (b) objective memory impairments, as demon-trated by a neuropsychologic memory test score 1.5 S.D.elow that of healthy elderly individuals of matching age,ex, and education; and (c) the subject not meeting crite-ia for dementia. The statistical distribution of the memoryest scores required in point b was derived from a databasef 282 longitudinal observations of healthy subjects 60–87ears of age. The diagnosis of AD was rendered according

o the National Institute of Neurologic and Communicativeisorders and Stroke and the AD and Related Disordersssociation (NINCDS-ADRDA) criteria (McKhann et al.,984).

iology o

2

3nM(a(rast1tee

dasp(α

iwst(wlhdmawtg

Afe

htdntTopcttarp

pfipiAeatvrat

3

TD

AGEATTBFBFBFBF

P

L.G. Apostolova et al. / Neurob

.3. Imaging data acquisition and analysis

Two MR examinations were performed in all subjects(±1) years apart, and within 2 months of the clinical andeuropsychological exams. Three-dimensional T1-weightedR imaging was performed with a 1.5-T MR imaging unit

Signa; GE Medical Systems, Milwaukee, WI) by usingspoiled gradient-recalled acquisition in the steady state

SPGR) sequence with the following acquisition parameters:epetition time ms/echo time ms, 35/9; number of signalscquired, one; flip angle, 60◦; acquisition matrix, 256 × 192;ection thickness, 1.3 mm; field of view, 18 cm; and 124 con-iguous coronal sections. The MR data from 10 out of the7 subjects were previously included in another publicationhat used a regional brain boundary shift integral approach tovaluate changes in the medial temporal lobe region (Rusinekt al., 2003).

The imaging analysis protocol has been explained inetails elsewhere (see (Thompson et al., 2004)). Briefly,fter image inhomogeneity correction and 9-parameter linearpatial normalization to the ICBM53 brain atlas, all hip-ocampi were manually traced by one experienced tracerAG, intra-rater reliability α = 0.98, inter-rater reliability= 0.9) following an extensively validated hippocampal trac-

ng protocol (Jack et al., 1995). 3D hippocampal mesh modelsere computed and the central core of each hippocampal

tructure derived, as a medial curve threading down the cen-roid of each individual’s hippocampus. The radial distancei.e., the distance from the central core to each surface point)as determined and digitally recorded at each 3D coordinate

ocation of the hippocampal mesh models. Average groupippocampal models for both baseline and follow-up wereerived and statistically compared. (For a schematic of thisethod see Apostolova and Thompson, 2007; Thompson et

l., 2004.) Permutation tests with a threshold set at p < 0.01ere applied to the hippocampal statistical maps (p maps)

o control for multiple comparisons and provide an overall,lobal p value for the group differences shown in the map.

jo

able 1emographic and cognitive data

NL-NL

Mean S.D.

ge (years) 69.1 2.6ender M:F 4:6 N/Aducation (years) 16.4 2.2poE4 (pos/neg) 3/7 N/Aime to follow-up MRI (years) 3.2 0.8ime to clinical follow-up (years) 2.8 1.1aseline MMSE 29.7 0.7ollow-up MMSE 29.2 1.3aseline GDS 2 0ollow-up GDS 1.9 0.32aseline paragraph recall 5.8 2.2ollow-up paragraph recall 6.9 1.6aseline PADR 10.5 3.0ollow-up PADR 10.6 2.1

ADR—paired associates delayed recall. Follow-up is at the 3-year time point.

f Aging 31 (2010) 1077–1088 1079

detailed explanation of the permutation-based correctionor multiple comparisons is provided elsewhere (Thompsont al., 2004).

Additionally we investigated the associations betweenippocampal radial distance and one global (MMSE) andwo verbal memory cognitive measures (paired associateselayed recall and paragraph delayed recall). Individual cog-itive scores were used as covariates in a general linear modelhat predicted hippocampal radial distance in each subject.he resulting 3D hippocampal maps showed the strengthf that correlation (i.e., the correlation coefficient) at eachoint of the hippocampal surface. Significance maps werereated and corrected for multiple comparisons by permuta-ion testing using a threshold of p < 0.01. We hypothesizedhat areas that showed the greatest between-group differencest baseline and follow-up would also demonstrate strong cor-elations between atrophy and verbal memory and MMSEerformance.

Between-group volumetric comparisons were separatelyerformed for each time point via two-sample T-test and con-rmed via one-way Analysis of Variance with Tukey-Kramerost hoc test. 3-year atrophy rates and 95% confidencentervals were computed with standard Repeated Measuresnalysis of Variance via Wilks’ Lambda test (i.e. null hypoth-

sis no time by group effects) with conventional Alpha sett 0.05. Two a priori assumptions underpinned our analyses:hat the annual atrophy rate was linear and that hippocampalolume loss was not reversible. 1-year atrophy rates and theirespective 95% confidence intervals were thereby determineds the negative one third of the natural logarithm of (1 minushe 3-year atrophy rate) and compared for changes over time.

. Results

Subjects’ demographic data are shown in Table 1. Sub-ects who declined to MCIAD (NL-MCIAD) had fewer yearsf education and contained a larger proportion of women

NL-MCIAD Statistic p value

Mean S.D.

72.2 4.5 T 0.140:7 N/A χ2 0.0614 2.3 T 0.0524/3 N/A χ2 0.32.9 0.8 T 0.442.8 0.8 T 0.9229.1 0.9 T 0.227.4 2.3 T 0.11.9 0.4 T 0.43.0 0 T <0.0013.7 3.6 T 0.212.7 2.8 T 0.0076.7 2.9 T 0.024.7 3.8 T 0.005

1080 L.G. Apostolova et al. / Neurobiology of Aging 31 (2010) 1077–1088

Table 2Hippocampal volumetric data

Group Side Volume, mm3—mean (S.D.) Annualized volume loss % (95% CI)

Baseline Follow-up

NL-NL Left hippocampus 3088 (527) 2963 (416) 1.1% (-1.1; 3.27)Right hippocampus 3194 (550) 2984 (514) 2.0% (-0.1; 4.2)

NL-MCIAD Left hippocampus 2602 (793) 2309 (774) 3.7% (1.1; 6.3)Right hippocampus 2697 (846) 2448 (813) 3.2% (0.7; 5.8)

Fig. 1. Baseline 3-D comparison maps of the NL-NL and the NL-MCIAD groups. The significance maps show regions with significantly more atrophy in theNL-MCIAD group at baseline (middle row). The ratio maps (bottom row) show the quantitative between-group differences (in %). The subfield definitions (toprow) are based on Duvernoy (1988) and West and Gundersen (1990). In the significance maps, dark blue colors denote p-values of 0.1 or higher.

iology o

astNpaTM

vgwNf

FN(

L.G. Apostolova et al. / Neurob

nd ApoE-4 carriers (57% NL-MCIAD vs. 30% NL-NLubjects) relative to the stable NL (NL-NL) group, althoughhese differences did not reach statistical significance. TheL-MCIAD group performed significantly worse on theaired associates test both at baseline and at 3-year follow-up

nd on the paragraph delayed recall test at 3-year follow-up.here were no baseline or 3-year follow-up differences inMSE scores between the two groups. The hippocampal

s(T

ig. 2. Follow-up 3-D comparison maps of the NL-NL and the NL-MCIAD groups.L-MCIAD group at follow-up (middle row). The ratio maps (bottom row) show t

top row) based on Duvernoy (1988) and West and Gundersen (1990). In the signifi

f Aging 31 (2010) 1077–1088 1081

olumetric data and the annualized atrophy rates for eachroup are shown in Table 2. The left hippocampal volumeas significantly smaller in the NL-MCIAD relative to theL-NL group both at baseline (p = 0.04) and at 3-year

ollow-up (p = 0.006) while the right hippocampus only

howed between-group difference at 3-year follow-upp = 0.05). Repeated measures ANOVA showed significantime and Group effects for both the left (Time F = 8.12,

The significance maps show regions with significantly more atrophy in thehe quantitative between-group differences (in %). The subfield definitionscance maps, dark blue colors denote p-values of 0.1 or higher.

1082 L.G. Apostolova et al. / Neurobiology of Aging 31 (2010) 1077–1088

F l atrophc anel) an

p(i

oa

ah

Fs

ig. 3. (A) 3-D longitudinal quantitative within-group hippocampal radiahange—right panel.) (B) 3-D statistical maps comparing the annual (left p

= 0.01; Group F = 7.8, p = 0.01) and right hippocampusTime F = 9.65, p < 0.01; Group F = 4.65, p < 0.05) while the

nteraction term (Time × Group) was not significant.

The baseline and 3-year follow-up 3-D comparison mapsf the hippocampal radial distances between the NL-NLnd NL-MCIAD groups are shown in Figs. 1 and 2. A 4D

∼fu1

ig. 4. 3-D quantitative maps depicting the absolute interim atrophy in NL-NL vsubtracting the ratio maps at baseline from those at follow-up).

y maps (in %). (Annualized atrophy change—left panel, 3-year atrophyd 3-year (right panel) atrophy rate of the NL-NL vs. the NL-MCIAD.

nimation of the spread of significant changes through theippocampus can be viewed at http://www.loni.ucla.edu/

lianaa/NLtoMCI.mov. The movie frames were computed

rom the maps shown in the middle rows of Figs. 1 and 2sing a previously described algorithm (Thompson and Toga,997). We observed significant difference in the left at

. NL-MCIAD (i.e., 3-year change from baseline to follow-up computed by

L.G. Apostolova et al. / Neurobiology of Aging 31 (2010) 1077–1088 1083

Fig. 5. 3-D radial distance variability maps (in S.D.) of the NL-NL and the NL-MCIAD groups at baseline (top row) and follow-up (bottom row).

Fig. 6. 3-D cognitive correlations maps. The significance maps (top) show the regions where significant negative correlations between the given cognitivemeasure and hippocampal atrophy was observed. The correlation maps (bottom) show the regional strength of the cognitive correlation. In the significancemaps, dark blue colors denote p-values of 0.1 or higher.

1 iology o

btfpAatrcBCtsdp

cpasbcp

Vpt

084 L.G. Apostolova et al. / Neurob

aseline (left pcorrected = 0.05) and a trend for significance onhe right (right pcorrected = 0.06; Fig. 1 middle row). At 3-yearollow-up the differences were much more pronounced (leftcorrected = 0.001, right pcorrected < 0.02; Fig. 2 middle row).t baseline, the NL-MCIAD group showed up to 30% greater

trophy in the subiculum and up to 20% greater atrophy inhe CA1 area relative to the NL-NL group (Fig. 1, bottomow). At 3-year follow-up, the subicular and CA1 atrophichanges appeared greater both spatially and in magnitude.etween-group differences were also seen in the CA2 andA3 areas at 3-year follow-up (Fig. 2, bottom row). Quanti-

ative maps of the longitudinal within-group changes can beeen in Figs. 3 and 4. As expected the NL-MCIAD groupemonstrated substantially higher annual and global atro-hy rates throughout the hippocampal surface. The interim

bsaF

Fig. 7. 3-D cognitive correlations m

Fig. 8. 3-D cognitive correlations ma

f Aging 31 (2010) 1077–1088

hange (i.e., 3-year change from baseline to follow-up com-uted by subtracting the ratio maps at baseline from thoset follow-up) can be seen in Fig. 4. To ascertain whether thepatial pattern of statistical differences was not influenced byetween-group differences in radial distance variability, wereated 3D variability maps for each group and each timeoint (Fig. 5).

The 3-D cognitive correlation maps are shown in Fig. 6.erbal memory performance (paragraph delayed recall andaired associates delayed recall) showed stronger correla-ions with hippocampal radial distance relative to MMSE

oth at baseline and at 3-year follow-up. As predicted,trongest cognitive correlations (r > 0.4) were observed inreas with the greatest between-group differences (seeigs. 1–2, bottom rows and Fig. 6). In agreement with the

aps in the NL-NL group.

ps in the NL-MCIAD group.

iology o

btcafptrripg

4

mA(emts3bslwtbojeaaaab(tstMd2

oaafitaUbp

tAdfioosA

Malldstdesvdmg

itcsiaaTcgbcln

hfurtatiig

bf

L.G. Apostolova et al. / Neurob

etween-group cognitive comparisons (Table 1, last column),he paired associates delayed recall score showed statisti-ally strongest associations with hippocampal atrophy botht baseline (left pcorrected = 0.06; right pcorrected = 0.007) andollow-up (left pcorrected = 0.03; right pcorrected = 0.005) whilearagraph recall only showed significant associations withhe right hippocampus in follow-up (left pcorrected = 0.17;ight pcorrected = 0.005). No significant associations betweenegional hippocampal atrophy and MMSE scores follow-ng stringent multiple comparison correction at the threshold< 0.01 were observed. The 3D correlation maps within eachroup are shown in Figs. 7 and 8.

. Discussion

Neuritic plaques and neurofibrillary tangles begin accu-ulating years before the earliest clinical features oflzheimer’s disease (e.g., memory loss) manifest themselves

Hyman, 1997; Price, 1997). The hippocampus is affectedarly in the disease process by neurofibrillary tangle accu-ulation, which spreads in a well-defined trajectory. The first

angles are typically seen in the entorhinal cortex; next, theypread to the CA1 and subicular areas, then to the CA2 andand finally the CA4 areas of the hippocampal formation

efore invading the neocortex (Schonheit et al., 2004). Theubfield-specific variability in tangle density and neuronaloss has been well documented. The subiculum and CA1ere reported to be the most affected and CA3 and CA4

he least affected hippocampal subfields in post-mortem ADrains (Bobinski et al., 1995, 1997). One post-mortem studyf 39 nondemented elderly reported that 56% of the sub-ects had neurofibrillary pathology in the transentorhinal andntorhinal areas (Braak stages I and II), 28% also had moder-te involvement of the CA1 and subiculum (Braak stage III)nd 13% demonstrated both heavy hippocampal involvementnd evidence of neocortical tangle pathology (Braak stage IVnd higher) (Knopman et al., 2003). Braak stages I–IV haveeen assigned to more than 80% of MCI subjects at autopsyPetersen et al., 2006). Prior MRI and PET data also suggesthe vulnerability of the hippocampal formation in the earlytages of AD. This has been observed in studies predictinghe decline from MCI to AD and most recently from NL to

CI and AD (Apostolova et al., 2006b; de Leon et al., 1993;en Heijer et al., 2006; Mosconi et al., 2007; Rusinek et al.,003).

We found that on average 3 years prior to the diagnosisf amnestic MCI and 6 years prior to the diagnosis of prob-ble AD, subjects who are predestined to become dementedlready show disease-associated changes. Our study is therst to demonstrate that these changes initially localize to

he subiculum and the CA1 subfield—the two hippocampal

reas pathologically affected very early in the disease course.pon imaging 3(±1) years later, subjects normal at baselineut now meeting MCI criteria manifested hippocampal atro-hy spread to the CA2 and CA3 subfields. To our knowledge,

airo

f Aging 31 (2010) 1077–1088 1085

his is the first longitudinal neuroimaging study of pre-clinicalD that maps this highly specific and orderly progression ofisease-associated changes through the hippocampus. Ourndings suggest CA1 and subicular involvement is predictivef cognitive decline to MCI while progressive involvementf the CA1 and subiculum and atrophy spread to the CA2-3ubfield in amnestic MCI is predictive of future diagnosis ofD.The annual atrophy rates in the NL-NL and the NL-

CIAD group (NL-NL 1.1–2% and NL-MCIAD 3.2–3.7%nnually) are comparable to those already reported in theiterature (Jack et al., 2000, 1998, 2004). Both the base-ine and the 3-year follow-up hippocampal comparisonsemonstrated restricted areas where the NL-MCIAD grouphowed greater radial distance (thicker hippocampus) rela-ive to the NL-NL group (Figs. 1 and 2, bottom row). Theseifferences were not statistically significant. One plausiblexplanation is between-subject variability of hippocampalhape (Carmichael et al., 2005). However, a reassuring obser-ation is that at 3-year follow-up, as the groups become moreivergent in their cognitive performance, these areas appearuch smaller relative to the baseline comparison where both

roups exhibited normal cognition.In addition we observed restricted areas of apparent gain

n radial distance over time that were more pronounced inhe NL-NL vs. the NL-MCIAD group (see Fig. 3). Thesehanges were not statistically significant. There are two pos-ible reasons for the occurrence of apparent positive changen the radial distance over time. Although presently accepteds the gold standard, hand tracing of the hippocampus orny other region of interest is not perfectly reproducible.his could lead to net gains in radial distance even when nohange has occurred (i.e., in the absence of true effect). Suchains should not be interpreted as meaningful unless theyeat the 0.05 significance level when corrected for multipleomparisons. Furthermore subtle movement artifacts wouldikewise decrease the signal-to-noise ratio and could result inon-biological effects.

In the current study we came across one occurrence of aippocampal fissure—in the left hippocampus of one subjectrom the NL-MCIAD group at baseline and at 3-year follow-p. We decided to use the most conservative approach inespect to our hypotheses and did not extend the trace inhe fissure (rather we have rounded the hippocampal contournd included the fissure in the trace). This ultimately leadso an increased left hippocampal volumetric measurementn this specific subject and works against our hypothesis ofncreased atrophy in the NL-MCIAD relative to the NL-NLroup at baseline and follow-up.

Two other hippocampal imaging studies used a differentut conceptually related computational anatomy techniqueor 3D hippocampal modeling to study the effects of normal

ging vs. questionable AD (defined as subjects whose Clin-cal Dementia Rating scale (CDR) scores were 0 and 0.5,espectively, including both MCI and mild AD patients). Onef the studies reported that subjects deemed to be cognitively

1 iology o

naCsCt4ytt(btdTnbc

p(osepCtiaasvpttjtae

bcAtfcwcrpmMlrie

Mdfiow1setlwlts

beptwfrhbisawrchoittbrcv

tivsmupsstoh

086 L.G. Apostolova et al. / Neurob

ormal over 2 years (i.e., who had CDR = 0 both at baselinend follow-up) showed progressive atrophy of the anteriorA1 region (localized to the hippocampal head) and the

ubiculum. Subjects with questionable AD at baseline (i.e.,DR = 0.5) showed additional progressive atrophy of the pos-

erior CA1 area (Wang et al., 2003). Another study followed9 nondemented subjects with CDR of 0 for an average of 5ears. They reported isolated progressive atrophic changes ofhe left CA1 subfield in subjects who declined from CDR = 0o CDR of 0.5 (cognitively normal to questionable AD)Csernansky et al., 2005). Our results build on these findingsy showing the spread of disease-associated changes overime, through the hippocampal subfields, following the well-ocumented pathologic progression of AD-type pathology.hese results raise hopes for the utility of this mapping tech-ique not only as a sensitive diagnostic and prognostic toolut also as a surrogate marker for future disease-modifyinglinical trials.

Two studies to date have segmented out the hippocam-al subfields in cognitively normal elderly and AD subjectsCsernansky et al., 2005; Mueller et al., 2007) and a thirdne in young adults (Zeineh et al., 2003). At 1.5 T there wasuboptimal resolution for subfield identification (Csernanskyt al., 2005). In a recent 4 T study using 2 mm thick T2 hip-ocampal image sections Mueller et al. resolved the CA1,A2, subiculum and the combined CA3-4 areas and traced

hem manually (Mueller et al., 2007). While identification ofndividual hippocampal layers at 4 T was still not possible, theuthors used a set of reliable anatomical landmarks for bound-ry approximation. The authors did not resolve and trace theubfields of the whole structure; they estimated the subfieldolumes from a set of three contiguous slices immediatelyosterior to the hippocampal head under the assumption thathis 3-slice measurement correlates well with the volume ofhe whole subfield. The study reported that the three AD sub-ects had smaller CA1 and subiculum measurements relativeo their age-matched control counterparts, while advancedge seemed to exert an effect on the CA1 subfield (Muellert al., 2007).

Several strengths and weaknesses of our study need toe recognized. Major strengths of the study are the well-haracterized patient population and the longitudinal design.s MCI is heterogeneous, a strength of the study design is

he inclusion of cognitively normal subjects who were closelyollowed-up until they met the stringent NINCDS-ADRDAriteria for probable AD, minimizing possible contaminationith other dementing disorders. The use of a state-of-the art

omputational anatomy technique enabled us to uncover sub-egionally specific hippocampal involvement in subjects withre-clinical AD and to dissociate different patterns of involve-ent in cognitively normal subjects who are transitioning toCI and MCI subjects who are transitioning to AD. The

imitations of our study lie in the relatively small sample sizeestricting our ability to generalize our findings and the strictnclusion of only the amnestic MCI subtype that preventsxtension of our conclusions to subjects with nonamnestic

limp

f Aging 31 (2010) 1077–1088

CI. The resolution at 1.5 T does not permit accurate subfieldelineation. Thus our conclusions in respect to selective sub-eld involvement are based on assumptions about the locationf each subfield boundary made in consultation with twoell-established hippocampal histologic atlases (Duvernoy,988; West and Gundersen, 1990). Although our findingseem consistent with the previously published pathologicvidence of the trajectory of neurofibrillary tangle spreadhrough the hippocampus (Bobinski et al., 1998, 2000), theack of histology-based subfield definition calls for cautionhen drawing conclusions from these data. Without patho-

ogical validation, both the diagnosis of AD and the inferencehat there is underlying neurofibrillary tangle pathology in thepecific hippocampal subfields remain to be ascertained.

Our algorithm, as well as several approaches developedy other groups (e.g., Csernansky et al., 2005), measures thextent and severity of hippocampal shape deformations as aroxy for hippocampal atrophy. To estimate the severity ofhe inward deformations or hippocampal thinning that occursith hippocampal atrophy we compute the radial distance

rom the central core of each hippocampal structure to theespective points on the surface. The areas of thinning thusave smaller radial distance to the core. Even so, it cannote ruled out that hippocampal deformations may have mim-cked subfield atrophy, and there are some limitations in usingurface mesh models to detect heterogeneous hippocampaltrophy. For example, if atrophy occurred in the dorsal sidehile volume increase occurred in the ventral side, this could

esult in a non-significant finding due to the shifting of theentral skeleton. This is unlikely, as hippocampal growth orypertrophy is not expected in aging or AD. Even so, the usef the central axis as a reference to gauge atrophy is a strengthn many ways as the radial atrophy measures will be invarianto any overall shifting of the structure in space. In addition,he manual delineations allow highly precise delineation ofoundaries for assessing atrophy, whereas a more automatedegistration method, like voxel-based morphometry, can typi-ally match the hippocampal boundaries across subjects onlyery approximately.

Another strong advantage of the surface-based mappingechnique relative to other voxel-based mapping approachess its imperviousness to shifts in stereotaxic space. In otheroxel-based mapping approaches, if a structure shifts intereotaxic space, the shift can be recovered using auto-ated nonlinear registration as a deformation, but it is highly

nlikely that an automated alignment will register the com-lex shape boundaries of the hippocampus accurately acrossubjects, and because the deformation is constrained to bepatially smooth, it is almost inevitable that some changes inhe structure would be incorrectly inferred if the structuresnly shifted. In the medial axis mapping approach we usedere, however, if the structure shifts, the medial axis is trans-

ated by a corresponding amount, so there is no net alterationn the amount of atrophy. This is an advantage of using the

edial axis curve as a reference, as the measures of 3D atro-hy relative to this curve are shift invariant, and as intrinsic

iology o

mt

C

flt

D

idi

CoiUhsnma

A

AASPLNPA

R

A

A

A

B

B

B

B

B

C

C

d

d

d

D

D

F

F

F

G

H

H

L.G. Apostolova et al. / Neurob

easures they should also not depend on the specifics of howhe images are registered.

onflicts of interest

Dr. Thompson, Ms. Green and Ms. Mistur have no con-icts of interest. Ms. Hwang and Mr. Ramirez have nothing

o disclose.

isclosures

Dr. Apostolova received personal compensation for speak-ng for Forest, OMN and Myriad. The reported financialisclosures are not in any way related to the work presentedn this manuscript.

Dr. Mosconi is a consultant for Abiant Imaging Inc.,hicago, IL. Drs Mosconi, Tsui, and de Leon are coinventorsf and own a patent on the HipMask technology. Abiant Imag-ng Inc. entered into a License Agreement with New Yorkniversity based on this technology and, as such, the authorsave a financial interest in this License Agreement and holdtock and stock options on the company. The HipMask tech-ology has not been used in the data analyses reported in thisanuscript and the reported financial disclosures are not in

ny way related to the work presented in this manuscript.

cknowledgments

This work was generously supported by NIA K23G026803 (jointly sponsored by NIA, AFAR, The John. Hartford Foundation, the Atlantic Philanthropies, thetarr Foundation and an anonymous donor; to LGA), NIA50 AG16570 (to LGA and PMT); NIBIB EB01651, NLMM05639, NCRR RR019771, NIH/NIMH R01 MH071940,IH/NCRR P41 RR013642 and NIH U54 RR021813 (toMT); the Alzheimer’s Association (to LM); and NIH-NIAG13616, AG12101, AG08051, and AG022374 (to MdL).

eferences

postolova, L.G., Dinov, I.D., Dutton, R.A., Hayashi, K.M., Toga, A.W.,Cummings, J.L., Thompson, P.M., 2006a. 3D comparison of hippocam-pal atrophy in amnestic mild cognitive impairment and Alzheimer’sdisease. Brain 129 (Pt 11), 2867–2873.

postolova, L.G., Dutton, R.A., Dinov, I.D., Hayashi, K.M., Toga, A.W.,Cummings, J.L., Thompson, P.M., 2006b. Conversion of mild cogni-tive impairment to Alzheimer disease predicted by hippocampal atrophymaps. Arch. Neurol. 63 (5), 693–699.

postolova, L.G., Thompson, P.M., 2007. Brain mapping as a tool to studyneurodegeneration. Neurotherapeutics 4 (3), 387–400.

ecker, J.T., Davis, S.W., Hayashi, K.M., Meltzer, C.C., Toga, A.W., Lopez,O.L., Thompson, P.M., 2006. Three-dimensional patterns of hippocam-pal atrophy in mild cognitive impairment. Arch. Neurol. 63 (1), 97–101.

obinski, M., de Leon, M.J., Tarnawski, M., Wegiel, J., Reisberg, B., Miller,D.C., Wisniewski, H.M., 1998. Neuronal and volume loss in CA1 of

H

f Aging 31 (2010) 1077–1088 1087

the hippocampal formation uniquely predicts duration and severity ofAlzheimer disease. Brain Res. 805 (1–2), 267–269.

obinski, M., de Leon, M.J., Wegiel, J., Desanti, S., Convit, A., Saint Louis,L.A., Rusinek, H., Wisniewski, H.M., 2000. The histological validationof post mortem magnetic resonance imaging-determined hippocampalvolume in Alzheimer’s disease. Neuroscience 95 (3), 721–725.

obinski, M., Wegiel, J., Tarnawski, M., Bobinski, M., Reisberg, B., de Leon,M.J., Miller, D.C., Wisniewski, H.M., 1997. Relationships betweenregional neuronal loss and neurofibrillary changes in the hippocampalformation and duration and severity of Alzheimer disease. J. Neu-ropathol. Exp. Neurol. 56 (4), 414–420.

obinski, M., Wegiel, J., Wisniewski, H.M., Tarnawski, M., Reisberg, B.,Mlodzik, B., de Leon, M.J., Miller, D.C., 1995. Atrophy of hippocampalformation subdivisions correlates with stage and duration of Alzheimerdisease. Dementia 6 (4), 205–210.

armichael, O.T., Aizenstein, H.A., Davis, S.W., Becker, J.T., Thompson,P.M., Meltzer, C.C., Liu, Y., 2005. Atlas-based hippocampus segmenta-tion in Alzheimer’s disease and mild cognitive impairment. Neuroimage27 (4), 979–990.

sernansky, J.G., Wang, L., Swank, J., Miller, J.P., Gado, M., McKeel, D.,Miller, M.I., Morris, J.C., 2005. Preclinical detection of Alzheimer’sdisease: hippocampal shape and volume predict dementia onset in theelderly. Neuroimage 25 (3), 783–792.

e Leon, M.J., Convit, A., Wolf, O.T., Tarshish, C.Y., DeSanti, S.,Rusinek, H., Tsui, W., Kandil, E., Scherer, A.J., Roche, A., Imossi,A., Thorn, E., Bobinski, M., Caraos, C., Lesbre, P., Schlyer, D.,Poirier, J., Reisberg, B., Fowler, J., 2001. Prediction of cognitivedecline in normal elderly subjects with 2-[(18)F]fluoro-2-deoxy-D-glucose/positron-emission tomography (FDG/PET). Proc. Natl. Acad.Sci. U.S.A. 98 (19), 10966–10971.

e Leon, M.J., Golomb, J., George, A.E., Convit, A., Tarshish, C.Y., McRae,T., De Santi, S., Smith, G., Ferris, S.H., Noz, M., 1993. The radiologicprediction of Alzheimer disease: the atrophic hippocampal formation.AJNR Am. J. Neuroradiol. 14 (4), 897–906.

en Heijer, T., Geerlings, M.I., Hoebeek, F.E., Hofman, A., Koudstaal, P.J.,Breteler, M.M., 2006. Use of hippocampal and amygdalar volumes onmagnetic resonance imaging to predict dementia in cognitively intactelderly people. Arch. Gen. Psychiatry 63 (1), 57–62.

ubois, B., Feldman, H.H., Jacova, C., Dekosky, S.T., Barberger-Gateau,P., Cummings, J., Delacourte, A., Galasko, D., Gauthier, S., Jicha, G.,Meguro, K., O’Brien, J., Pasquier, F., Robert, P., Rossor, M., Salloway, S.,Stern, Y., Visser, P.J., Scheltens, P., 2007. Research criteria for the diag-nosis of Alzheimer’s disease: revising the NINCDS-ADRDA criteria.Lancet Neurol. 6 (8), 734–746.

uvernoy, H., 1988. The Human Hippocampus. An Atlas of AppliedAnatomy. J.F. Bergmann Verlag, Munich.

licker, C., Ferris, S.H., Reisberg, B., 1993. A two-year longitudinal study ofcognitive function in normal aging and Alzheimer’s disease. J. Geriatr.Psychiatry Neurol. 6 (2), 84–96.

olstein, M.F., Folstein, S.E., McHugh, P.R., 1975. Mini-Mental State: apractical method for grading the cogntive state of patients for the clini-cian. J. Psychiatr. Res. 12, 189–198.

risoni, G.B., Sabattoli, F., Lee, A.D., Dutton, R.A., Toga, A.W., Thompson,P.M., 2006. In vivo neuropathology of the hippocampal formation in AD:a radial mapping MR-based study. Neuroimage 32 (1), 104–110.

ilbert, J.G., Levee, R.F., Catalano, F.L., 1968. A preliminary report on anew memory scale. Percept. Mot. Skills 27, 277–278.

aroutunian, V., Perl, D.P., Purohit, D.P., Marin, D., Khan, K., Lantz, M.,Davis, K.L., Mohs, R.C., 1998. Regional distribution of neuritic plaquesin the nondemented elderly and subjects with very mild Alzheimer dis-ease. Arch. Neurol. 55 (9), 1185–1191.

ebert, L.E., Scherr, P.A., Bienias, J.L., Bennett, D.A., Evans, D.A., 2003.

Alzheimer disease in the US population: prevalence estimates using the2000 census. Arch. Neurol. 60 (8), 1119–1122.

yman, B.T., 1997. The neuropathological diagnosis of Alzheimer’s dis-ease: clinical-pathological studies. Neurobiol. Aging 18 (4 Suppl.),S27–32.

1 iology o

J

J

J

J

K

M

M

M

M

P

P

P

P

R

R

S

S

T

T

W

W

W

088 L.G. Apostolova et al. / Neurob

ack Jr., C.R., Petersen, R.C., Xu, Y., O’Brien, P.C., Smith, G.E., Ivnik, R.J.,Boeve, B.F., Tangalos, E.G., Kokmen, E., 2000. Rates of hippocam-pal atrophy correlate with change in clinical status in aging and AD.Neurology 55 (4), 484–489.

ack Jr., C.R., Petersen, R.C., Xu, Y., O’Brien, P.C., Smith, G.E., Ivnik, R.J.,Tangalos, E.G., Kokmen, E., 1998. Rate of medial temporal lobe atrophyin typical aging and Alzheimer’s disease. Neurology 51 (4), 993–999.

ack Jr., C.R., Shiung, M.M., Gunter, J.L., O’Brien, P.C., Weigand, S.D.,Knopman, D.S., Boeve, B.F., Ivnik, R.J., Smith, G.E., Cha, R.H., Tan-galos, E.G., Petersen, R.C., 2004. Comparison of different MRI brainatrophy rate measures with clinical disease progression in AD. Neurol-ogy 62 (4), 591–600.

ack Jr., C.R., Theodore, W.H., Cook, M., McCarthy, G., 1995. MRI-basedhippocampal volumetrics: data acquisition, normal ranges, and optimalprotocol. Magn. Reson. Imaging 13 (8), 1057–1064.

nopman, D.S., Parisi, J.E., Salviati, A., Floriach-Robert, M., Boeve, B.F.,Ivnik, R.J., Smith, G.E., Dickson, D.W., Johnson, K.A., Petersen, L.E.,McDonald, W.C., Braak, H., Petersen, R.C., 2003. Neuropathologyof cognitively normal elderly. J. Neuropathol. Exp. Neurol. 62 (11),1087–1095.

cKhann, G., Drachman, D., Folstein, M., Katzman, R., Price, D., Stad-lan, E.M., 1984. Clinical diagnosis of Alzheimer’s disease: report ofthe NINCDS-ADRDA Work Group under the auspices of Departmentof Health and Human Services Task Force on Alzheimer’s Disease.Neurology 34 (7), 939–944.

osconi, L., 2005. Brain glucose metabolism in the early and specific diag-nosis of Alzheimer’s disease. FDG-PET studies in MCI and AD. Eur. J.Nucl. Med. Mol. Imaging 32 (4), 486–510.

osconi, L., De Santi, S., Li, J., Tsui, W.H., Li, Y., Boppana, M., Laska,E., Rusinek, H., de Leon, M.J., 2007. Hippocampal hypometabolismpredicts cognitive decline from normal aging. Neurobiol. Aging.

ueller, S.G., Stables, L., Du, A.T., Schuff, N., Truran, D., Cashdollar, N.,Weiner, M.W., 2007. Measurement of hippocampal subfields and age-related changes with high resolution MRI at 4T. Neurobiol. Aging 28(5), 719–726.

etersen, R.C., Doody, R., Kurz, A., Mohs, R.C., Morris, J.C., Rabins, P.V.,Ritchie, K., Rossor, M., Thal, L., Winblad, B., 2001. Current conceptsin mild cognitive impairment. Arch. Neurol. 58 (12), 1985–1992.

etersen, R.C., Parisi, J.E., Dickson, D.W., Johnson, K.A., Knopman, D.S.,Boeve, B.F., Jicha, G.A., Ivnik, R.J., Smith, G.E., Tangalos, E.G., Braak,

Z

f Aging 31 (2010) 1077–1088

H., Kokmen, E., 2006. Neuropathologic features of amnestic mild cog-nitive impairment. Arch. Neurol. 63 (5), 665–672.

rice, J.L., 1997. Diagnostic criteria for Alzheimer’s disease. Neurobiol.Aging 18 (4 Suppl.), S67–70.

rice, J.L., Morris, J.C., 1999. Tangles and plaques in nondementedaging and “preclinical” Alzheimer’s disease. Ann. Neurol. 45 (3),358–368.

eisberg, B., Ferris, S.H., de Leon, M.J., Crook, T., 1988. Global Deterio-ration Scale (GDS). Psychopharmacol. Bull. 24 (4), 661–663.

usinek, H., De Santi, S., Frid, D., Tsui, W.H., Tarshish, C.Y., Convit, A.,de Leon, M.J., 2003. Regional brain atrophy rate predicts future cog-nitive decline: 6-year longitudinal MR imaging study of normal aging.Radiology 229 (3), 691–696.

chonheit, B., Zarski, R., Ohm, T.G., 2004. Spatial and temporal relation-ships between plaques and tangles in Alzheimer-pathology. Neurobiol.Aging 25 (6), 697–711.

ilverman, D.H.S., Thompson, P.M., 2006. Structural and functional neu-roimaging: focussing on mild cognitive impairment. Appl. Neurol.,10–24.

hompson, P.M., Hayashi, K.M., De Zubicaray, G.I., Janke, A.L., Rose,S.E., Semple, J., Hong, M.S., Herman, D.H., Gravano, D., Doddrell,D.M., Toga, A.W., 2004. Mapping hippocampal and ventricular changein Alzheimer disease. Neuroimage 22 (4), 1754–1766.

hompson, P.M., Toga, A.W., 1997. Detection, visualization and animationof abnormal anatomic structure with a deformable probabilistic brainatlas based on random vector field transformations. Med. Image Anal. 1(4), 271–294.

ang, L., Swank, J.S., Glick, I.E., Gado, M.H., Miller, M.I., Morris, J.C.,Csernansky, J.G., 2003. Changes in hippocampal volume and shapeacross time distinguish dementia of the Alzheimer type from healthyaging. Neuroimage 20 (2), 667–682.

est, M.J., Gundersen, H.J., 1990. Unbiased stereological estimation of thenumber of neurons in the human hippocampus. J. Comp. Neurol. 296(1), 1–22.

imo, A., Jonsson, L., Winblad, B., 2006. An estimate of the worldwide

prevalence and direct costs of dementia in 2003. Dement. Geriatr. Cogn.Disord. 21 (3), 175–181.

eineh, M.M., Engel, S.A., Thompson, P.M., Bookheimer, S.Y., 2003.Dynamics of the hippocampus during encoding and retrieval of face-name pairs. Science 299 (5606), 577–580.