HIPPOCAMPUS 24:403–414 (2014)
Regional Vulnerability of Hippocampal Subfields to Aging Measured by Structural and Diffusion MRI Joana B. Pereira,1,2,3 Cinta Valls-Pedret,4,5 Emilio Ros,4,5 Eva Palacios,1,2 Carles Falc on,2,6 Nuria Bargall o,2,7 David Bartres-Faz,1,2* Lars-Olof Wahlund,3 Eric Westman,3 and Carme Junque1,2
Abstract: In the past few years, there has been an increasing awareness of the regional vulnerability of the hippocampus to age-related processes. However, to date, no studies have assessed the effects of age on different structural magnetic resonance parameters in the specific hippocampal subfields. In this study, we measured volume, mean diffusivity (MD) and fractional anisotropy (FA) in the presubiculum, subiculum, fimbria, cornu ammonis (CA) 1,2-3,4-DG and the whole hippocampus in fifty cognitively intact elder adults between 50 and 75 years of age (20 men, 30 women). Segmentation of hippocampal subfields was performed using FreeSurfer. Individual MD and FA images were coregistered to T1-weighted volumes using FLIRT of FSL. Linear regression analyses were performed to assess the effects of age on the anatomical measures of each subfield. In addition, multiple regression analyses were also carried out to assess which of the anatomical measures that showed a correlation with age in the previous analyses, were the best age predictors in the hippocampus. In agreement with previous studies, our results showed a significant association between age and volume (P < 0.001) as well as MD (P < 0.001) in the whole hippocampus. Regarding the specific hippocampal subfields, we found that age had a significant negative effect on volume in CA2-3 (P < 0.001) and CA4-DG (P < 0.001). Importantly, we found a positive effect of age on MD in CA2-3 (P < 0.001) and fimbria (P < 0.001) as well as a negative age effect on FA in the subiculum (P < 0.001). Multiple regression analyses revealed that the best overall predictors of age in the hippocampus were MD in the fimbria and volume of CA2-3, which explained 73.8% of the age variance. These results indicate that age has an effect both on volume and diffusion tensor imaging measures in different subfields, suggesting they provide complementary information on age-related C 2013 Wiley Periodicals, Inc. processes in the hippocampus. V
1
Department of Psychiatry and Clinical Psychobiology, University of Barcelona, Spain; 2 Institute of Biomedical Research August Pi i Sunyer (IDIBAPS), Barcelona, Spain; 3 Department of Neurobiology, Care Sciences and Society, Karolinska Institutet, Stockholm, Sweden; 4 Lipid Clinic, Endocrinology and Nutrition Service, Institut d’Investigacions Biomediques August Pi Sunyer (IDIBAPS), Hospital Clınic, Barcelona, n (CIBERobn), Spain; 5 CIBER Fisiopatologıa de la Obesidad y Nutricio Instituto de Salud Carlos III (ISCIII), Barcelona, Spain; 6 CIBER-BBN, stic per la Imatge Hospital Clinic Barcelona, Spain; 7 Centre de Diagno de Barcelona (CDIC), Barcelona, Spain Additional Supporting Information may be found in the online version of this article. Grant sponsor: Spanish Ministerio de Ciencia e Innovaci on; Grant number: SAF2009-07489; Grant sponsor: Ministerio de Economıa y Competitividad; Grant number: PSI2012-38257; Grant sponsor: Ministerio de Sanidad, Polıtica Social e Igualdad; Grant number: IMSERSO 200/2011. *Correspondence to: David Bartres-Faz, Departament de Psiquiatria i Psicobiologia Clınica, Universitat de Barcelona, IDIBAPS, Casanova 143, 08036 Barcelona, Spain. E-mail: [email protected] Accepted for publication 6 December 2013. DOI 10.1002/hipo.22234 Published online 12 December 2013 in Wiley Online Library (wileyonlinelibrary.com). C 2013 WILEY PERIODICALS, INC. V
KEY WORDS: volume; mean diffusivity; fractional anisotropy; age; hippocampus
INTRODUCTION The hippocampus is one of the most well studied cerebral structures in healthy aging due to its high susceptibility to degenerative processes. However, there is still debate on whether hippocampal volume loss is related to increasing age. Some studies have reported significant volume loss (Fjell et al., 2010; Zhang et al., 2010; Walhovd et al., 2011), while others have found that hippocampal volumes remain quite stable over aging (Head et al., 2005; Sullivan et al., 2005). These discrepancies have been related to age differences in the studied samples as studies showing hippocampal reductions with aging included elderly subjects and found these changes only take place in late adulthood. In line with this, recent longitudinal studies assessing the trajectories of regional brain volumes over aging reported that the hippocampus shows an accelerated volume decline with age rather than a linear decline (Raz et al., 2010; Pfefferbaum et al., 2013), particularly after 60 years (Pfefferbaum et al., 2013). Although most neuroimaging studies have assessed the hippocampus as a single structure, it is actually a complex circuit made up of functionally and molecularly distinct subfields (Small et al., 2011) comprising the subiculum (with the subdivisions pre-subiculum, para-subiculum and subiculum proper), the four cornu ammonis (CA1-4) and the dentate gyrus (DG) (Duvernoy, 1998). The increasing awareness of regional vulnerability within the hippocampus to disease and age-related processes has led researchers to measure these different hippocampal subfields rather than the hippocampus as a whole. For instance, using a manual technique on highresolution 4T MRI scans a significant association was found between increasing age and volume loss in CA1 (Mueller et al., 2007), CA3 and the DG (Mueller et al., 2008). In a subsequent study by the same group a negative effect of age was found in the CA3
404
PEREIRA ET AL.
and DG subfields (Mueller et al., 2009), providing consistency to their previous findings (Mueller et al., 2008). Using a surface-based method to map age-related changes to the shape of the hippocampus, Frisoni et al. (2008) found atrophy associated with age in CA1 and the presubiculum. By applying a method based on voxel-based morphometry, La Joie et al. (2010) found a negative correlation between age and the volume of the subiculum in healthy individuals between 19 and 68 years of age. In addition to studies assessing gray matter volume in the hippocampus, the application of diffusion tensor imaging (DTI) for measuring microstructural changes in deep gray matter structures has recently received a lot of interest. DTI is a MRI technique with an unparalleled sensitivity to water movements and a potential in probing neuroanatomy at a microscopic level that is well beyond image resolution (Le Bihan et al., 2001). Two quantitative parameters derived from DTI have been shown to be greatly influenced by physiological aging (Cherubini et al., 2009). One of them is mean diffusivity (MD), which characterizes the overall displacement of water molecules and the overall presence of obstacles to diffusion. The other is fractional anisotropy (FA), which describes how much molecular displacements vary in space and is related to the presence and coherence or oriented structures (Le Bihan et al., 2001). Increases in MD and reductions of FA have been described in several brain areas and structures in elder individuals (Hasan et al., 2007; Abe et al., 2008; Jeon et al., 2012). Previous histological studies suggest that these changes are related to the degradation of the axonal membranes, myelin sheaths (Sen and Basser, 2005) and general loss of ordered anisotropic tissue structures (Pierpaoli and Basser, 1996). These phenomena increase the extracellular volume fraction and allow larger net displacements of water molecules (Sodak, 2004) in all directions in the same amount. Moreover, loss of intracellular viscosity, macromolecular crowding (Hazlewood et al., 1991) and extracellular tortuosity have also been shown to affect diffusion (Chen and Nicolson, 2000; Norris, 2001). In both white and gray matter, MD is an indicator of microscopic barriers or obstacles to diffusion. The loss of microscopic barriers in white matter is associated with axonal degeneration and/or dysmyelination, while in gray matter is most likely related to axonal degeneration due to the relative absence of myelination in this brain tissue. In white matter, FA reflects the integrity of large fiber bundles such as the association, commissural and projection fibers. By contrast, in gray matter FA is harder to interpret due to the existence of multiple crossing axonal fibers that lower the FA values in this tissue compared to white matter. Despite of this, several studies have shown that FA in gray matter regions offer valuable information, suggesting they also reflect the integrity of oriented structures in gray matter including unmyelinated axons. For instance, previous studies have shown FA increases in the putamen and caudate in elder compared to younger individuals (Pfefferbaum et al., 2010) as well as in patients with epilepsy (Gerdes et al., 2012) and Huntington’s disease (Bohanna et al., 2011). In the hippocampus, previous studies showed significant Hippocampus
FA decreases in patients with Alzheimer’s disease (AD; Fellgiebel et al., 2004; Muller et al., 2007) or a negative correlation between hippocampal FA and age in elder individuals (Carlesimo et al., 2010; Heijer et al., 2012), suggesting that FA is sensitive to microstructural orientation in subcortical gray matter. Changes in DTI measures have been suggested to occur before neuronal degradation and atrophy are detectable at a macroscopic level (Muller et al., 2007; Heijer et al., 2012). In two recent studies comparing gray matter volume and DTI parameters in the whole hippocampus, the correlation coefficients between age and MD were higher than those between age and hippocampal volume and FA (Carlesimo et al., 2010; den Heijer et al., 2012). Reasons that may account for the higher sensitivity of MD to age compared to volume include the fact that it assesses subtle microstructural pathological alterations including the density of physical obstructions such as cellular membranes and the distribution of water molecules between cellular compartments (Beaulieu et al., 2002; Sen and Basser, 2005). By contrast, volume detects abnormalities only when significant tissue shrinkage at the macrostructural level has taken place, possibly due to reductions in neuropil, neuronal size, dendritic or axonal arborization (Mechelli et al., 2005). On the other hand, the higher sensitivity of MD to age compared to FA in the hippocampus could be related to the fact that MD is less affected by fiber crossing than FA because it reflects the magnitude of water diffusion, which is not influenced by direction (Pierpaoli et al., 1996). By contrast, FA is a measure of the directional dependence of diffusion (Basser et al., 1995), which is lower in gray matter structures such as the hippocampus due to the presence of multiple crossing fibers with different directions and lower intrahippocampal fiber coherence (van Norden et al., 2012). Despite providing relevant information in the whole hippocampus, to date no studies have assessed DTI measures within its specific subfields. Hence, the goal of the current study was to assess volume, MD, and FA in hippocampal subfields and test their association with age in a sample of nondemented older adults. First of all, we assessed the effects of age on both volume and DTI measures in the whole hippocampus to assess whether our findings agreed with those of previous studies (Carlesimo et al., 2010; den Heijer et al., 2012). Second, we evaluated the association between age and hippocampal subfields volumes using a novel neuroimaging technique developed by van Leemput et al. (2009) in ultra-high resolution MRI data, which can be applied to standard resolution MRI at 1.5 or 3 T (Engvig et al., 2012; Kuhn et al., 2012; Teicher et al., 2012; Pereira et al., 2013). In addition to measuring the volumes of CA1–4, DG and subiculum, this method also measures the white matter hippocampal band or fimbria. Moreover, it differentiates the presubiculum from the subiculum proper and delineates the hippocampal fissure, offering a more complete overview of hippocampal subregions than previous studies assessing hippocampal subfields volumes (Mueller et al., 2007, 2009; La Joie et al., 2010). Finally, we tested the association between age and MD as well FA in hippocampal subfields,
EFFECTS OF AGE ON VOLUME, MD, AND FA IN HIPPOCAMPAL SUBFIELDS which is the main novelty of our study. Based on previous studies showing a higher correlation between age and MD in the whole hippocampus (Carlesimo et al., 2010; den Heijer et al., 2012), we hypothesized that age would have a stronger effect on MD in the whole hippocampus compared to volume and FA in our study. Regarding the specific subfields, we expected a greater age effect in both volume and DTI measures in CA1, CA3, and DG compared to the other subregions. This hypothesis was based on previous studies showing a correlation between age and the volumes of CA1, CA3, and DG in a sample with a similar age range to ours (Mueller et al., 2007, 2009). Although previous studies did not include DTI subfields measures (Mueller et al., 2007, 2009), we predicted that if age correlated with macrostructural volume shrinkage in CA1, CA3, and DG then it would also strongly correlate with microstructural abnormalities in the same subregions given that, as previously suggested, changes in DTI measures occur before neuronal degradation and atrophy are detectable at a macroscopic level (Muller et al., 2007; Heijer et al., 2012). There are several reasons why volume and DTI measures might be sensitive to age effects in the previous subfields. For instance, the high expression of NMDA receptors in CA1 makes this subfield vulnerable to excitotoxicity due to hypoxia and ischemia associated with vascular problems that occur over aging (Small et al., 2011). These vascular problems can produce variations in extracellular water diffusion as well as decreases in FA (Sotak et al., 2002). The high superoxide levels in CA1 may also lead to greater susceptibility of these neurons to an excess in oxidative stress that can occur during aging and cause neuronal death (Nicolle et al., 2001; Wang et al., 2005), which would affect both DTI and volume measures in this region. Regarding CA3, the age-related reductions in the number and area of the large pyramidal cells (Kadar et al., 1998) as well as dendritic tree atrophy (Fuchs et al., 2001) could shrunk the volume of this subfield, while more subtle alterations in perivascular morphology, increased accumulation of lipofuscin in the neuropil (Ojo et al., 2013) and loss of mossy fiber densities (Stephens et al., 2011) could also potentially induce microstructural changes in DTI metrics. Finally, the DG is known for its role in adult neurogenesis, which decreases during aging, leading to a decline of production of new hippocampal neurons (Small et al., 2011). In particular, the high levels of mineralocorticoid receptors confer vulnerability of this subfield to the effects of circulating corticosteroids, which can suppress cell proliferation (Brummelte and Galea, 2010) and affect the volume of this subfield.
METHODS Subjects Fifty healthy individuals were invited to participate in this cross-sectional study. They were recruited from the Unitat de Lıpids at the Hospital Clinic of Barcelona, Spain between 2011 and 2012. The selection of suitable participants consisted
405
of a semistructured interview aimed at collecting sociodemographic data as well as past and current clinical history. In addition, all subjects were submitted to extensive neuropsychological assessment. Exclusion criteria in the selection of participating individuals involved: (1) diagnosis of clinical dementia as assessed by the mini-mental state examination (MMSE) (Folstein et al., 1975) or according to DSM-IV criteria (2003); (2) subjective complaints of memory difficulties or any other cognitive deficits, interfering or not with daily living activities; (3) history or clinical evidence of psychiatric or neurological disorders; (4) major medical illnesses (i.e. uncontrolled hypertension or diabetes mellitus, metabolic syndrome, heart, liver or renal failure, or history of cancer); (5) past or current alcohol or drug abuse; (6) sensory loss interfering with testing procedures. All MRI scans were examined to exclude potential brain abnormalities and microvascular lesions as apparent in conventional FLAIR or T2-weighted images. Anthropometric and blood pressure measurements were performed in all subjects using standard methods. Hypertension was defined as blood pressure above 140/90 mm Hg, selfreported hypertension diagnosis or use of anti-hypertensive medication. Fasting blood samples were collected from all subjects and lipids were measured in serum by standard enzymatic methods. The apolipoprotein (APO) E genotype was determined by using the method of Hixson and Vernier (1990). This study was approved by the local Institutional Review Board of Barcelona, Spain. Written informed consent was obtained from all subjects participating in this study.
MRI Protocol All subjects underwent the same imaging protocol at the Center of Imaging Diagnosis (CDI) of Hospital Clinic (Barcelona, Spain) using a 3.0 T Magnetom Trio Tim Siemens (Erlangen, Germany). First, T1-weighted structural images were acquired in the sagittal plane using the following parameters: Repetition Time (TR) 5 2,300 ms; Echo Time (TE) 5 2.98 ms; Inversion Time (TI) 5 900 ms; Flip angle: 9 ; 256 3 256 matrix; 1 3 1 3 1 mm3 voxel. Then, diffusion weighted images were acquired using spin echo-planar imaging (TR 5 7,700 ms; TE 5 89 ms; Flip angle; 90 ; 2-mm thickness; 122 3 122 matrix; 2 3 2 3 2 mm3 voxel size; NEX 5 2) with 30 noncollinear directions for the diffusion sensitizing gradients at a b-value of 1,000 s/mm2 and one b 5 0 image. The SNR was 78.6 6 10.5, calculated based on previously published methods (Choi et al., 2011). Finally, FLAIR images were acquired using the parameters: TR 5 9,000 ms; TE 5 96 ms; TI 5 2,500 ms; flip angle: 150 ; 256 3 171 matrix; 3 3 3 3 3 mm3 voxel size. All images were visually inspected prior to pre-processing to exclude potential susceptibility artifacts in the hippocampal area or around it using previously published procedures (Simmons et al., 2009, 2010).
Preprocessing T1-weighted images were preprocessed using the FreeSurfer software package (version 5.1, available at http://surfer.nmr. Hippocampus
406
PEREIRA ET AL.
FIGURE 1. Automated segmentation of hippocampal subfields. Sagittal and coronal views of hippocampal subfields in a magnetic resonance image from a subject of the sample. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
mgh.harvard.edu/). Briefly, the implemented processing stream involved: removal of nonbrain tissue using a hybrid watershed algorithm (Segonne et al., 2004), automated transformation to the Talairach reference space and segmentation of the subcortical white matter and deep gray matter (Fischl et al., 2002, 2004). The whole hippocampal formation was segmented using the standard segmentation procedure using a probabilistic brain atlas. Visual inspection and quality control of these segmentations was carried out for all subjects. The estimated intracranial volume (ICV) was also calculated based on an atlas normalization procedure for head size correction, which used manual total ICV measurement as a reference (Buckner et al., 2004). Then, the automated subfield segmentation of the hippocampus was performed using Bayesian inference and a probabilistic atlas of the hippocampal formation based on manual delineations of subfields in ultra-high T1-weighted scans from several subjects. Seven hippocampal subfields volumes were calculated including the fimbria (white matter, WM), presubiculum, subiculum, CA1, CA2-3, CA4-DG fields (gray matter, GM) as well as the hippocampal fissure (cerebrospinal fluid, CSF). The procedures for subfield segmentation have been described elsewhere (van Leemput et al., 2009; Hanseeuw et al., 2011). Briefly, these procedures consisted of delineating the medial border of CA1 by uniting the highest curvature points on the dorsolateral and ventrolateral edges of the hippocampus, which also defined the lateral boundary of CA2-3, CA4-DG and the subiculum. The subiculum was separated from the presubiculum through a vertical line drawn from the most medial edge of the fimbria and its dorsal boundary was drawn along the subicular clouds and the edge of the hippocampal fissure. The lateral boundary of CA1 and the dorsal boundary of CA2-3 was the border between the hippocampus and the temporal horn of the lateral ventricle. A line through the intersection of two points of maximum curvature and perpendicular to the medial boundary of CA1 formed the ventral border of CA2-3 and dorsal boundary for CA4-DG. Finally, the fimbria was defined as the dorsolateral white matter along the whole extent of the hippocampal formation. Figure 1 shows the subfield segmentation results for one of the subjects of the current study. Hippocampus
Regarding DTI, image distortions produced by motion effects and eddy currents were corrected by applying a 3D full affine alignment of each image to the mean diffusion weighting b 5 0 image using FSL (FMRIB Software Library, FMRIB, Oxford, UK) (Smith et al., 2004). DTI data were then averaged and concatenated into 31 (1 b0 1 30 b1000) volumes. To obtain individual MD and FA maps, a diffusion tensor model was fit at each voxel (Behrens et al., 2003). The individual FA maps of each subject were registered to the individual T1weighted volumes that were previously brain extracted using BET (Jenkinson et al., 2005). This registration was performed using FLIRT with an affine correlation ratio cost function alignment with the nearest-neighbor option (Jenkinson and Smith, 2001; Jenkinson et al., 2002). The transformation matrix obtained from the registration of FA maps to the T1weighted images was applied to the MD maps with the same resampling options. In this way, the FA and MD maps were resampled to the resolution of the T1 images. The mean MD and FA values for all hippocampal regions of each subject were extracted using FSL tools. This procedure has been previously implemented by other studies assessing DTI measures in the hippocampus (Cherubini et al., 2009; Carlesimo et al., 2010; Peran et al., 2010). To ensure that the MD and FA maps were well coregistered to the individual T1-weighted volumes, we created two binarized masks: one representing the sum of all white matter regions and the other as the sum of all CSF compartments including the choroid fissure. These regions were obtained from the FreeSurfer segmentation procedure and automated parcellation of the cerebral cortex of each individual subject (Desikan et al., 2006). These white matter and CSF masks were superimposed on the brain extracted T1 volumes and hippocampal subfields of each subject and detailed visual inspection was performed to ensure there was no overlap between the masks and the subfields. The only exception was the hippocampal white matter band or fimbria, for which only the CSF mask was superimposed. In addition, we also assessed the impact of potential partial volume CSF contamination on the MD and FA measures of hippocampal regions by contracting
EFFECTS OF AGE ON VOLUME, MD, AND FA IN HIPPOCAMPAL SUBFIELDS TABLE 1. Demographic and Clinical Characteristics of the Study Sample N 5 50 Mean age 50–65 years 66–75 years Sex (m/f) Education (years) MMSE ApoE e4a Hypertension LDL-C (150 mg/dl)
63.7 6 7.0 23 (46%) 27 (54%) 20/30 10.9 6 3.9 28.9 6 1.2 10 (20%) 8 (16%) 17 (34%)
a
Number of subjects carrying at least one e4 allele. The values are mean (standard deviation) or number (%). MMSE, mini-mental state examination; ApoE e4, apolipoprotein e4; LDL-C, low-density lipoprotein cholesterol.
the hippocampus segmentation on the interface with the CSF using a previously described procedure (den Heijer et al., 2012). In particular, a dilation of the CSF mask of each subject was performed with a spherical structuring element that expanded the CSF mask with one voxel in all directions. We subtracted this dilated CSF mask segmentation from each hippocampal region, measured again the MD and FA values and repeated the analyses with these shrunken hippocampal segmentations.
Statistical Analyses No errors were detected in any of the automated FreeSurfer segmentation outputs upon visual quality control. After excluding the presence of significant interactions between age and hemisphere (Supporting Information Table 1), the average of left and right subfield volumes, MD, and FA were used for all statistical analyses. In addition, volume measures of the whole hippocampus and hippocampal subfields were adjusted by ICV and normative measures of these volumes were used for subsequent analyses. Adjustment for ICV was performed by using the following formula: Volumeadjusted 5 Volumeobserved 2 b [slope from ICV vs. regional volume regression] 3 (ICVobserved 2 ICVsample mean) (Jack et al., 1989). After confirming the normality of all variables using the Kolmogorov-Smirnov test, parametric statistical tests were selected to test this study’s hypotheses. Within each subfield or the whole hippocampus, cross correlations between the three anatomical measures (volume, MD, and FA) were assessed using Pearson’s correlation coefficients to evaluate whether a change in macrostructural shrinkage, microscopic obstacles to diffusion and microscopic axonal orientation were affiliated phenomena or independent from each other. To evaluate the contribution of age to explain the variability in volume, MD, or FA of hippocampal regions, we performed linear regression analyses. In these analyses, age was included as the independent variable and volume, MD, or FA of each
407
region was included as the dependent variable. The measures that came out significant from these analyses were included in a step-wise multiple regression model that assessed which were the best overall predictors of age in the hippocampus. This model included age as the dependent variable and all the significant anatomical measures as predictors. In all regression analyses, the goodness of fit was assessed by the R coefficient. Finally, to evaluate the potential effect of genetic and clinical variables in our study, multivariate ANOVAs were carried out to compare hippocampal volumes, MD, and FA between ApoE e4 carriers and e4 noncarriers, subjects with and without hypertension, subjects with high levels of low-density lipoprotein cholesterol (LDL-C 150) and low levels (LDL-C < 150), and between men and women. To avoid type I errors or accepting false positive results derived from performing multiple correlation and regression analyses, a Bonferroni adjusted level of significance (0.05/22 subfields 5 0.002) was applied to all analyses. All statistical analyses were carried out using SPSS version 20.0.
RESULTS The clinical characteristics of the study population are shown in Table 1. There were 10 subjects that carried at least 1 ApoE e4 allele (20%), 8 that were being treated for hypertension (16%) and 17 with high LDL-C levels (34%). The number of voxels, volumes and DTI values of each hippocampal region for the whole sample are presented in Table 2. Regarding the cross-correlations between volume, MD, and FA, we observed a negative correlation between MD and FA in the presubiculum (r 5 20.748, P < 0.001) as well as in the subiculum (r 5 20.471, P < 0.001) (Table 3). No other significant correlations were found between volume or DTI measures in the same hippocampal region after correcting for multiple comparisons. To assess the effects of age on volume, MD, or FA of each hippocampal region, we performed linear regression analyses. The results showed that, in the whole hippocampus, age was a significant predictor of gray matter volume (R 5 0.443, t 5 23.422, P < 0.001) and MD (R 5 0.588, t 5 5.032, P < 0.001). Within the specific hippocampal subfields, age significantly predicted variability associated with volume in CA2-3 (R 5 0.588, t 5 5.032, P < 0.001) and CA4-DG (R 5 0.443, t 5 23.422, P < 0.001). Finally, a significant age effect was found in MD in CA2-3 (R 5 0.475, t 5 3.774, P < 0.001) and fimbria (R 5 0.647, t 5 5.877, P < 0.001) (Figure 2, Table 4) as well as FA in the subiculum (R 5 0.438, t 5 23.373, P < 0.001) (Figure 2, Table 4). These results remained significant after excluding CSF partial volume voxels from reduced hippocampal subfields masks (Supporting Information Table 2). When the significant results from the previous linear regression analyses where included in a step-wise multiple regression model to assess which were the best overall predictors of age in Hippocampus
408
PEREIRA ET AL.
TABLE 2. Voxel Number, Volume, MD, and FA of Hippocampal Regions Number of voxels Whole hippocampus CA 1 CA2–3 CA4-DG Presubiculum Subiculum Fimbria Hipp. fissure
5465.6 533.3 1638.0 765.8 826.9 898.4 134.3 71.7
Volume
(650.3) (76.8) (224.5) (108.1) (106.0) (117.6) (27.8) (28.2)
3187.8 311.7 921.2 513.9 426.3 595.5 57.8 45.4
(289.0) (32.1) (93.7) (51.8) (52.2) (63.2) (14.5) (16.9)
MD 1.03 1.21 1.08 1.04 0.96 0.84 1.09
(0.05) (0.14) (0.06) (0.06) (0.09) (0.04) (0.14) –
FA 0.19 0.19 0.17 0.14 0.23 0.19 0.32
(0.02) (0.02) (0.02) (0.02) (0.05) (0.03) (0.05) –
Data are presented in means followed by standard deviations. Volume data are shown in mm3 and MD values are presented as (31023). MD, mean diffusivity; FA, fractional anisotropy; CA, cornu armonis; DG, dentate gyrus; Hipp, hippocampal.
the hippocampus, we found that MD in the fimbria (t 5 5.725, P < 0.001) and CA2-3 volume (t 5 23.608, P < 0.001) were selected as the best predictors, explaining 73.8% of the age variance. We also performed partial correlation analyses between age and DTI metrics in hippocampal subfields, while controlling for the corresponding subfields volumes to assess the unique contribution of age on integrity measures, independently of age-related atrophy. These analyses showed that the correlation between age and the fimbria’s MD remained highly significant (t 5 0.651, P < 0.001), while controlling for the fimbria’s volume. In addition, the correlation between age and MD in CA2-3 decreased significance when CA2-3 volume was included as a covariate (t 5 0.354, P < 0.013), as did the correlation between age and the subiculum’s FA when the subiculum’s volume was controlled for (t 5 20.383, P < 0.007). These results indicate that of all integrity measures, the fimbria’s MD was the only one to show an age-related effect in hippocampal subfields that was independent of volume atrophy. TABLE 3. Correlations Between the Anatomical Measures Within Each Subfield Volume vs. MD r Whole hippocampus CA 1 CA2–3 CA4-DG Presubiculum Subiculum Fimbria
20.385 20.363 20.398 0.165 20.244 20.151 20.266
P
Volume vs. FA r
P
MD vs. FA r
P
0.006 0.239 0.094 20.371 0.008 0.010 0.073 0.616 20.099 0.496 0.004 20.144 0.318 0.221 0.123 0.251 0.108 0.454 20.039 0.789 0.088 0.052 0.718 20.748
Regional Vulnerability of Hippocampal Subfields to Aging Measured by Structural and Diffusion MRI Joana B. Pereira,1,2,3 Cinta Valls-Pedret,4,5 Emilio Ros,4,5 Eva Palacios,1,2 Carles Falc on,2,6 Nuria Bargall o,2,7 David Bartres-Faz,1,2* Lars-Olof Wahlund,3 Eric Westman,3 and Carme Junque1,2
Abstract: In the past few years, there has been an increasing awareness of the regional vulnerability of the hippocampus to age-related processes. However, to date, no studies have assessed the effects of age on different structural magnetic resonance parameters in the specific hippocampal subfields. In this study, we measured volume, mean diffusivity (MD) and fractional anisotropy (FA) in the presubiculum, subiculum, fimbria, cornu ammonis (CA) 1,2-3,4-DG and the whole hippocampus in fifty cognitively intact elder adults between 50 and 75 years of age (20 men, 30 women). Segmentation of hippocampal subfields was performed using FreeSurfer. Individual MD and FA images were coregistered to T1-weighted volumes using FLIRT of FSL. Linear regression analyses were performed to assess the effects of age on the anatomical measures of each subfield. In addition, multiple regression analyses were also carried out to assess which of the anatomical measures that showed a correlation with age in the previous analyses, were the best age predictors in the hippocampus. In agreement with previous studies, our results showed a significant association between age and volume (P < 0.001) as well as MD (P < 0.001) in the whole hippocampus. Regarding the specific hippocampal subfields, we found that age had a significant negative effect on volume in CA2-3 (P < 0.001) and CA4-DG (P < 0.001). Importantly, we found a positive effect of age on MD in CA2-3 (P < 0.001) and fimbria (P < 0.001) as well as a negative age effect on FA in the subiculum (P < 0.001). Multiple regression analyses revealed that the best overall predictors of age in the hippocampus were MD in the fimbria and volume of CA2-3, which explained 73.8% of the age variance. These results indicate that age has an effect both on volume and diffusion tensor imaging measures in different subfields, suggesting they provide complementary information on age-related C 2013 Wiley Periodicals, Inc. processes in the hippocampus. V
1
Department of Psychiatry and Clinical Psychobiology, University of Barcelona, Spain; 2 Institute of Biomedical Research August Pi i Sunyer (IDIBAPS), Barcelona, Spain; 3 Department of Neurobiology, Care Sciences and Society, Karolinska Institutet, Stockholm, Sweden; 4 Lipid Clinic, Endocrinology and Nutrition Service, Institut d’Investigacions Biomediques August Pi Sunyer (IDIBAPS), Hospital Clınic, Barcelona, n (CIBERobn), Spain; 5 CIBER Fisiopatologıa de la Obesidad y Nutricio Instituto de Salud Carlos III (ISCIII), Barcelona, Spain; 6 CIBER-BBN, stic per la Imatge Hospital Clinic Barcelona, Spain; 7 Centre de Diagno de Barcelona (CDIC), Barcelona, Spain Additional Supporting Information may be found in the online version of this article. Grant sponsor: Spanish Ministerio de Ciencia e Innovaci on; Grant number: SAF2009-07489; Grant sponsor: Ministerio de Economıa y Competitividad; Grant number: PSI2012-38257; Grant sponsor: Ministerio de Sanidad, Polıtica Social e Igualdad; Grant number: IMSERSO 200/2011. *Correspondence to: David Bartres-Faz, Departament de Psiquiatria i Psicobiologia Clınica, Universitat de Barcelona, IDIBAPS, Casanova 143, 08036 Barcelona, Spain. E-mail: [email protected] Accepted for publication 6 December 2013. DOI 10.1002/hipo.22234 Published online 12 December 2013 in Wiley Online Library (wileyonlinelibrary.com). C 2013 WILEY PERIODICALS, INC. V
KEY WORDS: volume; mean diffusivity; fractional anisotropy; age; hippocampus
INTRODUCTION The hippocampus is one of the most well studied cerebral structures in healthy aging due to its high susceptibility to degenerative processes. However, there is still debate on whether hippocampal volume loss is related to increasing age. Some studies have reported significant volume loss (Fjell et al., 2010; Zhang et al., 2010; Walhovd et al., 2011), while others have found that hippocampal volumes remain quite stable over aging (Head et al., 2005; Sullivan et al., 2005). These discrepancies have been related to age differences in the studied samples as studies showing hippocampal reductions with aging included elderly subjects and found these changes only take place in late adulthood. In line with this, recent longitudinal studies assessing the trajectories of regional brain volumes over aging reported that the hippocampus shows an accelerated volume decline with age rather than a linear decline (Raz et al., 2010; Pfefferbaum et al., 2013), particularly after 60 years (Pfefferbaum et al., 2013). Although most neuroimaging studies have assessed the hippocampus as a single structure, it is actually a complex circuit made up of functionally and molecularly distinct subfields (Small et al., 2011) comprising the subiculum (with the subdivisions pre-subiculum, para-subiculum and subiculum proper), the four cornu ammonis (CA1-4) and the dentate gyrus (DG) (Duvernoy, 1998). The increasing awareness of regional vulnerability within the hippocampus to disease and age-related processes has led researchers to measure these different hippocampal subfields rather than the hippocampus as a whole. For instance, using a manual technique on highresolution 4T MRI scans a significant association was found between increasing age and volume loss in CA1 (Mueller et al., 2007), CA3 and the DG (Mueller et al., 2008). In a subsequent study by the same group a negative effect of age was found in the CA3
404
PEREIRA ET AL.
and DG subfields (Mueller et al., 2009), providing consistency to their previous findings (Mueller et al., 2008). Using a surface-based method to map age-related changes to the shape of the hippocampus, Frisoni et al. (2008) found atrophy associated with age in CA1 and the presubiculum. By applying a method based on voxel-based morphometry, La Joie et al. (2010) found a negative correlation between age and the volume of the subiculum in healthy individuals between 19 and 68 years of age. In addition to studies assessing gray matter volume in the hippocampus, the application of diffusion tensor imaging (DTI) for measuring microstructural changes in deep gray matter structures has recently received a lot of interest. DTI is a MRI technique with an unparalleled sensitivity to water movements and a potential in probing neuroanatomy at a microscopic level that is well beyond image resolution (Le Bihan et al., 2001). Two quantitative parameters derived from DTI have been shown to be greatly influenced by physiological aging (Cherubini et al., 2009). One of them is mean diffusivity (MD), which characterizes the overall displacement of water molecules and the overall presence of obstacles to diffusion. The other is fractional anisotropy (FA), which describes how much molecular displacements vary in space and is related to the presence and coherence or oriented structures (Le Bihan et al., 2001). Increases in MD and reductions of FA have been described in several brain areas and structures in elder individuals (Hasan et al., 2007; Abe et al., 2008; Jeon et al., 2012). Previous histological studies suggest that these changes are related to the degradation of the axonal membranes, myelin sheaths (Sen and Basser, 2005) and general loss of ordered anisotropic tissue structures (Pierpaoli and Basser, 1996). These phenomena increase the extracellular volume fraction and allow larger net displacements of water molecules (Sodak, 2004) in all directions in the same amount. Moreover, loss of intracellular viscosity, macromolecular crowding (Hazlewood et al., 1991) and extracellular tortuosity have also been shown to affect diffusion (Chen and Nicolson, 2000; Norris, 2001). In both white and gray matter, MD is an indicator of microscopic barriers or obstacles to diffusion. The loss of microscopic barriers in white matter is associated with axonal degeneration and/or dysmyelination, while in gray matter is most likely related to axonal degeneration due to the relative absence of myelination in this brain tissue. In white matter, FA reflects the integrity of large fiber bundles such as the association, commissural and projection fibers. By contrast, in gray matter FA is harder to interpret due to the existence of multiple crossing axonal fibers that lower the FA values in this tissue compared to white matter. Despite of this, several studies have shown that FA in gray matter regions offer valuable information, suggesting they also reflect the integrity of oriented structures in gray matter including unmyelinated axons. For instance, previous studies have shown FA increases in the putamen and caudate in elder compared to younger individuals (Pfefferbaum et al., 2010) as well as in patients with epilepsy (Gerdes et al., 2012) and Huntington’s disease (Bohanna et al., 2011). In the hippocampus, previous studies showed significant Hippocampus
FA decreases in patients with Alzheimer’s disease (AD; Fellgiebel et al., 2004; Muller et al., 2007) or a negative correlation between hippocampal FA and age in elder individuals (Carlesimo et al., 2010; Heijer et al., 2012), suggesting that FA is sensitive to microstructural orientation in subcortical gray matter. Changes in DTI measures have been suggested to occur before neuronal degradation and atrophy are detectable at a macroscopic level (Muller et al., 2007; Heijer et al., 2012). In two recent studies comparing gray matter volume and DTI parameters in the whole hippocampus, the correlation coefficients between age and MD were higher than those between age and hippocampal volume and FA (Carlesimo et al., 2010; den Heijer et al., 2012). Reasons that may account for the higher sensitivity of MD to age compared to volume include the fact that it assesses subtle microstructural pathological alterations including the density of physical obstructions such as cellular membranes and the distribution of water molecules between cellular compartments (Beaulieu et al., 2002; Sen and Basser, 2005). By contrast, volume detects abnormalities only when significant tissue shrinkage at the macrostructural level has taken place, possibly due to reductions in neuropil, neuronal size, dendritic or axonal arborization (Mechelli et al., 2005). On the other hand, the higher sensitivity of MD to age compared to FA in the hippocampus could be related to the fact that MD is less affected by fiber crossing than FA because it reflects the magnitude of water diffusion, which is not influenced by direction (Pierpaoli et al., 1996). By contrast, FA is a measure of the directional dependence of diffusion (Basser et al., 1995), which is lower in gray matter structures such as the hippocampus due to the presence of multiple crossing fibers with different directions and lower intrahippocampal fiber coherence (van Norden et al., 2012). Despite providing relevant information in the whole hippocampus, to date no studies have assessed DTI measures within its specific subfields. Hence, the goal of the current study was to assess volume, MD, and FA in hippocampal subfields and test their association with age in a sample of nondemented older adults. First of all, we assessed the effects of age on both volume and DTI measures in the whole hippocampus to assess whether our findings agreed with those of previous studies (Carlesimo et al., 2010; den Heijer et al., 2012). Second, we evaluated the association between age and hippocampal subfields volumes using a novel neuroimaging technique developed by van Leemput et al. (2009) in ultra-high resolution MRI data, which can be applied to standard resolution MRI at 1.5 or 3 T (Engvig et al., 2012; Kuhn et al., 2012; Teicher et al., 2012; Pereira et al., 2013). In addition to measuring the volumes of CA1–4, DG and subiculum, this method also measures the white matter hippocampal band or fimbria. Moreover, it differentiates the presubiculum from the subiculum proper and delineates the hippocampal fissure, offering a more complete overview of hippocampal subregions than previous studies assessing hippocampal subfields volumes (Mueller et al., 2007, 2009; La Joie et al., 2010). Finally, we tested the association between age and MD as well FA in hippocampal subfields,
EFFECTS OF AGE ON VOLUME, MD, AND FA IN HIPPOCAMPAL SUBFIELDS which is the main novelty of our study. Based on previous studies showing a higher correlation between age and MD in the whole hippocampus (Carlesimo et al., 2010; den Heijer et al., 2012), we hypothesized that age would have a stronger effect on MD in the whole hippocampus compared to volume and FA in our study. Regarding the specific subfields, we expected a greater age effect in both volume and DTI measures in CA1, CA3, and DG compared to the other subregions. This hypothesis was based on previous studies showing a correlation between age and the volumes of CA1, CA3, and DG in a sample with a similar age range to ours (Mueller et al., 2007, 2009). Although previous studies did not include DTI subfields measures (Mueller et al., 2007, 2009), we predicted that if age correlated with macrostructural volume shrinkage in CA1, CA3, and DG then it would also strongly correlate with microstructural abnormalities in the same subregions given that, as previously suggested, changes in DTI measures occur before neuronal degradation and atrophy are detectable at a macroscopic level (Muller et al., 2007; Heijer et al., 2012). There are several reasons why volume and DTI measures might be sensitive to age effects in the previous subfields. For instance, the high expression of NMDA receptors in CA1 makes this subfield vulnerable to excitotoxicity due to hypoxia and ischemia associated with vascular problems that occur over aging (Small et al., 2011). These vascular problems can produce variations in extracellular water diffusion as well as decreases in FA (Sotak et al., 2002). The high superoxide levels in CA1 may also lead to greater susceptibility of these neurons to an excess in oxidative stress that can occur during aging and cause neuronal death (Nicolle et al., 2001; Wang et al., 2005), which would affect both DTI and volume measures in this region. Regarding CA3, the age-related reductions in the number and area of the large pyramidal cells (Kadar et al., 1998) as well as dendritic tree atrophy (Fuchs et al., 2001) could shrunk the volume of this subfield, while more subtle alterations in perivascular morphology, increased accumulation of lipofuscin in the neuropil (Ojo et al., 2013) and loss of mossy fiber densities (Stephens et al., 2011) could also potentially induce microstructural changes in DTI metrics. Finally, the DG is known for its role in adult neurogenesis, which decreases during aging, leading to a decline of production of new hippocampal neurons (Small et al., 2011). In particular, the high levels of mineralocorticoid receptors confer vulnerability of this subfield to the effects of circulating corticosteroids, which can suppress cell proliferation (Brummelte and Galea, 2010) and affect the volume of this subfield.
METHODS Subjects Fifty healthy individuals were invited to participate in this cross-sectional study. They were recruited from the Unitat de Lıpids at the Hospital Clinic of Barcelona, Spain between 2011 and 2012. The selection of suitable participants consisted
405
of a semistructured interview aimed at collecting sociodemographic data as well as past and current clinical history. In addition, all subjects were submitted to extensive neuropsychological assessment. Exclusion criteria in the selection of participating individuals involved: (1) diagnosis of clinical dementia as assessed by the mini-mental state examination (MMSE) (Folstein et al., 1975) or according to DSM-IV criteria (2003); (2) subjective complaints of memory difficulties or any other cognitive deficits, interfering or not with daily living activities; (3) history or clinical evidence of psychiatric or neurological disorders; (4) major medical illnesses (i.e. uncontrolled hypertension or diabetes mellitus, metabolic syndrome, heart, liver or renal failure, or history of cancer); (5) past or current alcohol or drug abuse; (6) sensory loss interfering with testing procedures. All MRI scans were examined to exclude potential brain abnormalities and microvascular lesions as apparent in conventional FLAIR or T2-weighted images. Anthropometric and blood pressure measurements were performed in all subjects using standard methods. Hypertension was defined as blood pressure above 140/90 mm Hg, selfreported hypertension diagnosis or use of anti-hypertensive medication. Fasting blood samples were collected from all subjects and lipids were measured in serum by standard enzymatic methods. The apolipoprotein (APO) E genotype was determined by using the method of Hixson and Vernier (1990). This study was approved by the local Institutional Review Board of Barcelona, Spain. Written informed consent was obtained from all subjects participating in this study.
MRI Protocol All subjects underwent the same imaging protocol at the Center of Imaging Diagnosis (CDI) of Hospital Clinic (Barcelona, Spain) using a 3.0 T Magnetom Trio Tim Siemens (Erlangen, Germany). First, T1-weighted structural images were acquired in the sagittal plane using the following parameters: Repetition Time (TR) 5 2,300 ms; Echo Time (TE) 5 2.98 ms; Inversion Time (TI) 5 900 ms; Flip angle: 9 ; 256 3 256 matrix; 1 3 1 3 1 mm3 voxel. Then, diffusion weighted images were acquired using spin echo-planar imaging (TR 5 7,700 ms; TE 5 89 ms; Flip angle; 90 ; 2-mm thickness; 122 3 122 matrix; 2 3 2 3 2 mm3 voxel size; NEX 5 2) with 30 noncollinear directions for the diffusion sensitizing gradients at a b-value of 1,000 s/mm2 and one b 5 0 image. The SNR was 78.6 6 10.5, calculated based on previously published methods (Choi et al., 2011). Finally, FLAIR images were acquired using the parameters: TR 5 9,000 ms; TE 5 96 ms; TI 5 2,500 ms; flip angle: 150 ; 256 3 171 matrix; 3 3 3 3 3 mm3 voxel size. All images were visually inspected prior to pre-processing to exclude potential susceptibility artifacts in the hippocampal area or around it using previously published procedures (Simmons et al., 2009, 2010).
Preprocessing T1-weighted images were preprocessed using the FreeSurfer software package (version 5.1, available at http://surfer.nmr. Hippocampus
406
PEREIRA ET AL.
FIGURE 1. Automated segmentation of hippocampal subfields. Sagittal and coronal views of hippocampal subfields in a magnetic resonance image from a subject of the sample. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
mgh.harvard.edu/). Briefly, the implemented processing stream involved: removal of nonbrain tissue using a hybrid watershed algorithm (Segonne et al., 2004), automated transformation to the Talairach reference space and segmentation of the subcortical white matter and deep gray matter (Fischl et al., 2002, 2004). The whole hippocampal formation was segmented using the standard segmentation procedure using a probabilistic brain atlas. Visual inspection and quality control of these segmentations was carried out for all subjects. The estimated intracranial volume (ICV) was also calculated based on an atlas normalization procedure for head size correction, which used manual total ICV measurement as a reference (Buckner et al., 2004). Then, the automated subfield segmentation of the hippocampus was performed using Bayesian inference and a probabilistic atlas of the hippocampal formation based on manual delineations of subfields in ultra-high T1-weighted scans from several subjects. Seven hippocampal subfields volumes were calculated including the fimbria (white matter, WM), presubiculum, subiculum, CA1, CA2-3, CA4-DG fields (gray matter, GM) as well as the hippocampal fissure (cerebrospinal fluid, CSF). The procedures for subfield segmentation have been described elsewhere (van Leemput et al., 2009; Hanseeuw et al., 2011). Briefly, these procedures consisted of delineating the medial border of CA1 by uniting the highest curvature points on the dorsolateral and ventrolateral edges of the hippocampus, which also defined the lateral boundary of CA2-3, CA4-DG and the subiculum. The subiculum was separated from the presubiculum through a vertical line drawn from the most medial edge of the fimbria and its dorsal boundary was drawn along the subicular clouds and the edge of the hippocampal fissure. The lateral boundary of CA1 and the dorsal boundary of CA2-3 was the border between the hippocampus and the temporal horn of the lateral ventricle. A line through the intersection of two points of maximum curvature and perpendicular to the medial boundary of CA1 formed the ventral border of CA2-3 and dorsal boundary for CA4-DG. Finally, the fimbria was defined as the dorsolateral white matter along the whole extent of the hippocampal formation. Figure 1 shows the subfield segmentation results for one of the subjects of the current study. Hippocampus
Regarding DTI, image distortions produced by motion effects and eddy currents were corrected by applying a 3D full affine alignment of each image to the mean diffusion weighting b 5 0 image using FSL (FMRIB Software Library, FMRIB, Oxford, UK) (Smith et al., 2004). DTI data were then averaged and concatenated into 31 (1 b0 1 30 b1000) volumes. To obtain individual MD and FA maps, a diffusion tensor model was fit at each voxel (Behrens et al., 2003). The individual FA maps of each subject were registered to the individual T1weighted volumes that were previously brain extracted using BET (Jenkinson et al., 2005). This registration was performed using FLIRT with an affine correlation ratio cost function alignment with the nearest-neighbor option (Jenkinson and Smith, 2001; Jenkinson et al., 2002). The transformation matrix obtained from the registration of FA maps to the T1weighted images was applied to the MD maps with the same resampling options. In this way, the FA and MD maps were resampled to the resolution of the T1 images. The mean MD and FA values for all hippocampal regions of each subject were extracted using FSL tools. This procedure has been previously implemented by other studies assessing DTI measures in the hippocampus (Cherubini et al., 2009; Carlesimo et al., 2010; Peran et al., 2010). To ensure that the MD and FA maps were well coregistered to the individual T1-weighted volumes, we created two binarized masks: one representing the sum of all white matter regions and the other as the sum of all CSF compartments including the choroid fissure. These regions were obtained from the FreeSurfer segmentation procedure and automated parcellation of the cerebral cortex of each individual subject (Desikan et al., 2006). These white matter and CSF masks were superimposed on the brain extracted T1 volumes and hippocampal subfields of each subject and detailed visual inspection was performed to ensure there was no overlap between the masks and the subfields. The only exception was the hippocampal white matter band or fimbria, for which only the CSF mask was superimposed. In addition, we also assessed the impact of potential partial volume CSF contamination on the MD and FA measures of hippocampal regions by contracting
EFFECTS OF AGE ON VOLUME, MD, AND FA IN HIPPOCAMPAL SUBFIELDS TABLE 1. Demographic and Clinical Characteristics of the Study Sample N 5 50 Mean age 50–65 years 66–75 years Sex (m/f) Education (years) MMSE ApoE e4a Hypertension LDL-C (150 mg/dl)
63.7 6 7.0 23 (46%) 27 (54%) 20/30 10.9 6 3.9 28.9 6 1.2 10 (20%) 8 (16%) 17 (34%)
a
Number of subjects carrying at least one e4 allele. The values are mean (standard deviation) or number (%). MMSE, mini-mental state examination; ApoE e4, apolipoprotein e4; LDL-C, low-density lipoprotein cholesterol.
the hippocampus segmentation on the interface with the CSF using a previously described procedure (den Heijer et al., 2012). In particular, a dilation of the CSF mask of each subject was performed with a spherical structuring element that expanded the CSF mask with one voxel in all directions. We subtracted this dilated CSF mask segmentation from each hippocampal region, measured again the MD and FA values and repeated the analyses with these shrunken hippocampal segmentations.
Statistical Analyses No errors were detected in any of the automated FreeSurfer segmentation outputs upon visual quality control. After excluding the presence of significant interactions between age and hemisphere (Supporting Information Table 1), the average of left and right subfield volumes, MD, and FA were used for all statistical analyses. In addition, volume measures of the whole hippocampus and hippocampal subfields were adjusted by ICV and normative measures of these volumes were used for subsequent analyses. Adjustment for ICV was performed by using the following formula: Volumeadjusted 5 Volumeobserved 2 b [slope from ICV vs. regional volume regression] 3 (ICVobserved 2 ICVsample mean) (Jack et al., 1989). After confirming the normality of all variables using the Kolmogorov-Smirnov test, parametric statistical tests were selected to test this study’s hypotheses. Within each subfield or the whole hippocampus, cross correlations between the three anatomical measures (volume, MD, and FA) were assessed using Pearson’s correlation coefficients to evaluate whether a change in macrostructural shrinkage, microscopic obstacles to diffusion and microscopic axonal orientation were affiliated phenomena or independent from each other. To evaluate the contribution of age to explain the variability in volume, MD, or FA of hippocampal regions, we performed linear regression analyses. In these analyses, age was included as the independent variable and volume, MD, or FA of each
407
region was included as the dependent variable. The measures that came out significant from these analyses were included in a step-wise multiple regression model that assessed which were the best overall predictors of age in the hippocampus. This model included age as the dependent variable and all the significant anatomical measures as predictors. In all regression analyses, the goodness of fit was assessed by the R coefficient. Finally, to evaluate the potential effect of genetic and clinical variables in our study, multivariate ANOVAs were carried out to compare hippocampal volumes, MD, and FA between ApoE e4 carriers and e4 noncarriers, subjects with and without hypertension, subjects with high levels of low-density lipoprotein cholesterol (LDL-C 150) and low levels (LDL-C < 150), and between men and women. To avoid type I errors or accepting false positive results derived from performing multiple correlation and regression analyses, a Bonferroni adjusted level of significance (0.05/22 subfields 5 0.002) was applied to all analyses. All statistical analyses were carried out using SPSS version 20.0.
RESULTS The clinical characteristics of the study population are shown in Table 1. There were 10 subjects that carried at least 1 ApoE e4 allele (20%), 8 that were being treated for hypertension (16%) and 17 with high LDL-C levels (34%). The number of voxels, volumes and DTI values of each hippocampal region for the whole sample are presented in Table 2. Regarding the cross-correlations between volume, MD, and FA, we observed a negative correlation between MD and FA in the presubiculum (r 5 20.748, P < 0.001) as well as in the subiculum (r 5 20.471, P < 0.001) (Table 3). No other significant correlations were found between volume or DTI measures in the same hippocampal region after correcting for multiple comparisons. To assess the effects of age on volume, MD, or FA of each hippocampal region, we performed linear regression analyses. The results showed that, in the whole hippocampus, age was a significant predictor of gray matter volume (R 5 0.443, t 5 23.422, P < 0.001) and MD (R 5 0.588, t 5 5.032, P < 0.001). Within the specific hippocampal subfields, age significantly predicted variability associated with volume in CA2-3 (R 5 0.588, t 5 5.032, P < 0.001) and CA4-DG (R 5 0.443, t 5 23.422, P < 0.001). Finally, a significant age effect was found in MD in CA2-3 (R 5 0.475, t 5 3.774, P < 0.001) and fimbria (R 5 0.647, t 5 5.877, P < 0.001) (Figure 2, Table 4) as well as FA in the subiculum (R 5 0.438, t 5 23.373, P < 0.001) (Figure 2, Table 4). These results remained significant after excluding CSF partial volume voxels from reduced hippocampal subfields masks (Supporting Information Table 2). When the significant results from the previous linear regression analyses where included in a step-wise multiple regression model to assess which were the best overall predictors of age in Hippocampus
408
PEREIRA ET AL.
TABLE 2. Voxel Number, Volume, MD, and FA of Hippocampal Regions Number of voxels Whole hippocampus CA 1 CA2–3 CA4-DG Presubiculum Subiculum Fimbria Hipp. fissure
5465.6 533.3 1638.0 765.8 826.9 898.4 134.3 71.7
Volume
(650.3) (76.8) (224.5) (108.1) (106.0) (117.6) (27.8) (28.2)
3187.8 311.7 921.2 513.9 426.3 595.5 57.8 45.4
(289.0) (32.1) (93.7) (51.8) (52.2) (63.2) (14.5) (16.9)
MD 1.03 1.21 1.08 1.04 0.96 0.84 1.09
(0.05) (0.14) (0.06) (0.06) (0.09) (0.04) (0.14) –
FA 0.19 0.19 0.17 0.14 0.23 0.19 0.32
(0.02) (0.02) (0.02) (0.02) (0.05) (0.03) (0.05) –
Data are presented in means followed by standard deviations. Volume data are shown in mm3 and MD values are presented as (31023). MD, mean diffusivity; FA, fractional anisotropy; CA, cornu armonis; DG, dentate gyrus; Hipp, hippocampal.
the hippocampus, we found that MD in the fimbria (t 5 5.725, P < 0.001) and CA2-3 volume (t 5 23.608, P < 0.001) were selected as the best predictors, explaining 73.8% of the age variance. We also performed partial correlation analyses between age and DTI metrics in hippocampal subfields, while controlling for the corresponding subfields volumes to assess the unique contribution of age on integrity measures, independently of age-related atrophy. These analyses showed that the correlation between age and the fimbria’s MD remained highly significant (t 5 0.651, P < 0.001), while controlling for the fimbria’s volume. In addition, the correlation between age and MD in CA2-3 decreased significance when CA2-3 volume was included as a covariate (t 5 0.354, P < 0.013), as did the correlation between age and the subiculum’s FA when the subiculum’s volume was controlled for (t 5 20.383, P < 0.007). These results indicate that of all integrity measures, the fimbria’s MD was the only one to show an age-related effect in hippocampal subfields that was independent of volume atrophy. TABLE 3. Correlations Between the Anatomical Measures Within Each Subfield Volume vs. MD r Whole hippocampus CA 1 CA2–3 CA4-DG Presubiculum Subiculum Fimbria
20.385 20.363 20.398 0.165 20.244 20.151 20.266
P
Volume vs. FA r
P
MD vs. FA r
P
0.006 0.239 0.094 20.371 0.008 0.010 0.073 0.616 20.099 0.496 0.004 20.144 0.318 0.221 0.123 0.251 0.108 0.454 20.039 0.789 0.088 0.052 0.718 20.748
