Subjects were 52 healthy volunteers, age 25–82 years, all right handed, 20 males and 32 females. Mean/median age was 51 years. Age distribution was approximately uniform across the age range. The local Tel Aviv-Sourasky Medical Center Helsinki Committee approved the research protocol, and all participants signed an informed consent. Subjects had no history of neurological diseases or psychological disorders, did not use neuropsychiatric medication, had no history of drug or alcohol abuse, and had intact vision. All subjects were right handed and had at least 15 years of education. Volunteers with common age-related diseases such as diabetes, high blood pressure, and high levels of cholesterol were not excluded from the cohort.

Participants underwent a series of computerized cognitive tests (Mindstreams, NeuroTrax Corp., Bellaire, TX) that evaluate performance across an array of cognitive domains known to deteriorate during aging (memory, executive function, visual spatial processing, verbal function, attention, information processing speed, and motor skills) and provide measurements of accuracy and reaction time (RT) shown to be valid and reliable (Russell, 2011). These tests have been used in clinical settings, as well as in studies of normal aging relating these neuropsychological measurements to, e.g., genetic findings (Thaler et al., 2012), psychiatric and neurodegenerative disorders (Doniger et al., 2006; Plotnik et al., 2011), and imaging parameters (Sasson et al., 2010, 2012). In the present study, we performed factor analysis of 44 cognitive outcome parameters. A detailed description of the cognitive tests used in the correlation analysis with DTI parameters appears below (in brief). A more detailed description of the cognitive tests was published previously (Sasson et al., 2012).

Factor analysis was performed using SPSS software (SPSS Inc., Chicago, IL, USA), as follows:

Factor analysis of 44 cognitive outcome parameter scores using principal component analysis (PCA) with orthogonal (VARIMAX) rotation and limiting the number of factors to 5 (although 10 factors obtained eigenvalues higher than 1) that explained 65% of the items total variance. The number of factors selected corresponded to the point of inflexion in the respective scree plot. The percentage of variance explained derives from that point.

(a) The executive function factor included scores from the Go-NoGo response inhibition test and Stroop interference test. The Go-NoGo test is a continuous performance test of simple RT and response inhibition. In the Stroop interference test performed here, the conflicting information provided by the meaning of a word and the color of its letters leads to a decrement in performance relative to the test phases where there is no conflict. In total, the test includes three phases—phase I (no interference: letter color), phase II (no interference: word meaning), and phase III (interference between letter color and word meaning). The executive function domain is an average of RT in phase III (interference), response time difference (RD) between RT in phase III and phase II, RT in phase I (no interference: letter color) and standard deviations (SD) in phase I (no interference: letter color).

MR imaging was performed on a 3T (GE) MRI system at the Tel Aviv-Sourasky Medical Center. The MRI protocol included conventional anatomy sequences and DTI acquired with a standard head-coil.

Correction of head motion image artifacts, normalization, and statistical analysis were performed using the SPM software (version 2, UCL, London, UK). The image analysis routine of the diffusion tensor data included the following steps:

All of the DWI images were coregistered using SPM2 (UCL, London, UK) to correct for head motion. Gradient orientations were compensated prior to the b-matrices calculation to account for the rotation component of the registration.

Tensor fitting was performed using the free water elimination method (Pasternak et al., 2009). The free water elimination method allows correcting for water contamination, especially in brain tissue adjacent to the lateral ventricles, for example the fornix. It is achieved by fitting a bi-tensor model for which a mathematical framework is introduced to stabilize the fitting. In this routine, we eliminate a tensor component that fits the diffusion properties of free water and use for tracking only the fiber component of the tensor. The free water elimination is important when focusing on elderly subjects as in the present work, since atrophy and enlarged ventricles might result in lower FA values in white matter structures such as the fornix, which is adjacent to the lateral ventricles.

The tensors obtained were spectrally decomposed to their eigen-components. The eigenvalues were used to calculate FA maps (Basser and Pierpaoli, 1998). Tractography was applied using the principal eigenvectors and FA: the brute force FACT algorithm was used to generate the fiber coordinates, terminating at voxels with FA lower than 0.2 or following tract orientation change higher than 60°. Fibers that passed through a manually chosen seed region of interest (ROI) were plotted. The fibers were plotted as streamlines using Matlab (Mathworks, Natick, MA). Once a subset of fibers had been found, a visitation map was generated, indicating voxels with at least one streamline passing through. The masks obtained were overlaid over the B₀ image. Overall 10 fibers tracts were plotted for each subject: five temporal projections (UF, fornix, cingulum, SLF, ILF), in both hemispheres.

Representative seed ROIs are shown in Figure 1, as well as the reconstructed fiber. The seed ROIs were chosen based on visual comparison with the MRI atlas of human white matter (Mori and Crain, 2005). The seed ROIs were drawn on the DTI color map for each temporal projection in the following manner:

The Cingulum: Three ROIs were used in order to reconstruct the cingulum bundle. Two seed ROIs were drawn in the mid-sagittal plane: first ROI within the body of cingulum (Figure 1A, ROI 1), second ROI in the posterior part of the cingulum (Figure 1A, ROI3). Another was drawn in a more lateral sagittal slice in the temporal part of the cingulum adjacent to the hippocampus (Figure 1A, ROI2). If necessary, in order to eliminate fibers which are not part of the cingulum, a “no-fiber” plane was drawn in a coronal plane posterior to the cingulum and in an axial plane dorsal to the cingulum (Figure 1A, no fiber).

After the tracking procedure, a mask was created from the tracts matrices of all subjects in order to create a tract mask. Overall 10 masks were created for each subject: five temporal projections (UF, fornix, cingulum, SLF, ILF), in both hemispheres, right and left. For each temporal projection, the tract masks (that are already normalized to the MNI template) of the different subjects were registered to a tract mask of one young subject. The registration procedure was performed using normalized mutual information (Studholme et al., 1999) in the SPM software (version 2, UCL, London, UK). Spatial registration involved finding parameters that either maximize or minimize the normalized mutual information function, followed by a voxel to voxel affine transformation.

Voxel-based analysis: VBA is a whole brain technique that allows the detection of regionally specific differences in brain tissue composition on a voxel-by-voxel basis (Ashburner and Friston, 2000). In this work we used SPM (version 2, UCL, London, UK) to perform voxel-by-voxel correlation between the DTI indices and performance in each of the cognitive domains. For each temporal projection, simple regression was performed between the DTI indices (ADC, FA, D_A, D_R) and age in each temporal projection and multiple regressions were performed between the DTI indices (ADC, FA, D_A, D_R) co-registered to the appropriate temporal projection, and scores in the four cognitive domains (executive function, information processing speed, and memory). The multiple regressions were controlled for age and motor ability using the multiple regression routine in SPM with age and scores in a simple finer tapping task were used as constancies. The fiber ROI was used as a mask. The statistical threshold was set at p < 0.05, after Hochberg sequential correction for multiple comparisons (Hochberg, 1988).

The multiple regression analysis generated statistical parametric maps. The multiple regression routine included applying implicit and explicit masks (the appropriate fiber ROI, see above), without applying grand mean scaling, threshold masking, or global calculation. Correction for multiple comparisons was performed on the statistical parametric map, and the five ROIs were used to constrain the analysis—reducing the multiple comparison correction to number of voxels contained within the 10 ROIs (5 tracts on L and R sides). The statistical parametric maps are presented superimposed on an average T1 image of all subjects, with the fiber ROIs drawn upon them allowing an anatomical informative reference.

Scatter plots of scores of the cognitive domains and the averaged DTI parameters extracted only in the voxels passing the significance threshold are presented. A linear fit (y = mx + b) was performed and the best linear fit is shown on the scatterplots.

A significant negative correlation was obtained in the right anterior cingulum between ADC and performance in the memory domain. This finding is consistent with previous reports of involvement of the anterior cingulate and cingulum in memory performance (Valenstein et al., 1987; Sepulcre et al., 2009; Charlton et al., 2010). A negative correlation was also obtained between ADC in the posterior cingulum with performance in the memory domain.

Lesion studies suggest that the fornix, which includes cholinergic afferents to the hippocampus, is important for learning and formation of new memories (Spiers et al., 2001).

Indeed in the present study we obtained that ADC in the right fornix is significantly negatively correlated with performance in the memory domain.

Various studies suggest an important role for the SLF in higher brain functions, and particularly language (Lichtheim, 1885; Damasio and Damasio, 1980; Tanabe et al., 1987; Catani et al., 2003; Wise, 2003; Geldmacher et al., 2007). We obtained a negative correlation between performances in the information processing speed domain, with bilateral FA in the SLF. The correlations were localized to parietal regions within the right SLF. The parietal lobe is well known from the literature to be related to calculation (Rocha et al., 2005; Cavanna and Trimble, 2006).

A strong significant positive correlation was also obtained between performance in the executive function domain and ADC in the left SLF. The Arcuate fasciculus (AF), a part of the SLF, has been central to the neurobiological interpretation of higher brain function generally and for language in particular (Lichtheim, 1885; Geschwind, 1965a,b; Heilman et al., 1970; Tanabe et al., 1987). The task we used here, stroop interference involves interference between semantic information and visual input, and hence involves linguistic abilities, and is also considered as a higher function task.

The ILF is known to play an important role in visual memory as demonstrated by post mortem studies (Bauer and Trobe, 1984) lesion studies (Shinoura et al., 2007) and imaging studies. In congenital prosopagnosia correlation was found between face recognition and FA in the ILF (Thomas et al., 2009; Tavor et al., 2010). In our work we found a significant positive correlation between memory performance and FA in the left ILF. The memory tasks were visual tasks and included visual spatial memory and visual verbal memory.

There was also negative correlation between FA in the left ILF, and performance in the information processing speed domain. There are some findings in the literature supporting an association between measurements of occipital white matter structures with information processing speed. Smaller occipital white matter volume was associated with slower information processing speed in schizophrenic patients (Antonova et al., 2005). Higher average fractional anisotropy (FA) in left occipitotemporal junction was obtained in children trained in mental calculation (Hu et al., 2011).

The UF plays an important role in the formation and retrieval of memories as shown by lesion (Levine et al., 1998) and animal studies (Squire et al., 2004). Indeed, we found a negative correlation between ADC in the right UF with performance in the memory domain. In a study using diffusion tensor imaging (DTI) (using tract based statistics—TBSS) increased axial diffusivity was found in the UF in patients with amnestic mild cognitive impairment (O'Dwyer et al., 2011). Lower white matter density was found in the UF (as well as in the fornix and anterior cingulum) in patients with hypoxic amnesia (Di Paola et al., 2011). Lesion studies have demonstrated that the frontal lobe is crucial for executive functions (Milner, 1971; Drewe, 1974; Perret, 1974; Rowe et al., 2001). Indeed, the ADC values in the UF which projects into the pre-frontal cortex were obtained in the present study to correlate with performance in the executive function domain. This finding is consistent with previous imaging studies; FA in the UF was correlated with executive function in subjects with schizotypal personality disorder (Nakamura et al., 2005).

Lower ADC and higher FA imply on higher white matter integrity, and indeed in the present paper, they were correlated with higher accuracy and lower RT. Higher FA values may indicate higher tissue directionality and organization, increased axonal and myelin density. Lower ADC values may indicate higher tissue density (Gouw et al., 2008; Blumenfeld-Katzir et al., 2011). Postmortem studies have shown that aging is accompanied by loss of 25–45% of the total length of myelinated fibers in the brain white matter (Marner et al., 2003). Degeneration of myelin sheaths and loss of nerve fibers in several but not all fiber bundles were correlated with cognitive decline in monkeys, among them the fornix (Peters and Kemper, 2011). Therefore, it is hard to interpret to which biological mechanism to relate the correlation of DTI parameters with cognitive decline. Based on the literature, it can be attributed to axonal loss, degeneration of myelin sheaths, loss of oligodendrocytes (Bartzokis, 2004), or other glial cells.

Correlation with ADC was accompanied by correlation with either D_R, D_A or both; however, correlation with FA did not correspond with correlation with either diffusivity except in the right ILF in which correlation between memory performance and D_R was obtained. This can be explained by the fact that ADC is an average of both diffusivities; however, change in FA is a result of change in the variance between the diffusivities.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.