The protocol was approved by the ethics committee of the University Hospitals of Geneva. All experimental procedures were carried out in accordance with the approved guidelines and with the principles of the Declaration of Helsinki. All participants were given written informed consent prior to inclusion. Participants were contacted via advertisements in local media to guarantee a community-based sample. An experienced neuropsychologist (C.R.) clinically assessed all individuals independently. Those who met dementia DSM-IV diagnostic criteria on the basis of the neuropsychological and clinical assessments were excluded. Moreover, exclusion criteria included psychiatric or neurologic disorders, sustained head injury, history of major medical disorders (neoplasm or cardiac illness). Moreover, subjects were included if they had no alcohol or drug abuse, no regular use of neuroleptics, antidepressants or psychostimulants and no contraindications to MR imaging. To control for the confounding effect of cardiovascular diseases, individuals with subtle cardiovascular symptoms and a history of stroke and transient ischemic episodes were also excluded from the present study. The inclusion period for control subjects and patients with MCI was from October 2010 to March 2016.

The final sample included 455 elderly individuals (mean age, 73.5 ± 4.2 years, 263 females). Among them, 182 were subsequently classified in the sCON group (mean age, 73.1 ± 3 years, 112 females), 192 in the dCON group (74.2 ± 4.1 years, 116 females), and 81 in the MCI group (73.9 ± 5 years, 35 females).

At baseline (T1), all individuals underwent neuropsychological assessment. The control and MCI participants were evaluated with an extensive neuropsychological battery, including:

Education level was defined according to the Swiss scholar system: level 1, less than 9 years (primary school); level 2, between 9 and 12 years (high school); and level 3, more than 12 years (university). Moreover, only cases with a CDR score of 0 and scores within 1.5 standard deviations of the age-appropriate mean in all other tests were included in the control group.

Participants with a CDR score of 0.5 but no dementia and a score more than 1.5 standard deviations below the age-appropriate mean in any of the previously mentioned tests were confirmed to have MCI, in agreement with the criteria of Petersen et al. (2001).

Eighteen months (±2 weeks) after the baseline evaluation (T2), control subjects underwent cognitive reassessment with the same neuropsychological battery. Participants were placed in the dCON group at follow-up if they had a performance 0.5 standard deviation lower than that at inclusion for two or more neuropsychological tests. Additionally, all individuals were clinically assessed independently by two board certified neuropsychologists (S.T., E.T.; 4 and 2 years of experience, respectively). The final classification of dCON was made blindly by a trained neuropsychologist (C.R., 10 years of experience) using both the neuropsychological scores and clinical assessment. For the extensive neuropsychological evaluation and clinical tests scores at T1 see Supplementary Material and previously published works (Xekardaki et al., 2015).

Whole blood samples were collected at baseline for all subjects for APOE genotyping. Standard DNA extraction was performed using either 9 ml EDTA tubes (Sarstedt, Germany) or Oragene Saliva DNA Kit (DNA Genotek, Inc., Ottawa, ON, Canada) which were stored at -20°C. APOE genotyping was done on the LightCycler (Roche Diagnostics, Basel, Switzerland) as described previously (Nauck et al., 2000). Subjects were divided based on if they were a carrier of the APOEε4 allele (ε4/ε3, ε3/ε3, ε3/ε2 carrier).

All participants underwent MRI examination including two structural (T1 and T2) sequences.

Images were obtained using a 3T scanner (Trio; Siemens, Erlangen, Germany) with a standard 32-channel head-coil. A magnetization-prepared rapid gradient-echo three-dimensional T1-weighted sequence was performed for spatial normalization and gray matter segmentation with the following fundamental parameters: 2300/2.3, 256 × 256 matrix, 176 sections, and 1 mm × 1 mm × 1 mm.

Additional sequences included axial fast spin-echo T2-weighted imaging (4000/105, 30 sections, 4-mm section thickness), susceptibility weighted imaging (28/20, 208 × 256 × 128 matrix, 1 mm × 1 mm × 1 mm voxel size) were performed to exclude brain disease, such as ischemic stroke, subdural hematomas, or space-occupying lesions.

FMRIB’s Integrated Registration and Segmentation Tool (FSL FIRST¹) was applied according to the standard procedure (de Jong et al., 2008; Patenaude et al., 2011) to estimate left and right gray matter volumes of the hippocampus and amygdala. FIRST is part of FMRIB’s Software Library (FSL) and performs registration and segmentation of these regions (Patenaude et al., 2011). The 3D T1 images are registered into MNI (Montreal Neurological Institute) standard space, by means of affine transformations based on 12 degrees of freedom. After registration, FSL-FIRST applies ad hoc masks to locate the structures by segmentation based on shape models and voxel intensities. ROIs were defined using the Harvard-Oxford Subcortical Structural Atlas implemented in FSL. The absolute volumes were calculated, taking into account the transformations made in the first stage. Subsequently, a boundary correction was used to determine which boundary voxels belong to the structure or not. In this study a Z-value of 3 was used, corresponding to a 99.998% certainty that the voxel belonged to the subcortical structure. All segmented regions were visually checked for errors in registration and segmentation.

Brain tissue volume was estimated with SIENAX² (Smith et al., 2002), part of FSL (Smith et al., 2004). SIENAX extracts brain and skull images from the single raw input data (Smith et al., 2002; Woolrich et al., 2009). Tissue-type segmentation with partial volume estimation was carried out to calculate total volume of brain tissue (including volumes of gray matter, white matter). For this study we used the absolute volumes generated by the algorithm. Intracranial volume was calculated by adding the volumes of cerebral spinal fluid, total gray matter and total white matter.

After extracting the volumes of the identified areas (FIRST and SIENAX-output) a one-way ANOVA group comparison was performed including GM volumes as dependent variable, and age, gender, educational level, APOE genotype and intracranial volume as non-explanatory covariates. Tukey corrections for multiple comparison was applied with a p < 0.05 threshold.

Moreover, after FIRST, vertex analysis was performed using the standard procedure (Patenaude et al., 2011) and shape changes were assessed on a per vertex basis. GM atrophy in the vertices across all subjects was assessed using a general linear model (GLM) with age as covariate using permutation-based non-parametric testing (Randomize, part of FSL). Threshold-free cluster enhancement (TFCE) multiple comparison correction was applied, considering corrected p-values < 0.05 as significant.

Group differences in age, gender, end education as well as changes in neuropsychological scores between T1 and T2 were assessed using a one-way independent ANOVA with Tukey correction for post hoc pair-wise comparisons. Correlations between neuropsychological scores changes (delta T1–T2) and GM volumes at baseline were assessed using Spearman rank correlation coefficients. In addition, linear regression models were built with the delta of cognitive scores as dependent variable and MRI parameters at baseline, age, years of education, and APOE genotyping as independent variables. Of note, the variable gender was not considered in this linear regression since a separate general regression analysis showed no differences in cognitive performance between males and females within each diagnostic group. Statistical analyses were performed using GraphPad Prism (Version 6, GraphPad Software, San Diego, CA, USA), R (Version 0.99.896, The R Project for Statistical Computing) FSL (Version 5.0.6, FMRIB, Oxford, UK).

The dCON cases were slightly older than MCI cases (p < 0.05) with no age differences between sCON and the other two diagnostic groups [F(2,452) = 4.34, η² = 0.2, p < 0.05]. A lower proportion of women was found in MCI group (43%) compared to sCON (61%) (p < 0.01) and dCON (60%) [F(2,452) = 4.34, η² = 0.02, p < 0.05]. Education level [F(2,452) = 2.42, η² = 0.01, p = 0.32] and APOE allele distribution did not differ between the three groups [F(2,452) = 4.34, η² = 0.02, p = 0.25].

Upon follow-up, higher MMSE scores were found in sCON compared to dCON (p < 0.001) and MCI (p < 0.001) and in dCON group compared to MCI cases (p < 0.01) [F(2,452) = 26.54, η² = 0.11, p < 0.001]. Moreover, significantly lower HAD scores [F(2,452) = 5.16, η² = 0.02, p < 0.01] were found in sCON compared to MCI cases (p < 0.01). As expected, dCON cases showed lower CDR scores compared to MCI cases (p < 0.001) [F(2,452) = 1.6, η² = 0.02, p < 0.001]. RI-48 performance was lower in MCI compared to dCON (p < 0.001) and sCON (p < 0.001) and in dCON compared to MCI (p < 0.001) [F(2,452) = 65, η² = 0.3, p < 0.001]. Finally, significantly higher CERAD scores were found in sCON compared to MCI (p < 0.05) [F(2,452) = 3.1, η² = 0.05, p < 0.05] (Table 1). For the results of the neuropsychological assessment at T1, see Supplementary Material and Supplementary Table A.

Significant shape differences were found using vertex analyses in the bilateral hippocampus and amygdala. In particular, GM atrophy was more pronounced at the superior- posterior lateral side of the structures both for the dCON compared to sCON and for MCI compared to sCON (Figures 1E–H). No significant contribution of age was observed.

Significant differences were found in the volumes of the left hippocampus [F(2,452) = 4.51, η² = 0.015, p < 0.05], right hippocampus [F(2,452) = 4.23, η² = 0.014, p < 0.05] and right amygdala [F(2,452) = 5.3, η² = 0.02, p < 0.01] between sCON and dCON (p < 0.05) as well as between sCON and MCI cases (p < 0.05) (Figures 1A,B,D). Significant differences were found in the volumes of the left amygdala between sCON and MCI cases (p < 0.05) [F(2,452) = 3.54, η² = 0.01, p < 0.05] (Figure 1C). No significant differences were found between dCON and MCI in the considered subcortical GM volumes.

In order to address the possible confounding effect of APOE genotype and gender on the observed differences, we performed a one-way ANOVA including GM volumes as dependent variable, and age, gender, educational level, APOE genotype and intracranial volume as non-explanatory covariates. Moreover, we also repeated the analysis inserting in the model the interaction APOExgender as non-explanatory variable. Our data persisted after controlling for APOE and gender. In fact both variables were not associated with GM volumes in the present sample.

Significant associations were found between gray matter volume of left hippocampus and delta of MMSE and RI-48 scores in dCON (MSSE: β-coefficient = 0.47, p < 0.001; RI-48: β-coefficient = 0.51, p < 0.001) and MCI (MSSE: β-coefficient = 0.65, p < 0.001; RI-48: β-coefficient = 0.18, p < 0.01) groups (Figure 2).

Strengths of the present study includes its longitudinal design, large numbers of community-dwelling cases, as well as detailed neuropsychological testing at inclusion and follow-up. However, some limitations should also be considered. Consistent with recent core clinical criteria for MCI (Albert et al., 2011), the identification of dCONs was based on the objective decline in cognitive functions measured using serial, comprehensive neuropsychological assessments. However, in the absence of longer follow-up the cognitive fate of these cases remains uncertain so that they cannot be a priori considered as incipient AD cases. No CSF measures of tau and Aβ protein were available in this work so that the real extent of AD pathology remains unknown. Future studies exploring the tau and amyloid status based on PET in our community-based cohort are warranted to examine the relationship between AD molecular events and MRI atrophy in limbic areas in dCON.