Study participants

The regional ethics committee approved the study, and written consent was obtained from the participants, family members, or caregivers. Subjects were first classified into three groups—normal control (NC), mildly cognitively impaired (MCI), and Alzheimer’s disease dementia (ADD). The clinical diagnosis of probable AD was made according to the National Institute of Neurological and Communicative Diseases and Stroke/Alzheimer’s Disease and Related Disorders Association Alzheimer’s criteria [4]. Controls had no evidence of neurological disease or impairment in cognitive function or activities of daily living. Individuals with a focal lesion on the brain identified using MRI (magnetic resonance imaging), a history of head trauma, or a psychiatric disorder that could affect mental function were excluded. The diagnosis of MCI was made according to the National Institute on Aging-Alzheimer’s Association’s (NIA-AA) criteria [28]. Depending on the PET scans (positive or negative) and as summarized in Table 1, (S2 Table and S2 Fig), 478 study subjects (NC: n = 192; β-amyloid negative cognitively unimpaired [NC-]), asymptomatic AD (n = 34; β-amyloid positive cognitively unimpaired [aAD or NC+]), cognitive impairments that are not dementia (CIND) (n = 118; β-amyloid negative mildly cognitively impaired [MCI-]), prodromal AD (n = 34; β-amyloid positive mildly cognitively impaired [pAD or MCI+]), Alzheimer’s disease dementia negative (n = 30; β-amyloid negative severely cognitively impaired [ADD-]), and Alzheimer’s disease dementia (n = 70; β-amyloid positive severely cognitively impaired [ADD+]) were recruited by the Gwangju Alzheimer’s disease and Related Dementias (GARD) cohort research center at Chosun University in Gwangju, Republic of Korea.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Demographic characteristics of the study population.

https://doi.org/10.1371/journal.pone.0275233.t001

Clinical and neuropsychological assessments

Participants underwent testing using the Seoul Neuropsychological Screening Battery, a comprehensive neuropsychological battery of tests and subtests assessing five cognitive domains: memory (subtests: Orientation and Verbal and Visual Memory), language (Korean version of the Boston Naming Test and written Calculation Trails), attention (Forward and Backward Digit Span), visuospatial functions (Copying Test from the Rey Complex Figure Test), and frontal/executive functions (Motor Impersistence, Contrasting Program, Go-no-go Test, Fist-edge-palm task, and the Luria Loop task) [29, 30]. In addition, the battery of tests included the Korean version of the Mini-Mental State Examination (MMSE) [31], the Clinical Dementia Rating [32], and the Seoul Instrumental Activities of Daily Living [33]. Control subjects did not show any evidence of neurological disease or impairment in cognitive function or activities of daily living.

MRI acquisition

Contiguous 0.8 mm sagittal Magnetization-Prepared Rapid Acquisition Gradient Echo (MPRAGE) images of the whole brain examined at the GARD cohort research center was acquired using a 3T MR scanner (Skyra, Siemens) with the following parameters: TR = 2300 ms; TE = 2.143 ms; TI = 900 ms; 9 flip angles; FoV = 256x256; matrix = 320x320; number of slices = 178.

MRI data processing

High-resolution structural T1-weighted images were processed using the FreeSurfer software package (v5.3.0 & v6.0.0; Athinoula A. Martinos Center for Biomedical Imaging, Harvard University, Cambridge, MA, USA) on a Linux environment using a 64-bit CentOS 7 operating system. The complete documentation of the FreeSurfer pipeline and methodologies can be found elsewhere [34–38]. Complete automated processing, including cortical and subcortical labeling using the Desikan–Killiany atlas, was performed on each subject. Then, the hippocampal subfields were accessed using FreeSurfer v6.0.0, sub-dividing the hippocampus into subfields, namely the hippocampal tail (Tail), Sub, CA1, hippocampal fissure (fissure), presubiculum (PrS), parasubiculum (PaS), molecular layer (ML), granule cells/molecular layer/dentate gyrus (GCMLDG), CA3, CA4, fimbria, and hippocampal-amygdala transition area (HATA) (S1 Fig) [23]. The hippocampal subfields segmentation was performed for all groups using only FreeSurfer (v6.0.0). It should be noted that the CA3 sector includes the CA2 sector in the hippocampal subfield atlas used. In the current study, we excluded the whole hippocampus volumes from all analyses, as the primary focus was on the subfields.

PET imaging

Subjects underwent a PET scan 90 min after intravenous injection of 300 MBq ¹⁸F-florbetaben using a dedicated Discovery ST PET-CT scanner (General Electric Medical Systems, Milwaukee, WI, USA). Non-contrast-enhanced computed tomography (CT) scans were used for attenuation correction with technical parameters of 120 Kvp, 10–130 mAs, eight slices, helical, and 3.79 mm slice thickness. PET and CT scan data were reconstructed using ordered subset expectation maximization after attenuation correction with two iterations and 21 subsets. A Gaussian filter was applied with 5.14 mm FWHM to reconstruct a 128 × 128 matrix with 3.27-mm slice thickness.

PET image data processing

PET images were assessed according to a predefined regional cortical tracer uptake (RCTU) scoring system (“1” = no uptake, “2” = minor uptake, “3” = pronounced uptake) for four brain regions (frontal cortex, posterior cingulate, lateral temporal cortex, parietal cortex). Details of a three-grade scoring system using RCTU scores for the amyloid plaque load have been provided previously [39]. For the PiB-PET images, the mean retention value of the global cortical region of interest (ROI) was used to define the global cerebral Aβ deposition as amyloid-positive if the mean standard value uptake ratio was > 1.4 in at least one of the ROIs, including the frontal, lateral parietal, lateral temporal, or posterior cingulate-precuneus [40].

Statistical analyses

All statistical analyses were performed using IBM SPSS Statistics (Version 23.0., Armonk, NY: IBM Corp.) All analyses were two-tailed and controlled for the covariates: age, sex, years of education, and estimated total intracranial volume. One-way analysis of variance and Bonferroni post hoc correction for multiple comparisons was used for continuous demographic variables, and chi-squared tests were performed for categorical demographic variables. A series of analysis of covariance (ANCOVA) on the estimated subfield volumes adjusting for the covariates were performed comparing the six diagnosis groups. The P-values of the comparisons between the diagnosis groups were corrected using the Bonferroni correction for multiple comparisons. The Bonferroni correction was performed across the 24 regions and six groups (Table 2). Further post hoc analysis of the subfields that met the minimum standards of significance after the Bonferroni correction for multiple comparisons were performed for pairwise comparisons between the two groups. Here, we considered p < 0.05 as significant (Table 3). The left-right hemispheric volumetric asymmetry comparisons were evaluated using the t-test (Table 4). p < 0.05 was considered statistically significant. Additionally, we investigated the hemispheric volumetric correlation among the six diagnostic groups using Pearson’s correlation, and again we used a statistical significance of p < 0.05.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Volume (mm³) of left and right hippocampal subfields.

https://doi.org/10.1371/journal.pone.0275233.t002

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Bonferroni pairwise post hoc analysis.

https://doi.org/10.1371/journal.pone.0275233.t003

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Comparison of hemispheric differences using paired t-tests.

https://doi.org/10.1371/journal.pone.0275233.t004

Demographics

Table 1 shows no significant differences in terms of age (F_{5, 478} = 1.29, p = 0.26) and gender (χ² test: χ² = 6.60, p = 0.25) were observed among the groups. Levels of education (F_{5, 478} = 2.54, p = 0.02) were significantly different when all the groups were compared, and an additional pairwise comparison revealed that the difference was only specific to CIND and ADD+ groups (post hoc: CIND versus ADD+, p = 0.01). The MMSE scores were significantly different between the groups (F_{5, 478} = 73.44, p = 1.08E-56). However, the post hoc analyses showed that the scores were not different between aAD and NC, pAD and CIND, and ADD- and ADD+ groups.

Group comparisons and hemispheric asymmetry findings

As summarized in Tables 2 and 3, comparisons of the six groups showed that hippocampal fissure volume was not significantly different between them (after adjusting for Bonferroni correction, p = 0.05/ 24 structures/ 6 groups = 3.47E-4). Additionally, the bilateral PaS, left CA3, and right fimbria did not show any significant volume differences between the aAD and pAD groups. The left fimbria and right CA3 showed significant atrophy in the supposed early AD progression stages, though they did not show any significant atrophy difference between the pAD and ADD+ groups. The percentile volume loss was higher in the prodromal AD to AD dementia stage than in the preclinical AD to prodromal AD stage. Predominantly, there was a higher percentile volume loss in the left hemisphere than in the right hemisphere.

Findings in NC and aAD groups

Subfield volumes between NC and aAD groups were not significantly different from each other (Tables 2 and 3; Fig 1). We observed left-right hemispheric differences in both groups, with significantly larger right hemisphere volumes than in the left hemisphere, except in the PrS, PaS, fimbria, and HATA (Table 4). In these regions, the volumes on the right were smaller than those on the left. However, in the aAD group, these differences were not statistically significant, except in the fimbria. In addition, the asymmetries were higher in the NC group than in the aAD group. Compared to other subfields, the left-right PaS volumes in NC and aAD groups were weakly correlated (Table 5).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Hippocampal subfield volumes in NC, aAD, pAD, and ADD groups adjusting for the covariates age, sex, level of education, and intracranial volume.

Error bars indicate two standard errors. There were seven prominent regions among the entire 13 regions. NC, normal controls; aAD, asymptomatic Alzheimer Disease; pAD, prodromal Alzheimer Disease; ADD, Alzheimer Disease Dementia; n.s, not statistically significant; *, p < 0.05; **, p < 0.01; ***, p < 0.001.

https://doi.org/10.1371/journal.pone.0275233.g001

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Correlations between the two hemispheres.

https://doi.org/10.1371/journal.pone.0275233.t005

Findings in CIND and pAD groups

As shown in Tables 2 and 3 and Fig 1, bilateral volumes of the hippocampal tail, CA1, ML, GCMLDG, CA4, and right CA3 significantly differed between CIND and pAD groups, with larger volumes in the CIND group than in the pAD group. In addition, left-right hemispheric asymmetries were significantly higher in CIND than in the pAD group. All other subfields except PrS, PaS, HATA, and fimbria (Table 4) had larger right hemisphere volumes. The left-right PaS volumes in CIND and pAD groups were moderately correlated (Table 5).

Findings in ADD- and ADD+ groups

The ADD- group had significantly larger volumes than the ADD+ group in the bilateral tail, SUB, PrS, and left ML (Tables 2 and 3; Fig 1). In addition, subfields other than the PrS, PaS, HATA, and fimbria (Table 4) had larger right hemisphere volumes than left hemisphere volumes. Inconsistent with the between NC and aAD and the between CIND and pAD group findings, the left-right hemispheric asymmetry was lower in the ADD- group than in the ADD+ group. In addition, the PaS left-right hemispheric volumes were weakly correlated in the ADD+ group and moderately correlated in ADD- group (Table 5).

Findings of the study

Neurodegeneration is believed to cause cognitive dysfunction, which is expected to be mediated by the deposition of β-amyloid [20]. Approximately 10–30% of cognitively normal individuals have abnormal levels of β-amyloid, suggesting a biological relevance to normal aging [41, 42] and also suggesting that the abnormal levels of β-amyloid are not sufficient to cause clear cognitive symptoms [6]. Prior studies show that β-amyloid positive cognitively normal, mild cognitively impaired, and AD dementia subjects at baseline showed greater cognitive and global deterioration compared to β-amyloid negative subjects [43–45]. Without neurodegeneration, which provides vital pathological staging, the differences between the plaques and tangles cannot be formally captured. Furthermore, standalone PET imaging cannot function as a significant prediction tool for future cognitive impairment without the use of MRI [46].

Differential volume loss among groups

The bilateral comparison of the six diagnostic groups showed that all the subfields volumes differed between the groups, except for the hippocampal fissure. However, when the diagnostic group volumes were compared using pairwise comparison, certain subfield volumes between a few groups were not different.

Comparison between NC and aAD

In line with previous studies [47, 48], there were no significant differences in volumes between the NC and aAD groups. Although, a few studies have shown marginal statistical significance in volume differences of one or two subfields between aAD and NC, with aAD showing lower volumes than the NC [49, 50]. These studies have failed to find clear differences in the other subfields they studied. The differences reported could be due to the sampling and sample size discrepancies, procedural variations, and most notably, ethnic variability. We observed a trend level increase in the subfield volumes of the aAD group compared to the NC group, which is similar to the observations made in the final model of temporal ordering of biomarkers in the Dominantly Inherited Alzheimer Network study [51].

No clear differences in the normal aging-associated hippocampal atrophy rates have been observed between the normal control and the MCI subjects with the same β-amyloid status [52], and an accelerated volume loss was seen specifically in the MCI group [53–55]. Thus, the current stage might be too early to see a clear volume loss, and a longitudinal study between the NC and aAD groups might provide a much clearer picture of the underlying mechanisms.

Comparisons between cognitively normal and mild cognitively impaired

It is shown that hippocampal atrophy can act as a strong predictor of AD progression and can discriminate the MCI from the cognitively normal [56]. When we compared the NC and aAD groups to the CIND group bilaterally, we observed no clear differences in the subfield volumes. Interestingly, when we compared the NC and aAD groups to the pAD group bilaterally, most of the subfield volumes showed clear differences; this indicates that the pAD is the transitional stage in the AD continuum, and the differences in the subfield volumes might be useful to classify subjects with mild cognitive impairment into pAD and CIND groups even before the β-amyloid imaging. A more extensive study might help clarify the underlying mechanisms of this transitional stage.

Comparisons between cognitively normal and severely impaired

In line with previous studies [25, 57, 58], comparing the NC and aAD groups with the ADD+ and ADD- groups yielded very clear differences in all the subfields, clearly differentiating the normal controls from subjects with dementia.

Comparisons between pAD and CIND

In line with an earlier study [59], in the MCI groups pAD and CIND, we observed larger subfield volumes in the CIND group than in the pAD group. Additionally, we observed no significant bilateral volume difference of the subicular complex (SUB, PrS, and PaS), but there was atrophy in the bilateral tail, CA1, ML, GCMLDG, CA4, and right CA3.

Comparisons between mild cognitively impaired and severely impaired

We then compared the pAD & CIND groups with the ADD+ & ADD- groups. There were clear differences when the CIND group was compared with the ADD+ and ADD- groups. The CIND might be a result of cognitive impairment resulting from the normal aging process, metabolic disturbance, substance abuse, or head trauma [28, 60]. pAD & ADD+ groups showed clear differences in all subfields, except the bilateral PaS, left fimbria, and right CA3, suggesting that these regions are atrophied only in the late AD stages, located some distance from the suspected pathological initiation sites, the CA1 and the subiculum [61–63].

Comparisons between ADD+ and ADD-

Finally, a comparison of ADD+ and ADD- groups showed differences in the bilateral SUB, PrS, tail, and left ML, suggesting that the volume atrophy in ADD+ might be more severe than in ADD- or other types of dementia [64]. All other subfields showed no clear differences. In addition, several subjects in the ADD- group has been shown to demonstrate Aβ positivity in a longitudinal study [65].

Comparisons between CIND and ADD-

Individuals in the group CIND or Aβ- MCI mostly have less cognitive and functional impairment at baseline than patients with pAD or Aβ+ MCI and individuals in the group Aβ- MCI are less likely to convert to AD. In Aβ- MCI, the simultaneous presence of several comorbidities makes it difficult to pinpoint the cause of the cognitive symptoms. In ADD- or Aβ- ADD individuals, non-AD etiologies like subclinical depression and vascular abnormalities may account for AD-like phenotype. However, individuals with ADD+ or Aβ+ ADD and Aβ- ADD do not have very strong differences in cognitive impairments [66].

The trajectory of volume loss

Based on the results, we expect that the trajectory of AD progression might be from NC to aAD to pAD to ADD+. In cognitively unimpaired and MCI subjects, those diagnosed with Aβ-negativity showed a higher hemispheric volume difference than those diagnosed with Aβ-positivity. In contrast, in severely cognitively impaired subjects, those diagnosed with Aβ-positivity showed a higher hemispheric volumetric difference than those diagnosed with Aβ-negativity [64]. Our study shows the importance of separating the Aβ positive subjects from the Aβ negative ones.

Selective vulnerability among subfields

Neuropathological studies have reported that the AD continuum reflects a complex and ordered sequential process with the atrophy beginning at the anterior CA1-subiculum regions and progressing toward other subfields [61, 62, 67, 68]. Also, another study has reported presubicular-subicular complex atrophy in the earliest stages of AD [69]. In line with the previous findings, we observed atrophy in both the CA1 and subiculum regions in a comparison between the cognitively unimpaired groups and the pAD group. Additionally, atrophy in other subfields was observed in a comparison between the cognitively unimpaired groups and the pAD group. The bilateral parasubiculum, left CA3, and right fimbria situated at some distance from the CA1-subiculum regions show no atrophy until the late AD stages. The parasubiculum is a transitional area sandwiched between the presubiculum and the entorhinal area [19] and is postulated to play a vital role in spatial navigation and the integration of head-directional information [70]. The fimbria extends to the fornix, the brain’s white matter, and the CA3 is expected to be the largest subregion in the hippocampus [71]. The atrophy of the parasubiculum and CA3 might suggest terminal disease propagation stages that later spread to other brain regions.

Hemispherical asymmetry in subfield volume loss

In line with an earlier study [72], we observed left-right hemispheric asymmetry in the atrophy pattern. In most subfields, the left hemispheric volumes showed more severe atrophy than the right hemispheres. The early atrophy of the left fimbria and the right CA3 was observed but not the right fimbria and the left CA3, which show a hemispherical difference in the atrophy patterns of these two regions [73]. These hemispheric volume differences are seen in almost all subfields, with larger right hemisphere volumes except in the PrS, PaS, and HATA.