A biomarker study in long-lasting amnestic mild cognitive impairment

The sample included 30 subjects (mean age 74.1 ± 4.8 years; mean education level 10.3 ± 4.5 years; mean disease duration at first evaluation 44.5 ± 25.5 months) fulfilling the Petersen criteria [1] for amnestic MCI (aMCI) and characterised by (1) a predominant episodic memory impairment, (2) a long-term clinical course (i.e., more than 4 years) and (3) a slow rate of progression of memory deficits. All subjects were consecutively referred to the neurology centres of San Raffaele Hospital (Milan, Italy). All of the included patients had a 3 to 5 years of clinical follow-up (i.e., 58.3 ± 10.1 months).

Conventional MRI was used to exclude the presence of white matter hyperintensities and lacunes of cerebrovascular origin as causes of cognitive impairments. Upon conventional inspection of MRI scans, hippocampal atrophy was found in the majority of subjects, without a clear radiological picture of hippocampal sclerosis (i.e., reduction of hippocampal volume with abnormal shape of mesial temporal lobe structures observed on T1-weighted images and increased signal intensity on T2-weighted and fluid-attenuated inversion recovery images) [18]. All of the included subjects had an FDG-PET scan, and 26 of 30 subjects with MCI had an amyloid biomarker evaluation (i.e., either with amyloid-PET and/or CSF Aβ₄₂ measure). See Tables 1 and 2 for details.

CSF was obtained from 20 of 30 subjects near the time of first clinical evaluation (< 3 months) by lumbar puncture in the L3-L4 or L4-L5 interspace. The procedure was performed early in the morning. No serious adverse events were reported. CSF (8–10 ml) was collected in sterile polypropylene tubes. Part of it was used to determine routine chemical parameters (leucocyte and erythrocyte cell count, glucose measurement, protein total content). The remaining CSF was centrifuged for 10 min at 4000 × g at 4 °C and stored at − 80 °C until analysis to ensure the stability of the CSF biomarkers. Measurement of CSF Aβ₄₂, total tau (t-tau) and phosphorylated tau (p-tau) levels was performed using commercial available enzyme-linked immunosorbent assay (ELISA) kits according to the manufacturer’s protocol and blinded to clinical data. Normal values were set ≥ 500 ng/L for Aβ₄₂ values, ≤ 450 ng/L (if age was 51–70 years) or < 500 ng/L (if age was > 71 years) for t-tau values and ≤ 61 ng/L for p-tau values, according to the ELISA kit guidelines and literature recommendations [19].

FDG-PET scans were acquired in 30 of 30 patients near the time of first clinical evaluation (< 3 months) according to European Association of Nuclear Medicine guidelines [20]. FDG-PET was performed in each subject at the Nuclear Medicine Unit, San Raffaele Hospital (Milan, Italy), with the Discovery STE multi-ring PET-computed tomography (CT) system (GE Medical Systems, Milwaukee, WI, USA). Before radiopharmaceutical injection of FDG (usually 185–250 MBq via a venous cannula), subjects were fasted for at least 6 h, and the measured blood glucose threshold was < 120 mg/dl. All images were acquired with an interval between injection and scan start of 45 min and scan duration of 15 min. Images were reconstructed using an ordered subset expectation maximisation (OSEM) algorithm. Attenuation correction was based on CT scans. Specific software integrated in the scanner was used for scatter correction. All subjects gave written informed consent after detailed explanation of the FDG-PET procedure.

Image pre-processing was performed using statistical parametric mapping (SPM; http://​www.​fil.​ion.​ucl.​ac.​uk/​spm/​software) according to an optimised SPM procedure with implementation of a standardised SPM FDG dementia-specific template [21] for spatial normalisation. This is an optimised method validated in MCI and different dementia conditions at the single-subject level and showing high accuracy in estimating specific metabolic patterns [22‐24]. In detail, images were smoothed with an 8-mm FWHM gaussian kernel; proportional scaling was used to remove inter-subject global variation in PET intensity; and each FDG-PET scan was then tested for relative “hypometabolism” by means of a two-samples t test implemented in SPM in a comparison with a normal FDG-PET image database (n = 112) on a voxel-by-voxel basis, with age as a covariate [22]. The FDG-PET image database included data acquired with different PET scanners. However, this did not affect data analysis with the optimised SPM procedure as evaluated and reported by Presotto et al. [25] with a large sample of FDG-PET images acquired with several PET scanners. The statistical threshold for the analysis at single-subject level was set at p = 0.05, familywise error (FWE)-corrected for multiple comparisons at the voxel level. Only clusters containing more than 100 voxels (i.e., 800 mm³) were deemed to be significant.

In order to compute brain metabolic changes at the group level, we also performed a second-level SPM analysis (one-sample t test) in the whole MCI group. The threshold was set at p < 0.05, FWE-corrected for multiple comparisons at the voxel level. In order to reveal more subtle effects in brain hypometabolism, a p < 0.001 uncorrected analysis (FWE-corrected at the cluster level) was also performed.

Amyloid-PET imaging was performed in 16 of 30 subjects using the [¹⁸F]florbetaben NeuraCeq™ tracer (Piramal Imaging, Berlin, Germany). All amyloid-PET acquisitions were performed within a mean of 3–12 months (i.e., 5.6 ± 1.9 months) from the first clinical evaluation. Patients received a single intravenous bolus injection of 315.6 ± 16.6 MBq of tracer. PET scans were acquired using a hybrid GE Discovery PET/CT 690 system (interval between injection and scan start 90 min; scan duration 15 min) [26]. CT scans were used for attenuation correction of PET data. All image reconstruction was performed using a 3D-OSEM algorithm. Images were scaled by division through the respective median cerebellar grey matter voxel intensity, and image processing resulted in regional cortical grey matter tracer uptake values relative to cerebellar grey matter (i.e., standardised uptake value [SUV]). SUVs were calculated for all regions according to the procedure reported by Villemagne et al. [27]. In particular, we considered ROIs in frontal (i.e., dorsolateral, ventrolateral and orbitofrontal regions), superior and inferior parietal, lateral temporal, lateral occipital, and anterior and posterior cingulate cortices. Each SUV was used to derive the composite SUV ratio (SUVr) referenced to the cerebellar cortex, a region most frequently used in SUV analysis because it is relatively unaffected by Aβ plaques in sporadic AD [27]. In addition to the above-mentioned cortical regions suggested in the literature to calculate the composite neocortical SUVr, we selected ROIs in hippocampal structures, amygdala and insula according to the Automated Anatomical Labelling atlas. This further analysis was done in order to test in vivo with PET whether amyloid deposits leading to a major effect on neurodegeneration in these brain regions might selectively exist. The long-lasting disease duration in subjects with MCI with selective long-term memory impairment or relatively stable or very slow progression has indeed been attributed to focal medial temporal lobe dysfunction possibly due to AD pathology [10‐12]. Finally, a non-parametric correlation analysis (Spearman’s rho) was performed to evaluate the relationship between mean neocortical amyloid-PET SUVr values and CSF Aβ₄₂ values in those patients who underwent both amyloid-PET and lumbar puncture.

At the time of the first clinical evaluation, the subjects’ age ranged from 65 to 84 years, and both sexes were similarly affected. Fifteen subjects were classified as having single-domain aMCI (s-aMCI), and 15 were classified as having multiple-domain aMCI (m-aMCI). Of note, 21 of 30 subjects showed behavioural disturbances (i.e., anxiety, irritability, aggressiveness, apathy and emotional blunting). At the clinical follow-up (58.3 ± 10.1 months), none had progressed to dementia. The majority of subjects with aMCI showed stable profiles (i.e., 9 with s-aMCI and 15 with m-aMCI). Mild progression of memory impairment and additional cognitive deficits were found in 6 subjects with s-aMCI, who were reclassified as having m-aMCI. See Tables 1 and 2 for demographic and clinical details.

Subjects with MCI who had some evidence of amyloid deposition (n = 15) visualised by CSF or PET imaging (see below) did not show any significant difference in clinical or neuropsychological features compared with amyloid-negative (n = 11) subjects, except for the female/male ratio and the disease duration in months, with amyloid-negative subjects showing longer disease duration (Table 3).

Among the 20 subjects with CSF data available, 7 had completely normal CSF values. Three subjects showed only slight elevation of p-tau levels, suggesting the presence of a neurodegenerative process. The remaining ten subjects showed Aβ values below the cut-off proposed for Aβ positivity [19]. Half of these subjects (i.e., five of ten) had only slightly altered Aβ results, with normal t-tau and p-tau values found in three of them. In these latter three subjects, MCI due to AD had a low probability according to the Erlangen Score Algorithm for the interpretation of CSF results [28, 29]. Pathological Aβ results, with slight alteration of p-tau levels, were present in three subjects. Only two subjects showed clearly pathological results for both Aβ and t-tau/p-tau, compatible with an AD-like CSF profile [29]. See Table 2 for further details on individual-subject values. The t-tau/Aβ ratio confirmed an AD-like profile in those subjects with clearly pathological Aβ values. Ratios above the cut-off proposed for AD positivity as reported by Shaw et al. [30] were also found in four of the subjects with an unclear single-biomarker profile. See Table 2 for details on individual-subject values.

No subject showed the typical AD temporoparietal hypometabolism on the basis of FDG-PET imaging [22‐24]. We found instead a consistent pattern of focal hypometabolism involving the hippocampus and/or hippocampal structures in every case. In half of the sample, additional metabolic changes were present in the frontomedial cortex, the amygdala, the posterior part of the insula extending to the parietal operculum and the superior temporal gyrus. Some patients (i.e., 11 of 30) showed posterior cingulate cortex hypometabolism. See Fig. 1a and b and Table 2 for details.

The group analysis of FDG-PET data using the SPM procedure showed a bilateral pattern of hypometabolism involving the hippocampal structures spreading to the insula on the left side (p < 0.05, FWE-corrected for multiple comparisons at the cluster level) (Fig. 1c). At the less stringent statistical threshold of significance (i.e., p < 0.001 uncorrected, FWE-corrected at the cluster level), we found a more extended hypometabolism only in the medial temporal lobe structures, without any involvement of the lateral temporoparietal regions (Fig. 1d).

Negative or very low levels of tracer uptake were found in the majority of subjects with MCI (Table 2). In particular, 8 of 16 patients had very low composite SUVr values (1.22 ± 0.12), comparable to those reported in the literature for Aβ-negative healthy control subjects [27, 31‐35], and notably well below the proposed cut-off for prodromal AD (i.e., < 1.45) [33]. Low to intermediate SUVr values (1.45 < SUVr < 1.80) below the mean values reported in the literature for AD Aβ positivity were found in 6 of 16 subjects (see Fig. 2). Only 2 of 16 subjects had slightly elevated composite SUVr values (i.e., 1.83 and 1.90) (Fig. 2). No regional effect of the tracer retention in the single cortical ROIs included in the composite SUVr measure emerged. Notably, in all subjects with MCI, there was a sparing of the hippocampal structures, amygdala and insula, in which SUVr values were always low overall (< 1.4). See Fig. 2 for details.

Amyloid-PET SUVr and CSF Aβ₄₂ values were consistent (Fig. 3). In detail, seven subjects had normal or slightly reduced CSF Aβ₄₂ values and no amyloid-PET burden, one had a normal CSF Aβ₄₂ value but a slight increase in SUVr value, and two had reduced CSF Aβ₄₂ levels and increased SUVr values. There was a significant negative correlation between mean neocortical SUVr and CSF Aβ₄₂ values (Spearman’s rho = − 0.83, p < 0.005).