We enrolled 1,175 non-demented participants (456 with CU and 719 with MCI) aged ≥ 45 years who underwent Aβ PET in the memory clinic at the Samsung Medical Center (Seoul, Korea) between August 2015 and August 2020. All participants underwent a comprehensive dementia examination including a standardized neuropsychological test [16], APOE genotyping, and brain magnetic resonance imaging (MRI). All participants with CU met the following criteria: (1) no medical history that was likely to affect cognitive function based on Christensen’s health screening criteria [17]; (2) no objective cognitive impairment in any cognitive domain on a comprehensive neuropsychological test battery (above at least -1.0 SD of age-adjusted norms on any cognitive test); and (3) independence in activities of daily living. All participants with MCI met the criteria for MCI with the following modifications [18, 19]: (1) subjective cognitive complaints by the participants or caregivers; (2) objective memory impairment below -1·0 SD on verbal or visual memory tests; (3) no significant impairment in activities of daily living; and (4) non-demented status.

We excluded participants with severe WMH (cap or band > 10 mm and longest diameter of deep white matter lesion > 25 mm), structural lesions including cerebral infarction, intracranial hemorrhage, brain tumors, and hydrocephalus on MRI, and abnormal laboratory results on complete blood count, electrolyte, vitamin B12 and folate levels, syphilis serology, and liver/kidney/thyroid function tests.

The Institutional Review Board of the Samsung Medical Center approved this study. Written informed consent was obtained from all the participants.

All participants underwent Aβ PET (18F-florbetaben) and 18F-flutemetamol PET scans using a Discovery STe PET/CT scanner (GE Medical Systems, Milwaukee, WI, USA). For 18F-florbetaben PET or 18F-flutemetamol PET, a 20-min emission PET scan in dynamic mode (comprising 4 × 5 min frames) was performed 90 min after an injection of a mean dose of 311.5 MBq 18F-florbetaben or 197.7 MBq 18F-flutemetamol, respectively. Three-dimensional PET images were reconstructed in a 128 × 128 × 48 matrix with 2 × 2 × 3·27 mm voxel size using the ordered-subsets expectation maximization algorithm (18F-florbetaben, iteration = 4 and subset = 20; 18F-flutemetamol, iteration = 4 and subset = 20).

Aβ uptakes were quantified using BeauBrain Morph of BeauBrain Healthcare Co., Ltd., which performs fully-automated image analysis of Aβ uptakes on PET images. We used a direct comparison of the FBB-FMM CL (dcCL) method previously developed by our group [20] to standardize the quantification of Aβ PET images obtained using different ligands. The dcCL method for FBB and FMM PET enables the transformation of the standardized uptake value ratio (dcSUVR) of FBB and FMM PETs to dcCL scales directly, without conversion to the 11C-labeled Pittsburgh compound SUVR.

There are three steps to obtaining dcCL scales [20]: 1) pre-processing of PET images, 2) determination of the global cortical target volume of interest (CTX VOI), and 3) conversion of dcSUVR to dcCL scales. First, to preprocess the Aβ PET images, PET images were co-registered to each participant’s MR image and then normalized to a T1-weighted MNI-152 template using the SPM8 unified segmentation method. We used T1-weighted MRI correction with the N3 algorithm only for intensity nonuniformities without applying corrections to the PET images for brain atrophy or partial volume effects. Second, we used the FBB-FMM CTX VOI, defined as areas of AD-specific brain Aβ deposition in our previous study [20]. In our previous study [20], we developed CTX VOI using the similar methods with Klunk’s Centiloid methods [21]. In their original study [21], they included 19 patients with AD dementia and 25 older controls to determine areas of AD-specific brain Aβ deposition. Briefly, to exclude areas of age-related brain Aβ deposition, the FBB-FMM CTX VOI was generated by comparing SUVR parametric images (with the whole cerebellum as a reference area) between 20 typical patients with Alzheimer’s disease-related cognitive impairment (ADCI-CTX) and 16 healthy elderly participants (EH-CTX) who underwent both FBB and FMM PET scans. To generate the FBB-FMM CTX VOI, the average EH-CTX image is subtracted from the average ADCI-CTX image. We then defined the FBB-FMM CTX VOI as the area of ADCI-related brain Aβ accumulation common to both FBB and FMM PET. Finally, the dcSUVR values of the FBB-FMM CTX VOI were converted to dcCL scales by using the dcCL conversion equation. The dcCL equation was derived from the FBB-FMM CTX VOI separately for FBB and FMM PET, and applied to FBB and FMM dcSUVR.

To determine the participants’ dcCL cut-off-based Aβ positivity, we applied the optimal cut-off value derived using k-means cluster analysis in 527 independent samples of participants with normal cognition. The cut-off value was set at 27.08, representing the 95th percentile of the lower cluster, and the whole cerebellum was used as a reference region.

All participants underwent blood tests. Blood samples were collected after an overnight fast. The mean ± SD of time interval between blood tests and Aβ PET was 3.9 ± 5.0 months. Cholesterol levels, including LDL-c, HDL-c, and triglycerides, were measured by an enzymatic colorimetric test using a Modular D2400 analyzer (Roche Diagnostics, Basel, Switzerland). Height and weight were measured for all participants. BMI was calculated as weight (kg) divided by the square of height (m). Cholesterol levels and BMI data were obtained by backtracking in the clinical data warehouse (CDW) of the Samsung Medical Center, which were measured within 12 months before or after Aβ PET scans. Prescriptions of lipid-lowering and dementia medications (donepezil, rivastigmine, galantamine, or memantine) were extracted from the CDW. Hypertension and diabetes were defined as a diagnostic history of hypertension and diabetes or current use of any antihypertensive medication and antidiabetic medication, respectively.

We acquired standardized three-dimensional T1 Turbo Field Echo and three-dimensional fluid-attenuated inversion recovery (FLAIR) images using a 3.0 T MRI scanner (Philips 3.0T Achieva; Philips Healthcare, Andover, MA, USA), as previously described [22].

The images were processed using the CIVET anatomical pipeline (version 2.1.0). The native MRIs were registered to the MNI-152 template by linear transformation and corrected for intensity nonuniformities using the N3 algorithm. The registered and corrected images were divided into white matter, grey matter, cerebrospinal fluid, and background. In addition, the inner and outer surfaces of the cortex were automatically extracted using the marching-cubes algorithm to obtain the cortical thickness, which was defined as the Euclidean distance between the linked vertices of the inner and outer surfaces.

As we extracted cortical surface models from MRI volumes transformed into stereotaxic space, the cortical thickness was measured in the native space by applying an inverse transformation matrix to the cortical surface and reconstructing them in the native space.

To measure hippocampal volume, we used an automated hippocampus segmentation method using a graph cut algorithm combined with atlas-based segmentation and morphological opening, as described in a previous study [23]. WMH segmentation was replicated using the Lesion Segmentation Tool (LST) in Statistical Parametric Mapping 12 [24]. LST is an automated segmentation approach for quantifying whole-brain WMH volume and shows high agreement with manual tracing of WMH in FLAIR.

To investigate the association between cholesterol levels and Aβ uptake, we performed linear regression analyses with each cholesterol type (LDL-c, HDL-c, and triglyceride) as a predictor and quantified dcCL scales as an outcome after controlling for age, sex, BMI, APOE e4 allele (APOE4) genotype, hypertension, diabetes, lipid-lowering medication, and dementia medication.

To investigate the association between cholesterol levels and neurodegeneration, we performed linear regression analyses with each cholesterol type (LDL-c, HDL-c, and triglyceride) as a predictor and hippocampal volume as an outcome after controlling for age, sex, BMI, APOE4 genotype, hypertension, diabetes, lipid-lowering medication, dementia medication, and intracranial volume.

To investigate the association between cholesterol levels and WMH volume, we performed linear regression analyses with each cholesterol type (LDL-c, HDL-c, and triglyceride) as a predictor and hippocampal volume as an outcome after controlling for age, sex, BMI, APOE4 genotype, hypertension, diabetes, lipid-lowering medication, dementia medication, and intracranial volume.

To determine whether Aβ uptake mediates the effect of cholesterol levels on neurodegeneration, we used path analyses with each cholesterol type as a predictor, Aβ uptake as a mediator, and hippocampal volume as an outcome after controlling for age, sex, BMI, APOE4 genotype, hypertension, diabetes, lipid-lowering medication, dementia medication, and intracranial volume.

The standardized β was also calculated to compare the strength of the effect of each independent variable on each dependent variable in all analyses.

Sensitivity analyses using cut-off-based categorization rather than quantified dcCL values were performed to further validate the relationship between cholesterol levels and Aβ uptake. We used logistic regression analysis with each cholesterol type (LDL-c, HDL-c, and triglyceride) as a predictor and Aβ positivity as an outcome after controlling for age, sex, BMI, APOE4 genotype, hypertension, diabetes, lipid-lowering medication, and dementia medication. In addition, we performed a single model with all cholesterol type (LDL-c, HDL-c, and triglyceride) together as predictors after age, sex, BMI, APOE e4 allele (APOE4) genotype, hypertension, diabetes, lipid-lowering medication, and dementia medication.

All reported p-values were two-sided and the significance level was set at 0.05. All analyses were performed using SAS version 9.4 (SAS Institute, Inc., Cary, NC, USA).