Categorization of participants
The cognitively normal (CN) participants were classified as CN- (PiB-PET negative, n = 185) and CN+ (PiB-PET positive, n = 36) groups. The details were shown in Table 1. There were no significant differences in sex, education level, mini-mental state examination (MMSE) score, apolipoprotein E ɛ4 (ApoE4) positivity, and clinical dementia rating (CDR) scores. In addition, the demographic data according to their age-groups (young, mid-age, and old-age) were shown in Supplementary Table 1.
Characterization of QPLEXTM biomarkers without any intervention of AD pathology
We had previously reported that our QPLEXTM biomarkers (beta-amyloid1-40, Ab1-40; periostin, POSTN; galectin-3 binding protein, LGALS3BP; angiotensin-converting enzyme, ACE) showed significant differences between age-matched (age > 55 years old) PiB− and PiB+ participants (14). In the current study, our purpose was to utilize the biomarkers to distinguish whether amyloid accumulates within the CN group (CN- vs CN+; cognitively normal participants of all ages). Before the start, we tested the effects of age and sex, which are generally known as confounders of AD, on our biomarkers (Figure 1A and 1B). Note that we only used CN- samples for Figure 1 to exclude the intervention of AD pathology from testing the confounders (age, sex). Interestingly, our monotone regression spline analysis found three different curve patterns of biomarkers according to aging (Figure 1A). The cut-off dividing lines (42 years old, 65 years old) separating age-groups were determined by the vertexes of each biomarker curve (POSTN and ACE showed the first vertex at the point of 42 years old; POSTN, ACE, and LGALS3BP showed the second vertex at the point of 65 years old; Ab1-40 showed unclear vertex). According to this cut-off criterion, we defined three age-groups as Young, Mid-age, and Old-age groups. Interestingly, the second cut-off point (65 years old) is exactly matched with the standard age for late-onset AD (Figure 1A) (18). Next, we identified that there were no differences in the biomarkers between male and female (Figure 1B). Furthermore, all of the biomarkers showed significant correlations between them (Figure 1C) but with low variance inflation factors (VIF < 2), which means they are in the most suitable situation to be used as variables for the logistic regression.
Age-dependent approaches for identifying the correlation between the QPLEXTM markers and cerebral amyloid deposition
Since the QPLEXTM biomarkers showed different patterns between the age groups, we checked relationships between the biomarkers and cerebral amyloid deposition (standard uptake value ratio; SUVR) for each age-group (Figure 2). The correlations between cerebral amyloid deposition and the QPLEXTM biomarkers were not significant in Young and Mid-age groups. In the Old-age group, we confirmed that periostin (POSTN) and galectin-3 binding protein (LGALS3BP) levels in blood showed remarkably significant correlations with cerebral amyloid deposition (Figure 2C). Thus, we decided to focus on the Old-age group and further investigated whether the QPLEXTM biomarkers can be utilized to discriminate PiB-PET positivity within this group. Young and Mid-age groups were not appropriate for this analysis because they had a too small number of PiB+ participants (Supplementary Table 1). However, since this model was only limited to old-age participants, we thought it was an insufficient model for satisfying our goal, which targets all ages of CN participants. Therefore, we decided to generate new models for all ages with the correction for covariates (age, sex) to compensate for the age-dependent difference in the level of biomarker pattern and the significant difference of age between CN- and CN+ (Supplementary Figure 1-2).
Relationship between the QPLEXTM biomarkers and cerebral amyloid deposition in Old-age group (>65 years)
First, we performed a comparison analysis between CN- and CN+ in the Old-age group. Although the biomarkers did not show any differences between the sexes (Figure 1B), we included this variable for a covariate because it has also been a well-known confounder of AD diagnosis (19). Interestingly, two biomarkers, periostin (POSTN) and galectin-3 binding protein (LGALS3BP) showed significant differences between CN- and CN+ (Figure 3A-D, left). The overall level of POSTN in blood was significantly increased in individuals with cerebral amyloid deposition above the threshold, whereas LGALS3BP showed the opposite tendency. Partial correlation analysis with correction for age and sex also revealed a significant association between the blood biomarker levels and cerebral amyloid deposition (Figure 3A-D, right). POSTN was positively correlated with cerebral amyloid deposition (Pearson’s correlation coefficient R = 0.2186, P-value = 0.0333), whereas LGALS3BP was negatively correlated (Pearson’s correlation coefficient R = -0.2626, P-value = 0.0094). When all age-groups were combined, all QPLEXTMbiomarkers except for ACE showed significant differences and correlations, shown in Supplementary Figure 1. Furthermore, the results from the analysis of covariance (ANCOVA) showed that All CN+ group had significantly different levels of Ab1-40, POSTN, and LGALS3BP than All CN− group (Supplementary Table 2). Our results showed the possibility of the QPLEXTM biomarkers as variables for the cerebral amyloid deposition, except for ACE.
Discriminative ability of the QPLEXTM Alz plus assay for CN- vs CN+ in Old-age group (>65 years)
Before further analyses for the practical use, we had to determine whether Ab1-40 and ACE could be used as variables because they did not show the significant results from the previous analyses (Figure 2-3, Supplementary Table 3). However, when we performed multiple regression analysis on all CN groups, we found that all QPLEXTM biomarkers, including Ab1-40 and ACE, showed a significant correlation with cerebral amyloid deposition (SUVR) in every combination of variables (Table 2). Thus, we decided to include Ab1-40 and ACE in our biomarker panel because they were verified to be useful variables in explaining cerebral amyloid deposition level, based on multiple regression analysis. Subsequently, we performed logistic regression and receiver operating characteristic (ROC) curve analysis to identify the discriminative ability of our QPLEXTM Alz plus assay (Figure 4A). When three ROC curves were compared with each other, we found adding QPLEXTM biomarkers dramatically increased area under curve (AUC) (0.622 to 0.878, Curve I vs Curve III; 0.684 to 0.878, Curve II vs Curve III) (Figure 4B). Each graph had high sensitivities or specificity (Curve I, 82.4% sensitivity and 42.9% specificity; Curve II, 52.9% sensitivity and 73.5% specificity; Curve III, 69.7% sensitivity and 98.4% specificity) (Figure 4C). When all age-groups were combined, QPLEXTMAlz plus assay still retained its discriminative ability, which is shown in Supplementary Figure 2. Therefore, the results suggest that QPLEXTM Alz plus assay can also be used for pre-screening amyloid deposition in the brain even when there are no apparent symptoms of cognitive disorders.