Social Networks and Cerebrospinal Fluid Biomarkers of Alzheimer’s Disease Pathology in Cognitively Intact Older Adults: The CABLE Study

CABLE study

CABLE is an ongoing large-scale cohort study initiated in 2017, mainly focusing on AD biomarkers and risk factors in the Chinese Han population [25, 26]. CABLE aims to identify the genetic and environmental modifiers for AD thereby advancing disease prevention and early diagnosis. All participants were administered clinical and neuropsychological assessments, blood and CSF sample collection, as well as biochemical testing. Demographic profile, possible risk factors of AD, and medical history data were acquired by comprehensive questionnaires and electronic medical record systems. All enrolled participants were Han Chinese aged between 40 to 90 years. The participants were excluded if they were diagnosed with: 1) epilepsy, multiple sclerosis, central nervous system infection, or other major neurological disorders; 2) major psychological disorders (e.g., severe depression, psychotic disorders); 3) severe systemic diseases that may influence neurochemical biomarkers of AD; 4) family history of genetic disease [25]. The protocol for the CABLE cohort in compliance with the Helsinki declaration was reviewed and approved by the Institutional Ethics Committee of Qingdao Municipal Hospital. All participants provided written informed consent or assent with proxy consent. This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Included participants

A total of 1,273 cognitively intact individuals who received systematic cognition assessments were initiatively recruited as part of the CABLE study, which may include a subset of individuals with abnormal CSF AD biomarkers characteristic of preclinical AD. This setting allows seeing modification impacts of social networks on these neurobiological alterations of AD pathology in an early stage. The general cognitive functions of included participants were evaluated by the recommended Subjective Cognitive Decline Initiative (SCD-I) basic scale [23, 27], the China-modified Mini-Mental State Examination (CM-MMSE) [25, 29], and the Montreal Cognitive Assessment (MoCA) [30]. Potential depressive symptoms were additionally assessed by the 17-item Hamilton depression rating scale, consistent with previous studies demonstrating abnormal depressive events using cutoffs higher than seven scores [31]. We followed the diagnostic recommendations incorporated in the National Institute on Aging–Alzheimer’s Association (NIA-AA) workgroup reports for the diagnosis of preclinical AD [32]. The cognitive status of SCD was determined according to the recommendation of the SCD-I Working Group [23, 27]. Like preclinical AD, SCD manifesting before cognitive impairment could act as a target population for early intervention trials in AD [24]. In addition, SCD is a harbinger of greater rates of clinical progression in preclinical AD [33]. All support promising evidence reflecting the at-risk nature of the SCD group. Therefore, we not only pay attention to a population during the preclinical AD stage but target SCD individuals expecting to advance perceptions of regulation mechanisms covering all risk groups. Each participant received a consensus diagnosis by the professional medical doctors through intact performance on neuropsychological testing, combined with CSF biomarkers and brain magnetic resonance imaging (MRI) examinations. After removal of extreme outliers, finally, 886 participants who had available social network and CSF biomarkers data were included in this cross-sectional study, which consisted of SCD individuals and non-SCD controls. Total of them were further grouped as preclinical AD subjects (n = 295) and healthy controls (n = 591) (Supplementary Figure 1).

The assessment of social networks

The social networks were measured using self-reported questionnaires about social ties (with friends, relatives, and neighbors) [11]. This study evaluated the closeness of every social tie by recording how subjects got along with their friends, relatives, and neighbors. Subjective options for the closeness in these questionnaires included distant, moderate, and close social ties, which were scored zero, one point, and two points in order. Social ties rating was completed in a blinded manner, regardless of other assessments and procedures. Also, the sum of scores of three social ties was adopted to construct a social network score ranging from 0 to 6 points, with a higher score indicating a closer social relationship [7, 11].

Measurements of CSF AD biomarkers

CSF was collected in 10 ml polypropylene tubes by lumbar puncture in the morning after overnight fasting and was gently mixed to avoid gradient effects. The samples were centrifuged at 2000 g for 10 min, snap-frozen, and stored at –80°C until assay. The separation and storage of samples were required within 2 h after standard lumbar puncture. Aβ_1–42, Aβ_1–40, T-tau, and P-tau were detected by the enzyme-linked immunosorbent assay (ELISA) kit (Innotest β-AMYLOID (1–42), β-AMYLOID (1–40), PHOSPHO-TAU (181p), and hTAU-Ag; Fujirebio, Ghent, Belgium) on the microplate reader (Thermo Scientific™ Multiskan™ MK3). The standards and samples were detected in duplicates, and the means of the duplicates were used for the statistical analyses. The mean inter-assay coefficient of variation was controlled under 15% (8.15% for Aβ_1–40, 8.79% for Aβ_1–42, 11.34% for P-tau, and 10.38% for T-tau). The mean intra-assay coefficient of variation was controlled under 10% (3.69% for Aβ_1–40, 5.27% for Aβ_1–42, 2.42% for P-tau, and 4.73% for T-tau). All analyses were performed by board-certified laboratory technicians who were blind to clinical information and diagnoses.

APOE genotyping

DNA was extracted from overnight fasting blood samples with the QIAamp®DNA Blood Mini Kit. And the extracted DNA was separated and stored at –80°C until APOE ɛ4 genotyping was performed. Specific loci related to APOE ɛ4 status (rs7412 and rs429358) were selected for genotyping with restriction fragment length polymorphism (RFLP) technology. The results were dichotomized into APOE ɛ4 carrier or non-carrier status.

Statistical analyses

Demographic characteristics were summarized as mean (SD) for continuous variables and number (column percentage) for categorical variables. p values for comparing the between-group difference using Chi-square tests for categorical variables, as well as using t-test (normally distributed data) or Mann-Whitney U test (skewed distribution) for continuous variables. As for the CSF AD biomarkers, the transformation was performed to achieve or approximate a normally distributed data (Kolmogorov-Smirnov test p > 0.05) via “car” package of R software in case of skewed distribution. Individuals who had a value which was three-fold standard deviation (SD) greater or smaller than the mean value were considered as extreme outliers in statistical data and were removed from this research cohort.

In step 1, we used the original social ties data from questionnaire and first analyzed them as three-category variables in the multiple linear models with age (continuous), sex (female = 0, male = 1), education (continuous), and APOE4 status (“44” or “43” or “42” = 1, other genotypes = 0) as covariates. In step 2, to assess the composite effects of the social networks on CSF AD biomarkers, we analyzed a social network summary index scored as a continuous variable in the linear model with same covariates. Additional factors including history of stroke (yes or no), hypertension (yes or no), diabetes (yes or no), marital status (married = 0, unmarried = 1, divorced = 2, being a widow or widower = 3), living status (living alone or not), physical exercise (regular exerciser or not), depressive symptoms (abnormal or normal), and CM-MMSE (continuous) were further corrected to exam the robustness of primary results and confounding effects. In step 3, we included interaction terms in the models to detect modification effects of sex, genetic factor, and cognition on these associations. Subgroup analyses were further performed based on factors that interacted with social networks to clarify an at-risk population.

We detected whether some underlying relationships of social networks with CSF biomarkers of AD pathology differed in preclinical AD and healthy control groups. In accordance with the biomarker-based definition of preclinical AD, asymptomatic β-amyloidosis (A +) was operationally determined by abnormal CSF assay of Aβ_1–42 and “downstream” neurodegeneration (N +) was defined by abnormal CSF assays of T-tau or P-tau [32, 34]. In the present study, the positive β-amyloidosis group represented preclinical AD [22, 34]. CSF biomarkers were binarized as abnormal and normal according to the finding that approximately one-third of cognitively intact older adults suffered from AD pathology [22, 36]. Similar distributions were also observed in Asian populations [37, 38]. Details regarding the CSF evidence of AD pathophysiology have been illustrated in our previous CABLE studies. Therefore, the levels of CSF Aβ_1–42 in the lower one-third of the distribution (A + , <116.74 pg/ml) and P-tau (N + , >39.00 pg/ml) or T-tau (N + ,>180.26 pg/ml) in the upper one-third of the distribution were considered as abnormal.

Statistical significance was set at a two-sided p < 0.05. Bonferroni method was used for multiple corrections except where specifically noted. All statistical analyses were performed using stats package in R version 3.5.1 software program.

Characteristics of participants

Table 1 summarizes the demographic characteristics and distributions of CSF biomarkers of AD in the whole participants. The mean age of included cognitively intact participants was 61.91 (SD = 10.51) years. The proportion of males was 58.92%, and the proportion of APOE ɛ4 carriers was 15.44%. As for detailed subcomponents of social ties, individuals that had distant ties with friends were older (p = 0.005) and showed lower years of education (p = 0.009) and lower CM-MMSE scores (p < 0.001) (Supplementary Table 1). There was a marked difference only in CM-MMSE performance regarding social ties with relatives (p = 0.001).

Social networks and CSF biomarkers of AD pathology in total participants

Firstly, we investigated whether social networks could affect the metabolisms of CSF proteins of AD pathology in all participants with normal cognition. Social network subcomponents showed several suggestive (p < 0.05) or significant (<0.008 after Bonferroni correction) associations with the levels of CSF amyloid proteins and indicators of multi-pathologies of AD in the whole participants (Fig. 1). Individual with relatively close social ties with relatives (close: β= 0.0015, p = 0.04) and neighbors (moderate, β= 0.0017, p = 0.001; close, β= 0.0015, p = 0.002) presented a higher level of CSF Aβ_1–42 (Supplementary Table 2). Notably, moderate, or close social ties (with friends, relatives, and neighbors) were associated with higher CSF Aβ_1–42/Aβ_1–40 and lower CSF P-tau/Aβ_1–42 and CSF T-tau/Aβ_1–42 (Supplementary Tables 2-3). No associations were found between social network subcomponents with CSF P-tau and T-tau.

OPEN IN VIEWER

Fig. 1

Correlation coefficient maps for social networks and CSF biomarkers of AD core pathology. The social networks showed several suggestive or significant associations with CSF indicators of multiple pathologies of AD in the whole participants, especially in APOE ɛ4 carriers and subjects with SCD. The color-coding and the number in each cell represent the transformed correlation coefficient between social networks and CSF biomarker of AD pathology. The absolute value in each cell represents the log-transformed absolute coefficient. Positive or negative number respectively reflect possible positive or negative correlation inferred from the linear model. “a” indicated a suggestive p value (<0.05) for social network subcomponents in associations with CSF biomarkers of AD pathology. “b” indicated a significant p value (<0.008) for social network subcomponents in associations with CSF biomarkers of AD pathology after Bonferroni correction for 6 tests. Aβ, amyloid-β; T-tau, total tau; P-tau, phosphorylated tau, APOE ɛ4, apolipoprotein E ɛ4 allele; SCD, subjective cognitive decline.

The multivariable linear models next detected the associations of social network score as a comprehensive continuous variable with levels of CSF biomarkers (Fig. 1). The social network score was positively associated with CSF Aβ_1–42/Aβ_1–40 (β=0.1156, p = 0.01) and had negative links with levels of CSF T-tau/Aβ_1–42 (β= –0.0254, p = 0.04) and P-tau/Aβ_1–42 (β= –0.0115, p = 0.03) (Supplementary Table 4). To clarify whether some covariates could confound the associations between the social network and CSF biomarkers, we repeated another model after adding potential terms of CM-MMSE, living status, marital status, chronic diseases (including stroke, hypertension, and diabetes mellitus), physical exercise, and depression. Collectively, the correlations with CSF Aβ_1–42/Aβ_1–40 and CSF P-tau/Aβ_1–42 were persisted and robustness (Supplementary Table 4).

Modification effects of APOE ɛ4 and subjective cognitive decline

Next, we detected whether potential genetic factor and cognitive status could modify the aforementioned associations. We noted significant interactions of social network score with APOE ɛ4 carrier status or with cognitive status (SCD or controls) (Supplementary Table 5). Further stratified results supported the modification effects of genetic factor and cognitive status. However, there were nonsignificant interaction effects by sex, indicating the associations independent of sex difference (Supplementary Table 5).

In APOE ɛ4 carriers, the social network score was inversely associated with the level of CSF P-tau (β= –0.044, p = 0.007) (Supplementary Table 6 and Fig. 1). While the link with CSF Aβ_1–42/_1–40 in APOE ɛ4 non-carriers (β= 0.122, p = 0.01) did not reach a significant level after Bonferroni correction (p > 0.008). Similarly, individuals with higher social network scores showed higher CSF Aβ_1–42/_1–40 (β= 0.228, p < 0.001) and lower CSF P-tau/Aβ_1–42 (β= –0.021, p = 0.007) and CSF T-tau/Aβ_1–42 (β= –0.045, p = 0.009) (Supplementary Table 6 and Fig. 1) in the SCD group. No correlation was revealed in those without SCD (i.e., controls).

Social networks associated with CSF biomarkers of multiple pathologies in preclinical AD stage

We finally repeated association explorations in an early risk stage, namely preclinical AD, intending to pinpoint advantageous influences of social networks in a particular group. Some underlying correlations were more pronounced in individuals with preclinical AD in comparison with cognitively intact participants without biomarkers evidence of AD (i.e., controls).

Linear regression models adjusted for age, sex, years of education, and APOE ɛ4 status in the preclinical AD group showed associations between increased social network score and higher CSF Aβ_1–42/_1–40 (β= 0.481, p = 0.001) and lower CSF P-tau (β= – 0.0006, p = 0.04), CSF T-tau (β= –0.005, p < 0.001), CSF P-tau/Aβ_1–42 (β= –0.039, p = 0.03), and CSF T-tau/Aβ_1–42 (β= –0.047, p < 0.001) (Fig. 2 and Supplementary Table 7). As for control groups with or without neurodegeneration, no significant associations were revealed after adjustment for covariates, although the univariable results implied that individuals with higher social network scores showed higher CSF Aβ_1–42/_1–40 (β= 0.314, p = 0.02) in the control group with neurodegeneration (Fig. 2). All results were barely changed in the sensitivity analyses after adjustment for the potential confounders (Supplementary Table 8).

OPEN IN VIEWER

Fig. 2

The associations of social network score with CSF AD biomarkers in preclinical AD and controls. In preclinical stage of AD, a higher social network score was associated with lower CSF T-tau (A) and CSF T-tau/Aβ_1–42 (B), and higher CSF Aβ_1–42/Aβ_1–40 (C), while no such relationships were noted in control 1 and control 2 groups. The red line represents the fitted univariable linear relationship in preclinical stage of AD. The green line represents the fitted univariable linear relationship in controls without β-amyloidosis and neurodegeneration (controls 1). The black line shows the fitted linear relationship in controls with neurodegeneration in the absence of β-amyloidosis (controls 2). Covariates of adjusted models comprised age, sex, years of education, and APOE4 status. CSF, cerebrospinal fluid; Aβ, amyloid-β; T-tau, total tau.