Cross-sectional and longitudinal data used in the current study were extracted from the ADNI database (adni.loni.usc.edu). ADNI was launched in 2003 as a collaborative effort between the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, the Food and Drug Administration, private pharmaceutical companies, and non-profit organizations. Study participants have been enrolled from more than 50 centers across the USA and Canada. The ADNI study aims to investigate whether a combination of biomarkers, including neurocognitive assessments, neuroimaging, and fluid markers, can be used to track the progression of mild cognitive impairment (MCI) and early AD dementia (Petersen et al., 2010). This study received approval from the institutional review boards of all participating institutions, and written informed consent was provided by all study subjects or their authorized representatives.

Our study sample included individuals with normal cognition (NC) and MCI (Petersen et al., 2010). Assignment to the NC group required a Mini-Mental State Examination (MMSE) (Folstein et al., 1975) score between 24 and 30 and a Clinical Dementia Rating (CDR) (Morris, 1993) score of 0. The diagnostic criteria for MCI included an MMSE score ranging from 24 to 30, a CDR of 0.5, the presence of a subjective memory complaint, evidence of objective memory loss as assessed by education-adjusted scores on the Wechsler Memory Scale Logical Memory II and essentially preserved ability to perform daily living activities. A total of 1821 participants with NC and MCI from ADNI1, ADNI-GO, ADNI2, and ADNI3 were initially considered for this analysis. From these, we selected 340 participants who met several criteria: availability of baseline plasma VWF data, at least two structural MRI scans, and demographic and clinical data at baseline, such as age, sex, education, and APOE4 genotype.

To examine the relationship between plasma VWF levels and cognitive decline over time, we used MMSE and Clinical Dementia Rating-Sum of Boxes (CDR-SB) (Williams et al., 2013) scores as cognitive outcomes. The MMSE is a measure of global cognition, with scores ranging from 0 to 30. Lower scores indicate greater cognitive impairment. The CDR-SB is an assessment of cognitive and functional capabilities, with scores ranging from 0 to 18. Higher scores indicate more severe cognitive impairment.

The imaging protocols for the ADNI are detailed on the official website¹ and have been previously documented (Jack et al., 2008). The ADNI team at the University of California, San Francisco, utilized FreeSurfer image analysis software² to process the T1-weighted sagittal 3D MPRAGE sequences obtained from MRI scans. From the ADNIMERGE dataset, we derived the volumes of several key brain regions, including the hippocampus, entorhinal cortex, fusiform gyrus, middle temporal gyrus, ventricles, and the whole brain. To normalize for variations in head size that could affect brain volume, we employed the following formula to calculate the adjusted volumes: Adjusted volumes = (raw regional brain volume / total intracranial volume) × 1,000. While there is no consensus on which normalization method should be preferred over another (Voevodskaya et al., 2014), the proportional approach was chosen because of its ease of application in practice and its intuitiveness.

Plasma levels of VWF were measured using a multiplex immunoassay panel based on Luminex technology, with specific procedures for plasma collection and measurement detailed on the website³ and described previously (Hu et al., 2012). In summary, a subset of plasma samples from the ADNI cohort was analyzed for VWF and other plasma proteins through a 190-analyte multiplex immunoassay panel. This panel, developed on the Luminex xMAP platform by Rules-Based Medicine (RBM), encompasses a broad range of proteins. The plasma VWF concentrations were reported in ug/mL. Samples were analyzed in singlicate, whereas quality control (QC) samples were run in duplicate. The lower assay limit of the analyte (i.e., VWF) is 0.7, the least detectable dose was 2.8, the RBM low plasma range was 5.3, and the RBM high plasma range was 74. To evaluate the variability across distinct concentration ranges for the analyte, three levels of QC (low, medium, and high) were run, yielding coefficient of variation (CV) values of 28.8, 14.6, and 13.7%, respectively. For our statistical analysis, we employed quality-controlled plasma VWF values, which were then log-transformed.

In cross-sectional analyses, descriptive statistics were performed to summarize the sample data. For example, mean (SD) was used to summarize continuous variables, and sample size (percentage) was used for categorical variables. Pearson’s correlation tests were conducted to investigate the correlations between plasma VWF levels and cognitive assessments, as well as structural MRI neuroimaging markers, among older adults without dementia. In longitudinal analyses, a total of eight linear mixed-effects models were built for eight different dependent variables, including cognitive measures (MMSE and CDR-SB) and neuroimaging markers (volumes of the hippocampus, entorhinal cortex, middle temporal gyrus, fusiform gyrus, ventricles, and whole brain). Each model adjusted for several potential covariates, including age, sex, education, and APOE4 status. For neuroimaging models, we additionally adjusted for baseline MMSE scores. Each model included the main effects of plasma VWF and potential covariates, as well as their interactions with time (follow-up duration in years). A random intercept for each participant was included in the model. All statistical analyses were conducted using R statistical software. We set the level of statistical significance at a two-tailed p-value of less than 0.05.

Table 1 demonstrates the demographic characteristics and neuroimaging data of the study sample: 340 older adults without dementia, comprising 52 cognitively unimpaired participants and 288 MCI subjects. The mean age in the study sample was 74 years (SD = 7), and the mean educational level was 16 years of education (SD = 3). Females constituted 38% of the total participants, and APOE4 carriers accounted for half of the study sample. The mean adjusted volumes of the hippocampus, entorhinal cortex, middle temporal gyrus, fusiform gyrus, ventricles, and whole brain were 4.21 (SD = 0.73), 2.19 (SD = 0.49), 12.10 (SD = 1.62), 10.59 (SD = 1.39), 26 (SD = 13), and 643 (SD = 42), respectively. The mean plasma VWF levels were 1.64 (SD = 0.36) ug/mL. The comparison of variables between CU and MCI individuals is also listed in Table 1.

To understand the potential role of VWF in cognitive function, the correlations between plasma VWF levels and cognitive measures were analyzed in this study sample using Pearson correlation tests (Figure 1). As shown in Figure 1A, plasma VWF levels were not correlated with MMSE score among older adults without dementia (r = −0.004; p = 0.94). Similarly, as shown in Figure 1B, levels of plasma VWF were also not related with CDR-SB score (r = 0.019; p = 0.73).

To understand the potential role of VWF in neurodegeneration, the correlations between plasma VWF levels and structural neuroimaging markers were analyzed in this study sample using Pearson correlation tests (Figure 2). As shown in Figure 2A, plasma VWF levels were not correlated with the volumes of the hippocampus among older adults without dementia (r = 0.01; p = 0.82). As shown in Figure 2B, levels of plasma VWF were also not related to the volumes of the entorhinal cortex (r = 0.04; p = 0.46). As shown in Figure 2C, levels of plasma VWF were also not related to the volumes of the middle temporal gyrus (r = 0.014; p = 0.8). As displayed in Figure 2D, levels of plasma VWF were not related to the volumes of fusiform gyrus (r = 0.06; p = 0.24). As demonstrated in Figure 2E, levels of plasma VWF were not related to the volumes of the ventricles (r = 0.03; p = 0.62). As displayed in Figure 2F, levels of plasma VWF were not related to the volumes of the whole brain (r = 0.002; p = 0.97).

Additionally, we further ran eight multivariable linear regression models for eight outcomes, adjusting for covariates. Model summaries are presented in Supplementary Tables S1, S2.

In a subset of 188 participants who had measurements for both plasma VWF and CSF AD biomarkers (i.e., Aβ42, total tau, and p-tau181 levels) at baseline, Pearson’s correlation tests were performed to examine the associations of plasma VWF levels with CSF Aβ42, total tau, and p-tau181 levels. No significant associations were found between plasma VWF levels and CSF Aβ42 (r = −0.036, p = 0.6), total tau (r = −0.008, p = 0.9), or p-tau181 (r = −0.02, p = 0.7) levels. Additionally, multivariable linear regression models were adjusted for several covariates, including age, education level, sex, and APOE4 status. The results remained unchanged. Plasma VWF levels were not significantly associated with CSF Aβ42 (coefficient = −86.2, SE = 78.9, p = 0.28), total tau (coefficient = −3.1, SE = 22, p = 0.89), or p-tau181 (coefficient = −0.7, SE = 2.5, p = 0.78) levels.

To explore the association between plasma VWF levels and longitudinal cognitive decline, the potential relationship between plasma VWF levels and changes in two cognitive measures was analyzed using linear mixed-effects models (Table 2). Plasma levels of VWF were significantly associated with changes in MMSE (lower scores represent worse cognitive performance; Coefficients [95% CIs] = 0.204 [0.030, 0.378]; p value = 0.021; Figure 3A) and CDR-SB (higher scores represent worse cognitive performance; Coefficients [95% CIs] = −0.268 [−0.374, −0.163]; p value < 0.001; Figure 3B) over time among older adults without dementia, suggesting that lower levels plasma VWF were associated with a steeper rate of cognitive decline. In addition, Supplementary Table S3 presents all coefficients of the main effects of the predictors.

To explore the association between plasma VWF levels and brain atrophy, the potential relationship between plasma VWF levels and changes in 6 structural neuroimaging markers was analyzed using linear mixed-effects models (Table 3). Plasma levels of VWF were significantly associated with changes in volumes of the hippocampus (Coefficients [95% CIs] = 0.016 [0.004, 0.027]; p value = 0.009; Figure 4A), entorhinal cortex (Coefficients [95% CIs] = 0.031 [0.014, 0.048]; p value < 0.001; Figure 4B), fusiform gyrus (Coefficients [95% CIs] = 0.047 [0.008, 0.085]; p value = 0.017; Figure 4D), and ventricles (Coefficients [95% CIs] = −0.380 [−0.558, −0.203]; p value < 0.001; Figure 4E) over time among older adults without dementia. In contrast, plasma levels of VWF were not associated with changes in volumes of the middle temporal gyrus (Coefficients [95% CIs] = 0.039 [−0.006, 0.083]; p value = 0.089; Figure 4C) or the whole brain (Coefficients [95% CIs] = 0.661 [−0.508, 1.830]; p value = 0.268; Figure 4F). In addition, Supplementary Table S4 presents all coefficients of the main effects of the predictors.

Additionally, for models with cognitive measures as outcomes, we further added cognitive status as a covariate. The results did not change (Supplementary Table S5).