Changes in CSF sPDGFRβ level and their association with blood–brain barrier breakdown in Alzheimer’s disease with or without small cerebrovascular lesions

The CANDI study was launched in 2018; it was a longitudinal study including individuals with normal cognition (CN), mild cognitive impairment (MCI), and dementia. All participants were recruited from the First Affiliated Hospital of the University of Science and Technology of China (USTC) in November 2018. CSF samples, plasma samples, imaging data, and cognition measurements were available and were used for this study [17].

For our study, 158 participants were from the CANDI cohort, and complete clinical data were required, after a detailed cognitive assessment, including the Mini-Mental State Examination (MMSE) and clinical dementia rating (CDR) scores (Fig. S1). Twenty-seven participants (MMSE ≧ 24), with conditions such as primary headache, facial neuritis, dizziness, and ophthalmoplegia, with normal CSF cell numbers and protein to exclude intracranial hemorrhage, infections and inflammation of the central nervous system, and significant blood–brain barrier damage, were included in the CDR 0 group. A total of 131 patients with cognitive impairment (CI) were divided into two groups: the CDR 0.5 group (48 cases) and the CDR 1–2 group (83 cases) based on the NIA-AA criteria (2011) and CDR score [19]. These criteria excluded vascular dementia (defined by a history of a stroke temporally related to the onset or worsening of cognitive impairment or the presence of multiple or extensive infarcts or severe white matter hyperintensity burden), Lewy body dementia, behavioral variant frontotemporal dementia, Parkinson’s disease dementia, semantic variant primary progressive aphasia, non-fluent/agrammatic variant primary progressive aphasia or other active neurological diseases, or a non-neurological medical comorbidity that could have a substantial effect on cognition. Meanwhile, to further investigate the association of CSF sPDGFRβ with Aβ pathology, we reclassified the whole cohort, and cases that were positive based on the ¹⁸F-florbetapir PET and/or CSF Aβ42/40 ratio were included in the A+ group. All cases in the CDR 1–2 group were A+, and only 3 cases were A+ in the CDR 0 group, but in the CDR 0.5 group, there were 20 A− cases and 28 A+ cases, using accepted cutoff values (0.06423) and/or visual judgment for ¹⁸F-florbetapir PET in the CANDI study [17, 20]. APOE genotypes were determined as described previously [21]. Each patient in this study provided written informed consent in accordance with the Declaration of Helsinki. The protocols used in this study were reviewed and approved by the ethics committee of our hospital.

Noanticoagulative blood samples were collected by venipuncture during the collection of CSF specimens. The collected blood samples were centrifuged, and serum albumin concentrations were determined using a bromocresol green dye binding assay (ADVIA 1800; Siemens, Berlin, Germany). The Q-alb value was calculated using the following formula: (CSF albumin/serum albumin) × 1000.

All images were acquired by using GE DISCOVER 750w 3.0T MRI scanner (GE Healthcare, USA), including DWI/FLAIR/SWI and T2-weighted images, which were assessed by two experienced neuroradiologists blinded to clinical information. According to the recently described score for small vascular lesions [22], we rated the total MRI burden of CSVD on an ordinal scale from 0–4 by counting the presence of each of the four MRI features of CSVD: lacunes (1 point if ≥ 1 lacune present), any cerebral microbleed (1 point if present), moderate to severe perivascular spaces (grade 2–4) in the basal ganglia (1 point if present), and periventricular white matter hyperintensities (WMH) meeting or exceeding Fazekas scale 3 and/or deep WMH meeting or exceeding Fazekas scale 2–3 (1 point if present). Based on the results of the analysis, the patients were divided into three groups: CSVD burden scores of 0, 1, and 2–4.

The IBM SPSS 23.0 software for Windows (SPSS; Chicago, IL, USA), GraphPad Prism 7.0 (GraphPad Software; La Jolla, CA, USA), and R 4.0.4 (ggplot2, ggpubr, mediation, and QuantPsyc) were used for the analysis. Statistical significance in all two-sided tests was defined as p value < 0.05. For describing demographic data, the Wilcoxon signed-rank test was used for continuous variable intergroup comparisons, and chi-square analysis was used for categorical variables. AD biofluid biomarker measurements including Aβ42, Aβ40, p-tau, t-tau, and the Aβ42/Aβ40 ratio were compared using analysis of covariance (ANCOVA), with age, sex, and the APOE genotype as covariates. ANCOVA was also used to compare CSF sPDGFRβ and Q-alb levels between different cognition groups. A multiple linear regression model was adopted to evaluate the associations between CSF sPDGFRβ, Q-alb, and AD biomarkers. In multiple linear regression models, ratios of Q-alb much greater or lower than the triple standard deviation from the mean value were regarded as outliers and discarded. The Q-albumin ratio and concentration measures of CSF sPDGFRβ were ln-transformed when necessary, and the Shapiro–Wilk test was used to test the normality of the transformed data. Age, sex, and APOE genotype were included in multiple linear regression models as covariances to adjust the effects. The extent of pericyte injury contributing to BBB damage was determined using mediation analysis. This was based on a multiple linear regression model adjusted for age, sex, and APOE genotype. Ranked data, such as CSVD scores, are shown as the median of the interquartile range (IQR).

As shown in Table 1, 158 patients were selected, including individuals in the CDR 0 group (27 cases), CDR 0.5 group (48 cases), and CDR 1–2 group (83 cases). In the CDR 1–2 group, the proportion of female patients was higher, as predicted (68.67%). The mean age of the CDR 0 group was significantly lower than that of the CDR 1–2 group (p = 0.013). MMSE scores were significantly different among the three groups. As expected, there were more APOE e4 gene carriers in the CDR 0.5 group (45.83%) and CDR 1–2 group (63.86%) than in the CDR 0 (14.81%) group.

The Q-alb exhibited no significant difference among the CDR 0, CDR 0.5 A−, CDR 0.5 A+, and CDR 1–2 groups (Fig. 1A). Interestingly, CSF sPDGFRβ in the CDR 0.5 A− and CDR 0.5 A+ groups was highly significantly elevated compared with that in CDR 0 patients (Fig. 1B, p < 0.001). However, its level was reduced in the CDR 1–2 group compared to that in the CDR 0.5 A− group (p = 0.013) and CDR 0.5 A+ group (p = 0.023). In addition, we compared the CSF sPDGFRβ levels between subjects with a different APOE e4 gene carrier status or different sex. In the CDR 0, CDR 0.5, or CDR 1–2 groups, no significant difference was observed between APOE e4 gene carriers and non-carriers or males and females (Fig. S2A and Fig. S2B). Meanwhile, CSF sPDGFRβ levels correlated with age (p = 0.040, Fig. S2C). We calculated the contributions of age, sex, and APOE e4 genotype to the variance in CSF sPDGFRβ levels. The proportions of explained variances in age, sex, and APOE genotype were 3.10%, 3.21%, and 0.14%, respectively, for CSF sPDGFRβ (Fig. S2D). Moreover, the correlation between Q-alb and sPDGFRβ levels was also assessed. As shown in Fig. 1C, D, there was no correlation between Q-alb and sPDGFRβ in all participants or in the CDR 1–2 group. However, a strong positive correlation between Q-alb and sPDGFRβ was observed in the CDR 0–0.5 group (β = 0.314, p = 0.008).

We then performed multiple linear analyses to evaluate the association between CSF sPDGFRβ and t-tau, p-tau, Aβ40, Aβ42, and the Aβ42/Aβ40 ratio. For patients with a CDR of 0–0.5, significant correlations were observed between CSF sPDGFRβ and p-tau (β = 0.302, p = 0.014), t-tau (β = 0.350, p = 0.003), and Aβ42/Aβ40 ratio (β = − 0.249, p = 0.049). In patients with CDR 1–2, there were positive correlations between sPDGFRβ and p-tau (β = 0.342, p = 0.003), t-tau (β = 0.374, p < 0.001), Aβ40 (β = 0.330, p = 0.004), and Aβ42 (β = 0.246, p = 0.027). Q-alb was correlated with t-tau (β = 0.266, p = 0.024) in the CDR 0–0.5 group, whereas a significant correlation was observed between Q-alb and Aβ42/Aβ40 (β = 0.294, p = 0.013) in the CDR 1–2 group (Table 2).

BBB breakdown and pericyte injury have been suggested to occur before cognitive impairment in patients [5]. Meanwhile, it has been suggested that Aβ pathology is suggested to be associated with BBB damage [15, 23]. To investigate whether the association between Aβ pathology and BBB damage was regulated by CSF sPDGFRβ, we performed a mediation analysis [24]. As shown in Fig. 2, mediation analysis revealed that from the cognitively unimpaired period to the early stage of cognitive impairment (CDR 0–0.5), CSF sPDGFRβ-mediated effect accounted for a statistically significant proportion of Aβ toxic effects on the BBB (p = 0.038) (28.5% for Aβ42/Aβ40). Meanwhile, there was no significant direct effect (i.e., ADE) or CSF sPDGFRβ-mediated effects (i.e., ACME) of Aβ42 or Aβ40 on Q-alb ratios. However, in the CDR 1–2 group, Aβ42/Aβ40 changed its CSF sPDGFRβ-mediated effect on BBB damage (1.16%, p = 0.862) to a direct effect on BBB damage (98.84%, p = 0.028).

As BBB breakdown is closely associated with cerebral vascular damage, it is necessary to analyze whether the association between CSF sPDGFRβ and Q-alb varies in subjects with or without CSVD burden. The characteristics of CSVD burden in all participants were evaluated, which confirmed significant differences in the combined total CSVD burden (p < 0.001), PVWMH (p < 0.001), DWMH (p = 0.035), and cerebral microbleeds (p = 0.028) among the CDR 0, 0.5, and 1–2 groups (Table 3). The levels of Q-alb and CSF sPDGFRβ at each CDR stage with different CSVD burdens were compared. Q-alb was significantly increased only in individuals with higher CSVD burdens (scores of 2 and higher) (Fig. 3A and B). However, CSF sPDGFRβ levels were not significantly different among the three groups (Fig. 3C, D).

A 1-year longitudinal follow-up study was conducted in the CDR 0.5 group. Among these, 25 gradually progressing cases were defined as the progressive mild cognitive impairment (PMCI) group, and one patient met the diagnostic criteria for AD (Fig. 4A, p = 0.009). The remaining 19 patients with stable disease pathology were defined as the stable mild cognitive impairment (SMCI) group (Fig. 4B, p = 0.918). No significant difference was observed between the PMCI and SMCI groups regarding the baseline age, sex ratio, years of education, MMSE, APOE genotype, Aβ40, and Aβ42 levels (Table 4). However, the concentrations of t-tau (p = 0.004) and p-tau (p = 0.004) were significantly higher in the PMCI group, whereas the ratio of Aβ42/Aβ40 was lower in the PMCI group (p = 0.005) (Table 4). Q-alb showed no distinctive difference between the PMCI and SMCI groups (p =0.554, Fig. 4C). However, CSF sPDGFRβ was significantly higher in the PMCI group than in the SMCI group (p < 0.001, Fig. 4D), even when adjusted for Aβ42/40. Baseline CSF sPDGFRβ levels were correlated with MMSE yearly change in the CDR 0.5 group (β = − 0.400, p = 0.020, Fig. 4E) and CDR 0.5 A+ group (β = − 0.542, p = 0.019), while the correlation in the A− group was not statistically significant (Fig. 4F).