Cerebrovascular and amyloid pathology in predementia stages: the relationship with neurodegeneration and cognitive decline

Two hundred seventy-one subjects were selected from memory clinics of the single-center BioBank Alzheimer Center Limburg (BBACL; n = 99) cohort and the LeARN (n = 50) [16] and DESCRIPA (Development of screening guidelines and criteria for predementia Alzheimer’s disease; n = 122) [17] multicenter studies. Inclusion criteria were (1) no diagnosis of dementia at baseline and (2) baseline data available for MRI and CSF measures. When subjects participated in more than one study, we included the data from the study with the longest follow-up. The medical ethics committee at each site approved the study. All subjects provided informed consent.

Clinical assessment included neuropsychological assessment and an assessment of medical history. Information on medical history (e.g., hypertension, diabetes, obesity) was provided by patients and/or their caregivers, or it was extracted from medical files. Neuropsychological assessment was performed according to local routine protocol of each site, including the Mini Mental State Examination (MMSE) and at least one test in the cognitive domains of memory and executive functioning. The delayed recall of a word list test was used to examine memory. For the BBACL and LeARN studies, the Rey Auditory Verbal Learning Test (RAVLT) was used [18]. For the DESCRIPA cohort, the RAVLT and Consortium to Establish a Registry for Alzheimer’s Disease word list were the primary memory tests used. DESCRIPA tests per center are described elsewhere [17]. The Trail Making Test part B (TMT-B) [19] was used to examine executive functioning. Raw scores on each test were converted to z-scores using local normative data. Z-scores below −5 (n = 7) were rounded to −5 to avoid bias through outliers in the data.

Diagnosis of MCI at baseline was made according to the criteria of Petersen [20]. Subjects with a z-score below −1.5 on the immediate recall or delayed recall of a word list test were classified as having amnestic MCI. Subjects with a z-score below −1.5 on any of the nonmemory tests were classified as nonamnestic MCI. Diagnosis of AD-type dementia at follow-up was made according to the criteria of the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition [21], and the National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer’s Disease and Related Disorders Association [22]. Etiological diagnoses of other types of dementia were made according to standardized clinical criteria for vascular dementia [23], frontotemporal dementia (FTD) [24], and dementia with Lewy bodies [25].

CSF was collected by lumbar puncture and thereafter centrifuged and stored at −80 °C in polypropylene tubes. CSF Aβ_1–42, t-tau, and p-tau were analyzed using the Innotest sandwich enzyme-linked immunosorbent assay (Innogenetics, Ghent, Belgium) in Gothenburg for the DESCRIPA cohort [17], in Amsterdam for the LeARN project [26], and in Nijmegen for the BBACL study [27]. To define abnormality of the CSF measures, the following predefined cutoffs were used: Aβ_1–42 ≤ 550 pg/ml, t-tau > 375 pg/ml, and p-tau₁₈₁ > 52 pg/ml [28].

The apolipoprotein E (APOE) genotype was determined in a subgroup of the sample (n = 165). Assessments were performed according to routine protocol at each site, as described elsewhere [17, 29].

All subjects were scanned according to the routine MRI protocol at each site (Additional file 1). Scanning was performed at 1.0 (n = 14), 1.5 (n = 108), or 3.0 (n = 149) Tesla, and all scans included a three-dimensional T1-weighted gradient echo sequence and a fast fluid-attenuated inversion recovery sequence. To determine MTA, the Scheltens MTA visual rating scale [30] was used. The score on the MTA scale ranges from 0 to 4 for each hemisphere. The summed score of both hemispheres was used, where an abnormal MTA was defined using a cutoff ≥ 2 [31]. White matter hyperintensities (WMH) were measured with the visually rated age-related white matter changes (ARWMC) scale (range 0–3) [32] for the DESCRIPA cohort and with the visually rated Fazekas scale (range 0–3) [33] for the BBACL and LeARN cohorts. For the ARWMC scale, a cutoff score ≥ 2 in at least one of the measured brain areas was used to define WMH status [34, 35]. For the Fazekas scale, a cutoff score ≥ 2 was also used to define WMH status [36].

To classify individuals into subgroups, we used Aβ as a measure of AD and WMH as a measure of CVD. Subjects were classified as Aβ + when CSF Aβ_1–42 levels were abnormal. Subjects were classified as WMH+ when the WMH score was high. We created four groups based on combinations of Aβ status and WMH status: Aβ − WMH−, Aβ − WMH+, Aβ + WMH−, and Aβ + WMH+.

We analyzed differences in clinical baseline and follow-up characteristics and neurodegeneration markers between groups using analysis of variance for continuous variables and chi-square tests for categorical variables. Prior to the continuous comparisons of biomarker values between groups (Tables 1 and 2), Aβ_1–42, p-tau, and t-tau values were log-transformed to approximate a normalized distribution required for statistical comparisons. The raw biomarker values are shown in the tables. Comparisons between Aβ/WMH groups regarding neurodegenerative markers (Table 2) were all corrected for demographics, study, and baseline diagnosis.

We included 271 individuals with a mean age of 65.6 (SD 9.0) years. One hundred sixty-five (61%) were female, and 153 (57%) had a diagnosis of MCI at baseline, of whom 99 (65%) were classified as having amnestic MCI. Follow-up data were available for 233 individuals (86%). The availability of follow-up data was not different among the Aβ/WMH groups (p = 0.396) or studies (p = 0.730). After a mean follow-up of 2.5 (SD 1.2) years, 46 (17%) subjects had progressed to dementia. The majority (80%) of the individuals who progressed to dementia had a clinical diagnosis of AD-type dementia.

Table 1 shows baseline and follow-up characteristics of the four Aβ/WMH groups. The group without pathology was younger (p < 0.001) and progressed less frequently to dementia than the other three groups (p < 0.001). We found no difference in the prevalence of several vascular risk factors between the four groups (hypertension, p = 0.563; obesity, p = 0.486; diabetes, p = 0.106). We found no difference in Aβ load between the group with only Aβ and the group with both Aβ and WMH pathologies (p = 0.502). Likewise, the proportion of WMH was not different between the two WMH+ groups (Aβ − WMH+ and Aβ + WMH+; p = 0.175).

Table 2 shows the values and frequency of abnormal neurodegenerative markers for the Aβ/WMH groups. We found that, compared with the group without pathology, MTA was more severe in the groups with only Aβ (p < 0.001) and with only WMH (p < 0.001), as well as in the mixed pathology group (p < 0.001). The Aβ + WMH+ group had higher MTA scores than the group with only WMH (p = 0.025) and the group with only Aβ (p = 0.002). t-tau was increased in all three groups with a form of pathology compared with the group without pathology (Aβ − WMH+, p < 0.001; Aβ + WMH−, p < 0.001; Aβ + WMH+, p = 0.047), but this effect was influenced by baseline diagnosis (Aβ/WMH group × baseline diagnosis, F = 3.20, p = 0.024). When stratified by diagnosis, the effect was found only in subjects with MCI. There was no difference in t-tau levels between the Aβ − WMH+, Aβ + WMH−, and Aβ + WMH+ groups, regardless of baseline diagnosis. p-tau was increased only in the group with only Aβ compared with the Aβ − WMH− group (p < 0.001), regardless of baseline diagnosis. The association between p-tau and Aβ/WMH group was influenced by APOE genotype because we found that the elevated p-tau levels in the Aβ + WMH− group were limited to APOE ε4 allele carriers (Aβ/WMH group × APOE status, F = 3.72, p = 0.013).

In the total sample, MMSE scores did not differ between the Aβ/WMH groups at baseline (Table 3, Fig. 1). Baseline MMSE scores of the individuals with MCI were lower than those of individuals with SCD, regardless of pathology (p < 0.001). In subjects with MCI, there was a difference in baseline MMSE score only between the Aβ − WMH− and the Aβ + WMH− groups (p = 0.020). In subjects with SCD, there was no difference in baseline MMSE scores. In the total sample, the groups with one or both pathologies declined in MMSE score over time (Aβ − WMH+, p = 0.014; Aβ + WMH−, p = 0.035; Aβ + WMH+, p = 0.045), whereas scores remained stable in the Aβ − WMH− group (p = 0.793). The rate of decline was higher in the Aβ − WMH+ group than in the Aβ − WMH− group (p = 0.035). There were no differences in the rate of decline between the three groups with pathology. In subjects with SCD, the Aβ/WMH groups showed no decline over time. In subjects with MCI, the results were similar to those found in the total sample. Baseline delayed recall memory scores were lower in the three groups with pathology than in the group without pathology (Aβ − WMH+, p = 0.004; Aβ + WMH−, p < 0.001; Aβ + WMH+, p = 0.009), which was not influenced by baseline diagnosis. None of the groups showed significant decline over time. TMT-B scores did not differ at baseline between the groups and did not change during follow-up (Table 3, Fig. 1). APOE genotype did not influence any of the baseline or longitudinal associations.

Table 4 and Fig. 2 show the risk of progression to dementia for the Aβ/WMH groups. Compared with the group without pathology, the groups with a form of pathology have an increased risk of progressing to dementia (Aβ- WMH+ HR: 3.25, p = 0.021, Aβ + WMH- HR: 4.89, p < 0.001, Aβ + WMH+ HR:3.00, p = 0.036), but this was influenced by baseline diagnosis (Aβ/WMH group*baseline diagnosis: HR = 2.89; p = 0.007) as the effect was mainly attributable to MCI subjects (Fig. 2). There was no difference in progression rates between the groups with isolated or coexisting Aβ/WMH pathology, when analyzing the total sample or only MCI subjects. Results were similar when using progression to AD-type dementia as outcome.

Because both Aβ and WMH were associated with MTA, we tested the interaction between the two pathologies on MTA. General linear model analyses showed no interaction between Aβ and WMH on MTA, using dichotomous variables created with cutoff points (p = 0.770) or continuous variables (p = 0.631).

We repeated the main analyses after exclusion of subjects with CSF Aβ_1–42 values 10% around the cutoff. The results remained similar after this exclusion. The results were also comparable after exclusion of the four subjects in the Aβ − WMH− group who progressed to FTD at follow-up, as well as when repeating the analyses using Aβ and tau for classification of AD profiles instead of only Aβ.

We conducted age sensitivity analyses in which we age-matched the Aβ/WMH groups by selecting only individuals between 64 and 79 years of age. Most results were similar to the original results. Results that were different are shown in Additional file 2. Associations that showed a similar direction but no longer reached significance because of a reduction in sample size (p values between 0.05 and 0.09) were considered unchanged.

Observed power calculations were done for the main analyses. For the comparisons of neurodegenerative markers (Table 2), the observed power ranged from 0.71 to 0.99. For the comparisons of cognitive performance and decline (Table 3), the observed baseline power ranged from 0.37 to 0.67, and for the slopes it ranged from 0.34 to 0.61. Regarding the comparisons in progression to dementia (Table 4), the observed power was 0.66.

The association between AD pathology and MTA has been well characterized in the literature. The first neuropathological changes underlying AD are thought to occur in the medial temporal lobe [37]. The relationship between MTA and CVD, however, is still somewhat controversial. Some studies have found that CVD was associated with MTA [12, 15, 38], whereas others did not find this relationship [7, 13]. Our results support an association between CVD and MTA, although we measured only one aspect of CVD (i.e., WMH). Interestingly, we found that MTA was most severe in the group with mixed Aβ/WMH pathology compared with the groups with a single form of pathology. In post hoc analyses, we found no interaction between Aβ and WMH on MTA, and therefore we conclude that WMH and Aβ are independent determinants of MTA severity and that when both are present, their effects are additive.

Increased CSF t-tau was associated with both WMH and Aβ pathology, which is consistent with previous studies where t-tau was considered a measure of neuronal damage [14, 39]. Our finding that t-tau levels were also elevated in patients with WMH and no Aβ pathology contradicts a previous study in which researchers concluded that elevated t-tau levels in patients with vascular damage could be the result of coexisting Aβ pathology [40]. That elevated t-tau levels in the groups with one or both pathologies were found only in subjects with MCI is in line with previous work that strongly related tau to cognitive dysfunction [41]. p-tau was significantly increased only in the Aβ + WMH− group and only slightly increased in the Aβ + WMH+ group, supporting the hypothesis that this could be a specific biomarker for AD [39].

We found similar levels of Aβ for the group with only amyloid pathology and the group with mixed Aβ/WMH pathology. This was in contrast to our expectations based on the literature, because we expected the group with mixed pathology to have a lower amyloid load (i.e., higher Aβ_1–42 levels) [5, 6]. Possibly, the cognitive status of the investigated population (i.e., individuals with SCD or MCI vs. cognitively normal individuals) might play a role, in particular in combination with the method of measuring amyloid load (by CSF or amyloid positron emission tomography), because a study comparing these two methods showed that discordance was dependent on disease stage [42]. Further studies should be done to determine the associations of both factors with amyloid load in mixed AD/CVD patients. The suggestion of expanding the recently proposed “A/T/N” classification system with a vascular component would be valuable in addressing these and other research questions [43].

The decrease in performance in global cognition over time was similar for the three groups with pathology, indicating that WMH and Aβ pathology are drivers of cognitive decline. This is in line with findings derived from a previous study of cognitively normal individuals in which investigators also found that both pathologies contribute to cognitive decline [9]. However, in contrast to this previous study, we did not find any differences in cognitive trajectories between individuals with only Aβ pathology and those with mixed Aβ/WMH pathology. This may relate to the fact that we included subjects with SCD and MCI instead of cognitively normal subjects or to the type of cognitive measures used. Also, risk of progression to dementia during follow-up did not differ between WMH and Aβ, and having both pathologies simultaneously did not increase the risk any further.

This study has several limitations. First, using WMH as a marker of CVD can be seen as a limitation because we did not take other forms of CVD such as lacunar infarcts or cortical microbleeds into account. Although using only WMH reflects a method of defining vascular damage frequently used in clinical practice [44], this makes our findings less generalizable to CVD in general. Also, WMH are heterogeneous in their etiology and pathophysiology, and the underlying mechanisms causing WMH are not yet completely understood [45]. However, in an aging population such as we used in the present study, WMH are mostly considered a consequence of cerebral vascular damage [46, 47]. Second, our follow-up length ranged from 1 to 4 years, which might have been too short to detect differences in cognitive decline in nondemented individuals. Third, our sample was derived from different studies, which might have led to variability in the data, despite adjustment for study in all of the analyses. However, our multistudy design makes our findings more generalizable to other memory clinic settings. Fourth, a methodological consideration of this study was that our results were based on both subjects with SCD and subjects with MCI. Although we did examine the influence of the baseline diagnosis in all analyses and when needed adjusted for this and reported the differences, the smaller sample sizes when analyzing per diagnosis could have influenced the results. The smaller sample sizes in general could reflect a lack of statistical power, and therefore our results should be interpreted with caution and validated in future studies. Although observed power calculations should be interpreted with caution [48], we recommend that researchers in future studies make group sizes more balanced and include a larger number of complete follow-up visits, in particular for outcome measures with smaller effect sizes (e.g., executive functioning measures). The major strengths of this study were the longitudinal setup, the reflection of clinical practice, and the availability of different neurodegeneration markers to provide novel insights into the role of neurodegeneration in relation to AD and CVD.