Race modifies the relationship between cognition and Alzheimer’s disease cerebrospinal fluid biomarkers

Ethics, consent, and permission

This study was approved by the Emory Institutional Review Board. Informed consent was obtained from all subjects or their authorized representatives.

Subjects

We enrolled 65 older African Americans and 70 older Caucasians into a cross-sectional study at Emory University from 1 July 2013 to 30 June 2015. Based on our preliminary data, this sample size was sufficient to detect a difference in CSF t-tau levels. Recruitment for this study was previously described [15]. Briefly, potential participants were identified through the Emory Alzheimer’s Disease Research Center, the Emory Cognitive Neurology Clinic, or community forums on aging and memory, and individuals were contacted if there was a subjective report of normal cognition or a diagnosis of MCI or AD dementia. Study protocols, including CSF and MRI analysis, were discussed with potential participants, and those diagnosed with non-AD dementia (e.g., vascular, dementia with Lewy bodies, frontotemporal dementia or primary progressive aphasia, normal pressure hydrocephalus) were excluded from the study. Subjects with MCI with features suggestive of non-AD pathology (e.g., rapid eye movement sleep behavior disorder) were also excluded. A total of 288 subjects were contacted, and 135 (47%) consented to participate. The study completion rate was 98% for MRI (133 of 135) and 93% for lumbar puncture (126 of 135). Subjects who did not undergo lumbar puncture were similar to those who did with regard to age, sex, race, education, and apolipoprotein E (APOE) and ABCA7 genotypes. All biochemical and MRI analyses were performed by experienced operators blinded to the subject’s race and cognitive functioning.

Clinical and neuropsychological characterization

MRI acquisition and analysis

MRI data were acquired using a MAGNETOM Trio, A Tim System 3.0-T scanner (Siemens Healthcare, Erlangen, Germany), including 3D T1-weighted magnetization-prepared rapid acquisition with gradient echo anatomical imaging (repetition time/inversion time/echo time, 2300/900/3.0 milliseconds; flip angle, 9 degrees; voxel size, 1 × 1 × 1.2 mm³; 176 slices). Cortical and subcortical volumes were extracted by employing a user-independent parcellation process in FreeSurfer (version 5.03; Massachusetts General Hospital/Harvard University, Boston, MA, USA) running on the CentOS 6.6 operating system [30]. Subcortical white matter and deep gray matter structures are parcellated by combining data from voxel intensity, probabilistic atlas locations, and local relationships between anatomical structures. Total white matter hyperintensity (WMH) volume was derived from T2-weighted fluid-attenuated inversion recovery sequences as previously described [31]. WMH volume was not normally distributed and was therefore log-transformed for statistical analysis.

CSF collection and analysis

CSF samples were collected between 8:00 a.m. and 2:00 p.m. without requiring fasting. This window was chosen because CSF Aβ42 levels during these times represent approximately 95–105% of average Aβ42 over time [32]. CSF was immediately aliquotted after collection and before freezing, and otherwise we used the ADNI biofluid protocols, including the use of 24-gauge Sprotte needles, aspiration syringes, and transfer into 15-ml polypropylene tubes. To avoid measurement bias related to well positions, sample positions were randomized for each biomarker assay.

CSF biomarkers were measured following the manufacturers’ protocols: Aβ42, t-tau, and p-tau₁₈ levels were measured using the INNO-BIA AlzBio3 immunoassay (Fujirebio, Malvern, PA, USA) in a Luminex 200 platform (Luminex, Austin, TX, USA) [33] (average interplate coefficients of variation of 13% for Aβ42, 10% for t-tau, and 11% for p-tau₁₈₁), Aβ40 levels using the INNOTEST® enzyme-linked immunosorbent assay (ELISA) (Fujirebio), α-synuclein levels using the Novex® ELISA (Thermo Fisher Scientific, Waltham, MA, USA), NfL levels using the NF-light® ELISA (Uman Diagnostics, Umeå, Sweden), and sICAM-1 and sVCAM-1 levels using a commercial multiplex kit (MILLIPLEX® MAP Human Neurodegenerative Magnetic Bead Panel 3 [HNDG3MAG-36 K]; EMD Millipore, Billerica, MA, USA).

DNA extraction and analysis

DNA was extracted from buffy coat as previously described [34]. Single-nucleotide polymorphism (SNP) genotyping for APOE and ABCA7 was performed using TaqMan® assays (Thermo Fisher Scientific). ABCA7 SNP rs3764650 (catalogue number 4351379; Thermo Fisher Scientific) was analyzed because of its association with AD [6] and because it is in linkage disequilibrium with SNP rs115550680 implicated in AD in African Americans [4]. Statistical analysis was restricted to those with successful genotyping (n = 124 cases; all had CSF, 123 had MRI).

Statistical analysis

Statistical analysis was performed using IBM SPSS version 22.0 software (IBM, Armonk, NY, USA). Baseline demographic variables and AD risk alleles were compared between African Americans and Caucasians within each diagnostic category (normal cognition, MCI, and AD) using the chi-square test or Fisher’s exact test for categorical variables and Student’s t test for continuous variables, with p < 0.005 to adjust for multiple comparisons. Among subjects with normal cognition, all biomarkers were normally distributed except for NfL and WMH, which were then log₁₀-transformed. Analysis of covariance (ANCOVA) was used to determine race-associated differences in CSF AD biomarkers (t-tau, p-tau₁₈₁, Aβ40, t-tau/Aβ42, p-tau₁₈₁/Aβ42, Aβ42/Aβ40), α-synuclein, and p-tau₁₈₁-to-t-tau ratio (p/t-tau), adjusting first for cognitive functioning (diagnostic category, Mini Mental State Examination [MMSE], or cognitive Z-score). Race and the interaction between race and cognitive function were both included in these models. When appropriate, age, sex, APOE, ABCA7 risk allele status, and CSF Aβ42 levels were further included in the ANCOVA. F-statistic and p values for main effects were reported, along with coefficient estimate and 95% CI after stepwise removal of factors and covariates with p > 0.10. To account for coexisting cerebrovascular risks, hypertension, diabetes, and WMH volume were then added. Furthermore, sVCAM-1 was analyzed first through Pearson’s correlational analysis and then in its own ANCOVA to determine its suitability as a CSF endothelial marker. Finally, because African Americans had lower CSF tau markers than Caucasians but similar WMH volumes, linear regression analysis was used to model potential interactions between race and tau and between race and WMH volume to determine if cognition was more affected in African Americans than in Caucasians on the basis of the same unit of change in CSF tau markers or WMH volume. In this model, composite cognitive Z-score was the dependent variable, with race, sex, having the APOE ε4 allele, having an ABCA7 risk allele, hypertension, and diabetes as categorical variables and age, education, Aβ42, t-tau, log₁₀ (WMH), and education as continuous variables. Independent variables were entered in a stepwise fashion to achieve a model consisting of only main effects. Race and education were not significant factors in this model, because the cognitive Z-scores had been adjusted for these two factors, but sex and age remained significant to account for variance from the sex- and age-adjusted cognitive Z-scores. To test the hypothesis that race modified the relationship between cognition and t-tau or WMH, an interaction term of t-tau × race or log(WMH) × race was added to the model.

Baseline and genetic characteristics

Among the 135 participants, African Americans were more likely than Caucasians to have the ABCA7 risk allele (46.7% vs. 22.9%, p = 0.005) and the ICAM1 Lys56Met polymorphism (33.3% vs. 5.7%, p < 0.001) (Table 1). African Americans were also more likely to have hypertension (72% vs. 46%, p = 0.003) and diabetes (34% vs. 6%, p < 0.001) (Table 1). The two groups were otherwise similar in age, sex, education, proportion with the APOE ε4 allele, and other vascular risk factors (Table 1).

Relationship between race and CSF biomarkers for amyloid and tau

As a group (regardless of diagnosis), African Americans had lower CSF levels of p-tau₁₈₁ (17.9 vs. 25.6 pg/ml, p < 0.001) (Fig. 1a), t-tau (47.0 vs. 71.5 pg/ml, p = 0.001) (Fig. 1b), and Aβ40 levels (7.88 vs. 9.29 ng/ml, p = 0.017) (Fig. 1d) than Caucasians, but they had similar Aβ42 levels (Fig. 1c). Univariate analysis of these biomarkers revealed the greatest differences in subjects with normal cognition (Table 1), but the relationship between biomarker levels and demographic variables necessitated the use of ANCOVA to determine whether race influenced CSF biomarker levels in a cognition-dependent or cognition-independent fashion. We found that, independent of cognitive functioning (data for cognitive Z-scores presented; diagnostic category and MMSE produced similar results), age, sex, APOE ε4 and ABCA7 risk alleles, and Aβ42 levels, African Americans had lower levels of t-tau [difference of 23.6 pg/ml; 95% CI, 9.5–37.7; F(2,122) = 10.99; p = 0.001], p-tau₁₈₁ levels [difference of 7.4 pg/ml; 95% CI, 3.7–11.2 pg/ml; F(2,122) = 15.79; p < 0.001], and Aβ40 [difference of 1.355 ng/ml; 95% CI, 0.293–2.417 ng/ml; F(2,122) = 6.385; p = 0.013]. Race also affected the ratio biomarker of t-tau/Aβ42 (Fig. 1e) and p-tau₁₈₁/Aβ42 (not shown) according to cognitive functioning, but it had minimal effect on Aβ42/Aβ40 (Fig. 1f). For both tau markers, cognitively impaired African Americans had lower CSF t-tau/Aβ42 [difference of 0.255 per 1-SD change in composite cognition; 95% CI, 0.100–0.409; F(2,122) = 10.67; p = 0.001] and p-tau₁₈₁/Aβ42 [difference of 0.076 per 1-SD change in composite cognition; 95% CI, 0.031–0.122; F(2,122) = 10.94; p = 0.001] ratios than cognitively impaired Caucasians.

CSF neurodegenerative and endothelial markers do not account for race-associated AD biomarker differences

Lower t-tau and p-tau₁₈₁ levels in African Americans may be explained by misdiagnosis, less neurodegeneration, or greater contribution from non-AD copathology toward cognitive impairment. Because CSF Aβ42 levels were indistinguishable between African Americans and Caucasians, cognitively impaired subjects in the two racial groups were equally likely to have β-amyloidopathy, which makes misdiagnosis less probable. To determine if there was a difference in neurodegeneration between the two races, we examined the relationship between race, CSF NfL levels, and hippocampal atrophy because the latter two are influenced by AD as well as non-AD disorders. ANCOVA showed that cognitively normal African Americans had lower log(NfL) than cognitively normal Caucasians [F(2,122) = 3.525; p = 0.017], but log(NfL) did not differ between the two races for those with cognitive impairment. Race also had no effect on hippocampal volumes [F(2, 122) = 1.78; p = 0.185] to suggest less neurodegeneration in cognitively impaired African Americans. Similarly, among those with CSF t-tau/Aβ42 not consistent with AD, race had no effect on CSF α-synuclein levels [implicated in Lewy body disease; F(2,75) = 1.078; p = 0.303] or p/t-tau ratio [implicated in frontotemporal lobar degeneration with TDP-43 immunoreactive inclusions; F(2,75) = 0.775; p = 0.382].

Because African Americans had a higher prevalence of hypertension and diabetes than Caucasians in our cohort (Table 1), cerebrovascular disease could account for cognitive impairment in the setting of lower CSF tau levels. Including hypertension, diabetes, and total WMH volume in the ANCOVA did not change the effect of race on p-tau₁₈₁ [F(2,121) = 18.13; p < 0.001), t-tau [F(2,121) = 6.790; p = 0.010], or Aβ40 levels [F(2,121) = 5.084; p = 0.008]. Therefore, vascular risk factors and total WMH volume do not sufficiently account for the CSF AD biomarker differences between the two races. At the same time, peripheral vascular risk factors may not adequately reflect cerebrovascular disease burden. We thus tested whether CSF levels of two candidate endothelial markers (sVCAM-1 and sICAM-1) could be used as surrogate markers of endothelial dysfunction. As a proof of principle, we first analyzed the relationship between these two markers and race, peripheral vascular risks, and WMH among cognitively normal subjects enriched for those without AD pathology (CSF Aβ42, > 192 ng/ml; n = 34). At the univariate level, CSF sVCAM-1 levels, but not sICAM-1 levels, correlated positively with older age (R = 0.514; p < 0.001), Caucasian race (R = 0.35; p = 0.026), higher WMH volume (log-transformed; R = 0.433; p = 0.005), greater number of vascular risk factors (R = 0.357; p = 0.020), history of congestive heart failure (R = 0.444; p = 0.003), and history of atrial fibrillation (R = 0.399; p = 0.009). ANCOVA showed that race modified the relationship between sVCAM-1 levels and WMH volume (Fig. 2a) and between sVCAM-1 levels and vascular risk (Fig. 2b). Similar results were obtained when all subjects with CSF t-tau/Ab42 < 0.39 were included (Fig. 2c, d). In sum, whereas sVCAM-1 correlated with known vascular disease markers in Caucasians, it is a poor measure of endothelial dysfunction in African Americans.

Race modifies the impact of WMH on cognition