Longitudinal decreases in multiple cerebrospinal fluid biomarkers of neuronal injury in symptomatic late onset Alzheimer’s disease

2.1. Alzheimer’s Disease Neuroimaging Initiative study design

CSF Aβ42, tTau, and pTau181 demographic, imaging, and cognitive data were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adniloni.usc.edu). The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of MCI and early AD. ADNI participants have been recruited from more than 50 sites across the USA and Canada. Regional ethical committees of all institutions approved of the study, and all participants provided written informed consent. For up-to-date information, see www.adni-info.org.

2.2. Study participants

The ADNI cohort in the present study consisted of all cognitively normal (CN) participants and those with MCI or AD dementia (AD) with available CSF samples from at least two visits as of April 2012. This cohort included 152 individuals across ADNI1, ADNI GO, and ADNI2 (n = 56 CN, n = 73 MCI, and n = 17 AD). Demographic and cognitive data were downloaded in August 2015 and were collected as described (adni.loni.usc.edu/methods/documents/). By definition, individuals in the CN group all had a clinical dementia rating (CDR) score of 0 at the time of lumbar puncture (LP) and a Mini–Mental State Examination (MMSE) score ≥ 24. Individuals with MCI also scored ≥24 on the MMSE but exhibited subjective memory loss (> 1 standard deviation [SD] below the normal mean of the delayed recall of the Wechsler Memory Scale Logical Memory II), received a CDR of 0.5, and preserved activities of daily living and the absence of dementia. The AD group met the definition of probable AD according to the criteria established by the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association [8] and had MMSE scores of 20–26 and CDRs of 0.5 or 1. Groups were designated by clinical diagnosis at the time of initial available CSF sample in the longitudinal cohort (defined herein as baseline).

2.3. ADNI clinical, CSF and imaging data

Scores for MMSE and Alzheimer’s Disease Assessment Scale-cognitive 11 (ADAS11) and ADAS13 were downloaded from the LONI site in August 2015 via ADNIMerge R Package. Values for CSFAβ42 (INNO-BIA AlzBio3;Fujirebio, Ghent, Belgium) were downloaded at the same time from two data sets (UPENNBIOMK4 and UPENNBIOMK6) and were used to define amyloid-positivity based on a published, autopsy-confirmed cutoff value (<192 pg/mL) [9]. For statistical analyses, values for Aβ42, tTau, and pTau181 generated by a single lot number of the novel, fully automated, electrochemiluminescent Elecsys® immunoassays (Roche Diagnostics, Basel, Switzerland) were downloaded from the LONI site in March 2017 from a single data set (UP-ENNBIOMK9). The Elecsys® system aims to offer a fully automated CSF biomarker test for AD capable of achieving In Vitro Diagnostic capability and offers some improvements over current Research Use Only assays including the following: reduction in manual steps, improved precision and accuracy both within labs and between labs, and improved lot-to-lot reagent performance. The Elecsys® Aβ42 immunoassay in use is not a commercially available In Vitro Diagnostic assay. It is an assay currently under development and used only for investigation purposes. The measuring range of the assay is 200 (lower technical limit)—1700 (upper technical limit) pg/mL. The performance of the assay beyond the upper technical limit has not been formally established. Therefore, values above the upper technical limit have been truncated at 1700 pg/mL. In the present study, baseline analyses excluded these data. Longitudinal statistical analyses were run with and without these truncated values and performed nearly identically.

MRI data for the left and right hippocampal (HP) volume (white matter parcellation) and left and right entorhinal cortex (EC) thickness, two regions known to be affected early in AD, were also analyzed. EC thickness and HP volume were downloaded in November 2016 from the file UCSFFSL_02_01_16. Acquisition of 1.5 Tesla MRI and data processing methods are as described (adni.loni.usc.edu/methods/mri-analysis/). Data were processed using FreeSurfer, version 4.4, and only values that passed all quality control standards were included in the analyses. Values for left and right HP and EC thickness were added together to create a value for “total” HP volume and EC thickness. In analyses evaluating potential effects of ventricular volume on CSF biomarker concentrations, we created a variable termed “ total ventricular volume” by summing left [ST37SV] and right [ST96SV] lateral ventricle, left [ST30SV] and right [ST89SV] inferior lateral ventricle, and third ventricle [ST127SFV] from the ADNI data set so to best capture ventricular volume in its entirety.

2.4. Novel CSF analytes

Samples were analyzed for YKL-40 (also known as chitinase 3-like 1, a marker of gliosis/neuroinflammation) [10], VILIP-1 (a neuronal calcium sensor protein and marker of neuronal injury) [11], Ng (a postsynaptic protein and marker of synaptic dysfunction) [12], and SNAP-25 (a presynaptic protein and marker of synaptic dysfunction) [13]. YKL-40 was measured with a plate-based enzyme-linked immunoassay (MicroVue ELISA; Quidel, San Diego, CA) [14]. VILIP-1 [15,16], Ng [17,18], and SNAP-25 were measured using microparticle-based immunoassays using the Singulex (now part of EMD Millipore; Alameda, CA) Erenna system, and employed antibodies developed in the laboratory of Dr. Jack Ladenson at Washington University. All samples (each on the same freeze/thaw cycle) were run in triplicate on a single lot number for VILIP-1, SNAP-25, and Ng and in duplicate for YKL-40. Within-person longitudinal samples were run on the same assay plate to reduce interplate and intraplate variability. Quality control for VILIP-1, SNAP-25, and Ng included analysis of three internal standard CSF pools run on each plate and two internal pools for YKL-40. See Supplementary Text for assay details.

2.5. Statistical analysis

Because the study intent was to compare baseline biomarker levels and their longitudinal change over time in individuals who span the AD continuum (from no disease [normal] to preclinical AD, to MCI due to AD, and to AD), participants in the three diagnostic categories (CN, MCI, and AD) were further stratified into β amyloid-positive (Aβ+) versus β amyloid-negative (Aβ−) at baseline based on the published ADNI CSF Aβ42 cutoff of < 192 pg/mL [9]. Baseline characteristics for the five resultant groups (CN−, CN+, MCI−, MCI+, and AD+) were summarized as mean (SD) for continuous variables or number (percentage) for categorical variables. Group differences among the various measures were assessed using one-way analysis of variance and post hoc Tukey tests. Correlations between measures were assessed via Spearman correlation.

Biomarker concentrations, cognitive performance, and MRI measures within individuals over time were compared among the five groups (all AD individuals were Aβ+) by general linear mixed models with random intercepts/slopes at the subject level to allow estimation and comparison of within-person rates of change [19]. In addition to the mean intercept and slope for each group (unadjusted models), covariates including age at baseline, apolipoprotein E (APOE) ε4 carriage, sex, education, and ventricular volume, their interactions with subject groups on the intercepts and slopes, were also included as fixed effects (see Supplementary Text). All general linear mixed models assumed a subject-level random effect on intercept and slope and were fitted using the maximum likelihood method. Statistical tests were based on the approximate F or t-tests with denominator degrees of freedom approximated by the Satterthwaite methods [13]. All analyses were performed using SAS software, version 9.4 (SAS Institute Inc.), with statistical significance defined as P < .05.

3.1. Demographics

Of the 152 ADNI participants who met the criteria for having longitudinal CSF samples (range 2–7 LPs over 1–7 years of follow-up [mean (SD) = 4.0 (1.62)] and a mean [SD] LP interval of 16 [8.6] months), four were omitted from the data set due to missing values for CSF Aβ42 (via AlzBio3) required to define baseline amyloid status (Aβ+ vs. Aβ−). Participants in the final data set of n = 148 were 38% female, between 58 and 90 years of age at the time of initial LP (mean [SD] = 75 [7.13]), and 68% were APOE ε4-positive (Table 1). All individuals in the MCI group were classified by ADNI as “late MCI”. As expected, baseline HP volume and EC thickness were different among the groups (CN > MCI > AD) (P < .0001). Performances on MMSE, ADAS11, and ADAS13 were also as expected, with the MCI and AD groups performing worse than the CN group (P < .0001).

When the clinical groups were dichotomized into Aβ+ and Aβ− [9], neuronal injury/inflammation biomarker levels were higher (more AD-like) in the Aβ+ than those in the Aβ− groups, both among and within each clinical group (Table 2). Positive correlations were observed among the injury markers at baseline, strongest among tTau, VILIP-1, and Ng (Spearman r = 0.798–0.853) (Supplementary Table 1). SNAP-25 was moderately correlated with the other injury markers (r = 0.619–0.720), and as expected, tTau and pTau exhibited the highest positive correlation (r = 0.975). Elecsys Aβ42 was positively correlated with AlzBio3 Aβ42 (r = 0.869) and negatively correlated with tTau, pTau, and SNAP-25 (r = −0.214, −0.324 and −0.240, respectively). YKL-40 was significantly, but weakly, correlated with the injury markers (r = 0.307–0.422) but not Aβ42.

3.2. Patterns of neuronal injury and neuroinflammatory markers

Participant-level CSF biomarker trajectories were plotted for each of the five amyloid-defined clinical groups (see Supplementary Fig. 1 for spaghetti plots). General linear mixed models (with random intercepts/slopes at the subject level) were then used to estimate and compare baseline biomarker levels and within-person rates of change in the five groups. Results adjusting for sex, APOE ε4 status, education, baseline age, and total ventricular volume are provided in the Supplementary Text.

3.3. Elecsys® tTau

Baseline tTau levels were significantly elevated in the AD+ group compared with all other groups (all P ≥ .01) and the MCI+ compared with the MCI− and CN− (P < .0001) and CN+ groups (P = .02) (Table 2). Longitudinally, tTau levels significantly increased in both CN (both P < .05) and the MCI+ groups (P < .0001) (Fig. 1, Table 2) tTau levels decreased longitudinally in the AD+ group, but this change did not reach statistical significance (P = .095).

3.4. Elecsys® pTau

pTau levels at baseline were significantly elevated in the AD+ compared with all other groups (all P < .01), MCI+ compared with MCI− and CN− (both P < .0001) and CN+ groups (P = .02), and the CN+ compared with the MCI− and CN− groups (both P < .03) (Table 2). Longitudinally, pTau levels significantly increased in the CN+ (P = .001) and trended toward increase in the MCI+ group (P = .055). Strikingly, pTau levels significantly declined in the AD+ group (P ≤ .0001) (Fig. 1, Table 2), with rate of change greater than the change in all other groups (P < .001).

3.5. VILIP-1

Levels of baseline VILIP-1 were significantly higher in the MCI+ and AD+ compared with both the MCI− and CN− groups (all P ≤ .01) (Table 2). The amyloid-positive groups did not differ from one another (all P > .05). Longitudinally, as with pTau, VILIP-1 levels strongly and significantly decreased in the AD+ group (P = .006), whereas no significant changes were observed in the other groups (Fig. 1, Table 2).

3.6. SNAP-25

SNAP-25 values at baseline were significantly higher in the AD+ and MCI+ compared with the CN− (both P < .0003), CN+ (P = .001 and P = .01, respectively), and MCI− groups (both P < .0001) (Table 2). Longitudinally, SNAP-25 levels declined significantly in the AD+ group (P = .05), whereas no significant changes were observed in the other groups (Fig. 1, Table 2).

3.7. Ng

Baseline levels of Ng were significantly higher in the AD+ group than the CN− (P = .003), CN+ (P = .02), and MCI− (P = .0006) groups, although not between the MCI+ and AD+ groups (P = .10) (Table 2). Levels were also higher in the MCI+ compared with the CN− (P = .004) and MCI− (P = .02) groups. Longitudinally, Ng markedly and significantly decreased in the AD+ group (P < .0001), whereas no significant changes were observed in the other groups (Fig. 1, Table 2).

3.8. YKL-40

In contrast to the markers of neuronal injury, baseline levels and longitudinal patterns of change in the neuroinflammatory marker, YKL-40, exhibited a large degree of within-group variability. Baseline YKL-40 was significantly higher in the AD+ compared with the MCI− (P = .04) but not the other groups (Table 2). Longitudinally, all groups showed an increase in mean levels over time, but this increase was statistically significant only in the MCI+ group (P = .03) (Fig. 1, Table 2), perhaps due to less variability (smaller SD) within that group.

3.9. Elecsys® Aβ42

Although CSF Aβ42 (as measured in ADNI by AlzBio3) was used a priori to define amyloid status in the clinical groups, we were also interested in evaluating the patterns of this biomarker using the novel Elecsys® platform. As expected, baseline Aβ42 levels (via Elecsys®) were significantly lower in all Aβ+ than those in Aβ42− groups (all P < .0001) (Table 2). Longitudinally, levels decreased in all groups (and at similar rates), although only the AD+ and CN− groups reached statistical significance (P = .04 and P = .0004, respectively) (Fig. 1, Table 2).

3.10. Cognitive measures

As expected, cognitive performance differed with clinical diagnosis, particularly in the Aβ+ symptomatic groups. Furthermore, Aβ+ individuals exhibited longitudinal changes in MMSE and ADAS11/13 that are consistent with a worsening of cognitive performance and often at a faster rate than the Aβ− groups. See Supplementary Fig. 2 for spaghetti plots.

3.11. MMSE

Baseline MMSE was lower (indicative of worse performance) in the AD+ group than any other group (all P < .0001), lower in the MCI+ compared with the MCI− (P = .03) and both CN groups (both P < .0001), and in the MCI− compared with both CN groups (both P < .03) (Table 3). In the AD+ and MCI+ groups, MMSE was decreasing significantly (both P < .0001) and at a faster rate in the AD+ compared with the MCI + group (P <.0001) (Table 3).

3.12. ADAS11 and ADAS13

At baseline, ADAS11 was significantly elevated (indicating worse performance) in the AD+ compared with both CN groups (both P < .0001), both MCI groups compared with both CN groups (both P < .02), and in the AD+ compared with both MCI groups (both P < .0001) (Table 3). Longitudinally, ADAS11 score significantly increased in the AD+ and MCI+ groups (both P < .0001) and at a significantly faster rate in the AD+ versus the MCI+ group (P < .0001) (Table 3).

Baseline ADAS13 performance was similar to ADAS11 except that the MCI+ group was also significantly elevated (worse performance) compared with the MCI− group (P = .05) (Table 3). Longitudinally, ADAS13 was significantly increasing in all three Aβ+ groups (all P < .004), at a faster rate in the AD+ compared with the MCI+ (P = .0005) and CN+ (P < .0001) groups, and at a faster rate in the MCI+ than the CN+ group (P = .02) (Table 3).

3.13. Volumetric MRI measures

As expected, HP volume and EC thickness were smaller at baseline in the AD+ than those in the other groups. However, all but the CN− group exhibited significant atrophy over time, albeit at different rates. See Supplementary Fig. 3 for spaghetti plots.

3.14. HP volume

HP volume at baseline was significantly smaller in the AD+ compared with all other groups (P < .001 for both CN groups; P = .03 for both MCI groups) and in both MCI groups compared with the CN groups (MCI− vs. CN− [P = .003] and CN+ [P = .01]; MCI+ vs. CN− and CN+ [both P ≤ .0007]) (Table 3). Longitudinally, all groups exhibited significant HP shrinkage over time (all P ≤ .0001) (Table 3). Volume in the AD+ and MCI+ groups decreased at a significantly faster rate than in both CN groups (P ≤ .003 and P ≤ .001, respectively) and the MCI− group (P = .04 and P = .003, respectively). The rate of atrophy in the MCI− group was faster than the CN− group (P = .0009) and in the CN+ compared with the CN− group (P = .03).

3.15. Entorhinal cortex thickness

At baseline, EC thickness was significantly smaller in the AD+ compared with all other groups (P ≤ .0003), in the MCI+ compared with the CN groups (P = .0004 for CN− and P = .01 for CN+) (Table 3). MCI− was also significantly thinner than the CN− group (P = .03) and at the significance level compared with the CN+ group (P = .05). Longitudinally, EC thickness was declining in all but the CN− group (all P ≤ .0003) and at a faster rate in the AD+ compared with the CN+ (P = .005) and MCI− (P = .007) groups (Table 3). The EC in the MCI+ group was also shrinking more quickly than the CN+ and MCI− groups (both P ≤ .0001).