Research ArticleAlzheimer’s Disease

Longitudinal Change in CSF Biomarkers in Autosomal-Dominant Alzheimer’s Disease

See allHide authors and affiliations

Science Translational Medicine  05 Mar 2014:
Vol. 6, Issue 226, pp. 226ra30
DOI: 10.1126/scitranslmed.3007901

Abstract

Clinicopathological evidence suggests that the pathology of Alzheimer’s disease (AD) begins many years before the appearance of cognitive symptoms. Biomarkers are required to identify affected individuals during this asymptomatic (“preclinical”) stage to permit intervention with potential disease-modifying therapies designed to preserve normal brain function. Studies of families with autosomal-dominant AD (ADAD) mutations provide a unique and powerful means to investigate AD biomarker changes during the asymptomatic period. In this biomarker study, we collected cerebrospinal fluid (CSF), plasma, and in vivo amyloid imaging cross-sectional data at baseline in individuals from ADAD families enrolled in the Dominantly Inherited Alzheimer Network. Our study revealed reduced concentrations of CSF amyloid-β1–42 (Aβ1–42) associated with the presence of Aβ plaques, and elevated concentrations of CSF tau, ptau181 (phosphorylated tau181), and VILIP-1 (visinin-like protein-1), markers of neurofibrillary tangles and neuronal injury/death, in asymptomatic mutation carriers 10 to 20 years before their estimated age at symptom onset (EAO) and before the detection of cognitive deficits. When compared longitudinally, however, the concentrations of CSF biomarkers of neuronal injury/death within individuals decreased after their EAO, suggesting a slowing of acute neurodegenerative processes with symptomatic disease progression. These results emphasize the importance of longitudinal, within-person assessment when modeling biomarker trajectories across the course of the disease. If corroborated, this pattern may influence the definition of a positive neurodegenerative biomarker outcome in clinical trials.

INTRODUCTION

Alzheimer’s disease (AD), the most common cause of dementia in the elderly, is a progressive and fatal neurodegenerative disorder. AD currently affects ~10.6 million people in the United States and Europe, with numbers expected to double every 20 years unless effective disease-modifying treatments are developed (http://www.alz.org/national/documents/Facts_Figures_2011.pdf). Suboptimal clinical diagnostic accuracy (1) and the existence of a long preclinical phase during which the hallmark pathologies (amyloid plaques, neurofibrillary tangles, and neuronal injury/death) develop before the appearance of clinical symptoms (27) have propelled efforts to identify biomarkers to aid disease diagnosis and prognosis, especially during the preclinical and early clinical stages.

Results from scores of biomarker studies have contributed to hypothesized trajectories of fluid and imaging biomarker changes that take place over the natural course of the disease, from the asymptomatic/preclinical stage to the earliest symptomatic stage (variously termed “mild cognitive impairment,” “prodromal AD,” and “very mild AD”) to the end stages characterized by advanced dementia (8, 9). Substantiating the longitudinal change in biomarkers over time will advance our basic understanding of the pathobiology of the disease and also provide information critical for the design and interpretation of disease-modifying clinical trials that use biomarkers for subject enrollment or as outcome measures.

In studies of older adults at risk for late-onset AD (LOAD), comparison among independent cohorts with different clinical characteristics has been used as a proxy for evaluating longitudinal change within individuals over time. For example, cerebrospinal fluid (CSF) concentrations of amyloid-β1–42 (Aβ1–42), the primary component of amyloid plaques, inversely correlate with plaque load (1016) and are lower at the group level in individuals with clinically expressed AD compared to those with normal cognition. Conversely, CSF concentrations of the microtubule-associated protein tau and/or hyperphosphorylated forms of tau (ptau), the primary constituents of neurofibrillary tangles, positively correlate with brain atrophy measures (17, 18) and tangle load at autopsy (14, 19, 20), although one study failed to find such associations (21). Mean concentrations of CSF tau are higher in individuals with clinically expressed AD compared to those with normal cognition, consistent with these analytes being considered markers of neuronal pathologies (tangles, neuronal injury, and/or death). However, such a categorical grouping according to cognitive status is only a gross estimate of disease stage during a dynamic neuropathological process and, furthermore, does not necessarily accurately capture the pathological changes in the earliest preclinical and clinical stages.

Study of autosomal-dominant AD (ADAD) is particularly well suited to investigations of biomarker trajectories because the 100% penetrance of the mutations and the relative consistency of age at symptom onset (AAO) within families overcome the limitations of disease uncertainty and unpredictability of symptom onset inherent in studies of LOAD. In our initial cross-sectional study of ADAD individuals enrolled in the Dominantly Inherited Alzheimer Network (DIAN), we analyzed baseline measures of plasma Aβ1–42 and CSF Aβ1–42 and tau in 79 mutation carriers (MCs) and 34 mutation noncarriers (NCs) as a function of their estimated number of years to symptom onset (EYO) (22). Statistical modeling of these cross-sectional measures in the two genetic groups suggested the presence of amyloid and neuronal pathologies at least 10 to 15 years before symptoms, with biomarker abnormalities progressing in severity over time. However, true longitudinal evaluation within individuals with disease progression was not possible in that initial cross-sectional cohort.

To more fully characterize the patterns of fluid marker evidence of amyloid and neuronal pathologies and to test the hypothesis that the degree of biomarker abnormality increases over time with disease progression, the current follow-up study evaluated cross-sectional baseline data for four biomarkers in plasma and five analytes in CSF from a much larger cohort of DIAN participants (146 MCs and 96 NCs) spanning a wide range of EYOs. We also analyzed longitudinal CSF samples collected from a subset of individuals (n = 37). In asymptomatic MCs, we observed patterns of longitudinal change in CSF biomarkers within individuals that were similar to those inferred from cross-sectional analyses based on dementia severity and from modeling cross-sectional baseline measures as a function of EYO [that is, reductions in CSF Aβ1–42 and elevations in CSF tau, ptau181, and a new marker of neuronal injury/degeneration, visinin-like protein-1 (VILIP-1)]. However, analysis of longitudinal samples revealed decreases, not increases, in the concentrations of markers of neuronal injury/death within individuals over time once they are older than their expected age of symptom onset. These findings have important implications for understanding the dynamics of disease pathobiology and interpreting neuronal injury biomarker concentrations in response to AD therapies.

RESULTS

Baseline demographics and cross-sectional biomarker data are presented in Table 1. Two-hundred forty-two participants had at least one plasma or CSF sample evaluated (n = 237 with plasma, n = 206 with CSF). For analysis purposes, participants were grouped by genetic (MCs and NCs) and cognitive (asymptomatic and symptomatic) status. The cognitively normal [defined by a Clinical Dementia Rating (CDR) score of 0] MC group is termed MC-AS (asymptomatic), and MC individuals with CDR >0 are termed MC-S (symptomatic). CDR scores in the MC-S group ranged from 0.5 (very mild) to 3 (severe), although the majority (65%) was CDR 0.5. The CDR Sum of Boxes (CDR-SB) in the symptomatic group ranged from 0.5 to 17 (with 18 indicative of worst performance), and the Mini-Mental State Exam (MMSE) (23) scores ranged from 3 to 30 (with 30 indicative of perfect performance). As a group, MC-AS individuals were younger than NCs (P < 0.05), who were, in turn, younger than MC-Ss (P < 0.05), although the mean parental AAO was not different among the groups (45 to 48 years) (P > 0.05). The EYO for the entire cohort ranged from about −30 (30 years before the parental AAO) to +30 (30 years after the parental AAO). Nearly 60% of the participants were female, and ~25% of individuals carried at least one APOE ε4 allele. Five participants were ε4 homozygotes. Most participants (70%) were from families with mutations in PSEN1.

Table 1. Characteristics of the cross-sectional DIAN cohort at baseline.

Demographic variables correspond to mean (SD) or n (%) as indicated. Biomarker variables correspond to mean (SD). Continuous demographic variables were compared with general linear mixed models, gender and APOE4 status were compared with mixed-effects logistic regression models, and family mutations were compared with a mixed-effects multinomial logistic regression model. Pairwise testing was conducted only after an omnibus test suggested significant (P < 0.05) differences between groups. Units for single biomarker analytes are pg/ml. Analyte combinations are represented as ratios. AAO, age at symptom onset; Aβ, amyloid-β; APOE ε4+, presence of at least one ε4 allele of apolipoprotein E; APP, amyloid precursor protein; CDR, Clinical Dementia Rating score (0, cognitively normal; 0.5, very mild; 1, mild; 2, moderate; 3, severe); CDR-SB, CDR Sum of Boxes (range, 0 to 18, with 0 indicating no impairment); EYO, estimated years to symptom onset; INNO, INNOTEST ELISA; MC-AS, asymptomatic mutation carrier (CDR 0); MC-S, symptomatic mutation carrier (CDR >0); MMSE, Mini-Mental State Exam score (range, 0 to 30, with 30 as perfect score); NC, mutation noncarrier; PSEN1, presenilin 1; PSEN2, presenilin 2; ptau, phosphorylated tau; VILIP-1, visinin-like protein 1.

View this table:

Trajectories of plasma analytes estimated from baseline cross-sectional data

In contrast to LOAD in which disease certainty is unknown and age of symptom onset is unpredictable, asymptomatic individuals with deterministic mutations will develop dementia and at an age that is relatively predictable within a given family. To take advantage of this characteristic of ADAD, a metric defined as the EYO for each DIAN participant was calculated as the participant’s age at clinical evaluation minus the reported parental AAO, thus providing an estimate of where along the disease trajectory a specific individual falls at a given point in time, regardless of their chronological age. This allows for comparison among individuals from different families with different parental AAOs. General linear mixed models, accounting for within-family dependencies, were used to estimate the trajectory of biomarker concentrations in the two genetic groups to derive predicted mean biomarker concentrations at defined EYOs (−25 to +10). Estimated mean concentrations of plasma Aβ1–40 in MCs did not differ significantly from those of NCs at any EYO point (P = 0.8720) (Fig. 1A; see table S1 for EYO-specific statistical differences). In contrast, whereas plasma Aβ1–42 concentrations in the two groups were stable across EYO points (trajectories not different from 0; NC, P = 0.2903; MC, P = 0.7710), overall concentrations in MCs were significantly higher at all but the earliest (EYO −25) EYO point evaluated (all P < 0.0343) (Fig. 1B and table S1). However, small participant numbers at this early EYO warrant statistical caution. Similar results were obtained for the ratio of plasma Aβ1–42/Aβ1–40 (Fig. 1C and table S1). The estimated trajectories and the group differences for plasma Aβx–40, Aβx–42, and the Aβx–42/Aβx–40 ratio were virtually identical to those of the full-length species (Aβ1–40 and Aβ1–42) (table S1). Plasma concentrations of VILIP-1, a new marker of neuronal injury/death, did not differ significantly between the two groups (P = 0.0881), nor consistently as a function of EYO (Fig. 1D and table S1). However, at every time point examined, mean concentrations of plasma VILIP-1 were higher in MCs compared to NCs (table S1). For all analytes, statistical comparisons could not be made at the far extremes of the EYO distribution (earlier than EYO −25 and later than EYO +10) because of the small number of individuals at these points.

Fig. 1. Estimated mean plasma biomarker concentrations at defined EYO.

Predicted mean plasma biomarker concentrations in NCs and MCs at various EYO were determined by general linear mixed models, accounting for within-family dependencies, and plotted as histograms. (A to D) Estimated trajectories of plasma (A) Aβ1–40, (B) Aβ1–42, (C) Aβ1–42/Aβ1–40 ratio, and (D) VILIP-1 for NCs and MCs. Values represent means ± SE. *P < 0.05, **P < 0.01, P < 0.001, ††P < 0.0001, MC significantly different from NC at a given EYO.

Trajectories of CSF amyloid-related analytes estimated from baseline cross-sectional data

The changes in the various CSF biomarkers estimated from cross-sectional baseline measures were more complex than those for plasma (Fig. 2). CSF Aβ1–40 concentrations did not consistently differ as a function of EYO between the NC and MC groups. Although the overall approximate F test (P = 0.0021) suggested that concentrations were not equal between the groups over the EYO range, the approximate t tests at 5-year intervals did not reveal persistent differences across the entire EYO range (Fig. 2A and table S2). In contrast, concentrations of CSF Aβ1–42 in MCs were significantly lower than those in the NC group at least 10 years before their parental AAO (EYO −10), whereas concentrations were stable over the range of EYOs in NCs (linear trajectory = −0.1534, P = 0.9088) (Fig. 2B and table S2). In agreement with results from our initial report evaluating a smaller subset of this DIAN cohort (22), the Aβ1–42 concentrations appeared to diverge from one another even earlier (earlier than EYO −10), with concentrations in the MC group higher than those in NCs at very early EYO points. However, there were too few individuals at these early EYO points to make any statistical conclusions. Similar to what was observed for CSF Aβ1–42, the ratios of CSF Aβ1–42/Aβ1–40 in MCs were lower than those of NCs in individuals who were approaching or surpassing their estimated age at symptom onset (EAO) (EYO 0) (table S2). Consistent with previous studies (2426), absolute concentrations of CSF Aβ1–42 obtained by the two assay platforms (INNOTEST versus AlzBio3) were different but highly correlated (Pearson r = 0.8559, P < 0.0001).

Fig. 2. Estimated mean CSF biomarker concentrations at defined EYO.

Predicted mean CSF biomarker concentrations in NCs and carriers at various EYO were determined by general linear mixed models, accounting for within-family dependencies, and plotted as histograms. (A to H) Estimated trajectories of CSF (A) Aβ1–40, (B) Aβ1–42, (C) tau, (D) ptau181, (E) tau/Aβ1–42 ratio, (F) ptau181/Aβ1–42 ratio, (G) VILIP-1, and (H) VILIP-1/Aβ1–42 ratio for NCs and MCs. Values for Aβ1–42 were obtained with AlzBio3. Values represent means ± SE. *P < 0.05, **P < 0.01, P < 0.001, ††P < 0.0001, MC significantly different from NC at a given EYO.

Relationship between fluid Aβ analytes and in vivo amyloid imaging

To better understand the potential etiology of the alterations in fluid Aβ measures in MCs, concentrations of CSF and plasma Aβ1–40 and Aβ1–42 were compared to cortical retention of the Aβ-binding agent Pittsburgh compound B (PIB), a marker of fibrillar Aβ deposition, as assessed by positron emission tomography (PET). This cohort consisted of DIAN participants who had undergone both baseline fluid collection and PET PIB procedures (82 NC, 61 MC-AS, and 35 MC-S with CSF and PIB; 94 NC, 68 MC-AS, and 43 MC-S with plasma and PIB) within a short time interval [mean (SD), 22.7 (36.9) days]. A preliminary analysis estimated PIB positivity in this ADAD cohort to be a standardized uptake value ratio (SUVR) ≥0.85 when normalized to brainstem (mean cortical SUVR) using a k-means clustering algorithm (see Materials and Methods). As shown in Fig. 3 (A, C, E, and G), all NCs were PIB-negative as expected given their young age [mean (SD), 39.6 (9.8) years]. Concentrations of plasma and CSF Aβ species differed up to about sixfold among these normal individuals. In contrast, 58% of MCs were considered to be PIB-positive (PIB+) (53 of 93), including 40% (23 of 59) of those who were asymptomatic and 88% (30 of 34) of those who were symptomatic. PIB+ individuals generally had low concentrations of CSF Aβ1–42 (Fig. 3D), whereas concentrations of CSF Aβ1–40 (Fig. 3B) and plasma Aβ species (Fig. 3, F and H) were not associated with cortical PIB retention. Similar to what is observed in LOAD, the distribution of biomarker values in the MC group included both normal (high CSF Aβ1–42, low PIB) and AD-like (low CSF Aβ1–42, high PIB) patterns. The normal pattern was observed more often in individuals who were farthest from their EAO (that is, earlier in the disease course, dark/light blue symbols), whereas the AD pattern was more common in individuals closer to their parental AAO (that is, later in the course of the disease, pink/red symbols). However, in individuals who were considered PIB+, concentrations of CSF Aβ1–42 were only weakly negatively correlated with amyloid load (Pearson correlation = −0.3008, P = 0.0120; Pearson correlation = −0.2569, P = 0.0470 when the high outlier is omitted).

Fig. 3. Association between fluid Aβ measures and mean cortical PIB retention in NCs and MCs.

(A to H) Concentrations of CSF (A and B) Aβ1–40 and (C and D) Aβ1–42, and concentrations of plasma (E and F) Aβ1–40 and (G and H) Aβ1–42 for NCs (left panels) and MCs (right panels). Units on the y axes are pg/ml. Units on the x axes are mean cortical PIB SUVR calculated from prefrontal cortex, gyrus rectus, lateral temporal, and precuneus regions using a brainstem (pons) gray matter reference following application of partial volume correction. Cortical PIB positivity in this ADAD cohort is defined as SUVR ≥0.85 (vertical dashed line) based on a k-means clustering algorithm implemented in R (see Materials and Methods). Symbol colors identify groupings of the EYO in the participant groups, with blue-to-red gradations extending from dark blue (EYO earlier than −20 years) to dark red (EYO later than +11 years).

Trajectories of CSF neuronal injury–related analytes estimated from baseline cross-sectional data

The observed changes in CSF tau and ptau181 estimated from cross-sectional baseline measures were similar to each other. Consistent with our earlier report in a much smaller cohort (22), concentrations of tau in MCs were significantly higher than those of NCs as early as 15 years before their EAO (EYO −15), with concentrations even higher in individuals who were approaching or who had surpassed their EAO (linear trajectory = 3.7394, P < 0.0001) (Fig. 2C and table S2). Similar patterns were observed for ptau181 (Fig. 2D), although concentrations were higher in MCs than in NCs even earlier (EYO −20) (table S2). The changes in the ratios of tau/Aβ1–42 (Fig. 2E) and ptau181/Aβ1–42 (Fig. 2F and table S2) were similar to those for tau and ptau181. In contrast to the pattern of VILIP-1 in plasma, but similar to those of CSF tau and ptau181, concentrations of CSF VILIP-1 (Fig. 2G) in MCs were significantly higher than those in NCs as early as EYO −15 (table S2), with concentrations even higher in individuals who were approaching or had surpassed their EAO (EYO 0) (linear trajectory = 2.3050, P = 0.0004). In addition, the estimated trajectory of the ratio of VILIP-1/Aβ1–42 (Fig. 2H) and group differences as a function of EYO were similar to those observed for the tau(s)/Aβ1–42 ratios (table S2). Concentrations of CSF VILIP-1 were strongly correlated with both tau (Spearman correlation = 0.7610, P < 0.0001) and ptau181 (Pearson correlation = 0.7320, P < 0.0001), and less so with Aβ40 (Pearson correlation = 0.2429, P = 0.0001) and Aβ42 (Pearson correlation = −0.1694, P < 0.0001). Trajectories of cognitive measures estimated from cross-sectional data revealed differences between mutation groups later than differences in biomarker measures (CDR-SB and MMSE at EYO −5, compared to CSF biomarkers at EYOs −10 and earlier) (table S2).

Longitudinal change in CSF biomarkers within individuals over time

The biomarker trajectories estimated from the cross-sectional baseline data from MCs (Fig. 2) over a range of EYOs support a dynamic model of AD neuropathologic changes characterized by early reductions in CSF Aβ1–42 (marker of amyloid) and continued elevations in tau, ptau181, and VILIP-1 (markers of neurofibrillary tangles and/or neuronal injury/death), a pattern that suggests an increase in biomarker abnormality with disease progression. This pattern is consistent with what has been proposed from cross-sectional analyses in LOAD in which mean biomarker concentrations are compared as a function of clinical diagnosis and/or staging (for example, AD versus controls, CDR 0 versus CDR >0, decreasing MMSE) (27, 28). To investigate this pattern in an ADAD cohort, we analyzed the baseline cross-sectional DIAN data as a function of CDR as opposed to EYO. Similar to what has been reported in LOAD, mean concentrations of CSF Aβ1–42 in MCs appear to decrease with advancing dementia severity (especially at early stages) (Fig. 4A), whereas concentrations of the neurodegenerative markers tau, ptau181, and VILIP-1 appear to increase as dementia progresses (Fig. 4, B to D). However, given the fairly wide distribution and overlap between individual CSF analyte values among the groups, we sought to better understand the nature of true longitudinal change within individuals by analyzing two to three samples obtained from the same person over time [lumbar puncture (LP) interval ranged from 10 to 38 months; mean (SD), 16.7 (8.9)].

Fig. 4. Mean baseline concentrations of CSF biomarkers in NCs and MCs according to dementia severity.

(A to D) Box and whisker plots (median ± 25th/75th percentiles) of CSF (A) Aβ1–42, (B) tau, (C) ptau181, and (D) VILIP-1 obtained at baseline are shown for NCs (white box, as the normal reference group) and MCs (purple boxes) with the indicated CDR scores (0, cognitively normal; 0.5, very mild; 1, mild; 2, moderate; 3, severe). The number of participants in each group is shown in italics in parentheses. Units on the y axes are pg/ml. Horizontal lines indicate statistical significance between the groups as determined by general linear mixed models, accounting for within-family dependencies: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; different from NC: +P < 0.05, ++P < 0.01, +++P < 0.001, ++++P < 0.0001.

The longitudinal cohort (n = 37) was composed of 11 NCs and 26 MCs (9 MC-AS and 17 MC-S at baseline), with demographics similar to the larger cross-sectional cohort [with the exception of a higher than expected percentage of APOE4+ individuals in the small MC-AS group (7 of 9, 78%)] (Table 2). Because of current limitations regarding biomarker assay precision and reproducibility (2931), longitudinal samples from a given individual were analyzed together on the same assay plate, and all plates were from a single lot number to eliminate between-plate and between-batch variabilities as potential confounders. As expected, the mean baseline concentration of CSF Aβ1–42 in MCs in this longitudinal cohort was lower than that observed in NCs (P < 0.0001), whereas baseline concentrations of tau, ptau181, and VILIP-1 were higher (P < 0.0001). Concentrations of VILIP-1 in MC-Ss were significantly higher than those in NCs (P < 0.0050). When comparing within-person biomarker concentrations, these relatively young NCs [mean age (SD), 40.0 (8.3) years] exhibited little change in their high concentrations of CSF Aβ1–42 and low concentrations of tau, ptau181, and VILIP-1 over the 1- to 3-year time period, although changes in Aβ1–42 concentrations were more variable (Fig. 5A). In general, concentrations of Aβ1–42 in MCs appeared to decrease, albeit to differing degrees, across the range of baseline EYOs (Fig. 5A, top), consistent with what we observed in cross-sectional analyses (Fig. 2B). However, we observed an interesting finding with respect to longitudinal changes in markers of neuronal pathology and injury in MCs. In contrast to a pattern of steady increase in tau, ptau181, and VILIP-1 with disease progression that was inferred from cross-sectional analyses (Fig. 2, C, D, and G) or CDR (Fig. 4, B to D), the within-person longitudinal changes of these markers differed in direction depending on where a person fell with respect to their baseline EYO. Specifically, concentrations of tau, ptau181, and VILIP-1 increased longitudinally in those at early stages of the disease (EYO ≤ 0) but decreased in those at later stages (EYO > 0) (Fig. 5, B to D, top).

Table 2. Baseline characteristics of the longitudinal DIAN subcohort.

Demographic variables correspond to mean (SD) or n (%) as indicated. Biomarker variables correspond to mean (SD). Continuous demographic variables were compared with general linear mixed models, whereas gender and APOE4 status were compared with mixed-effects logistic regression models. Pairwise testing was conducted only after an omnibus test suggested significant (P < 0.05) differences between groups. Units for biomarker analytes are pg/ml. Aβ, amyloid-β; APOE ε4+, presence of at least one ε4 allele of apolipoprotein E; CDR, Clinical Dementia Rating score (0, cognitively normal; 0.5, very mild; 1, mild); CDR-SB, CDR Sum of Boxes (range, 0 to 18, with 0 indicating no impairment); EYO, estimated years to symptom onset; LP, lumbar puncture; MC, mutation carrier; MC-AS, asymptomatic mutation carrier (CDR 0); MC-S, symptomatic mutation carrier (CDR >0); MMSE, Mini-Mental State Exam score (range, 0 to 30, with 30 as perfect score); NC, mutation noncarrier; ptau, phosphorylated tau; VILIP-1, visinin-like protein 1.

View this table:
Fig. 5. Evaluation of CSF biomarker change over time in NCs and MCs in the longitudinal cohort.

(A to D) Concentrations of CSF (A) Aβ1–42, (B) tau, (C) ptau181, and (D) VILIP-1 as a function of the EYO at the time of sample collection. To maintain participant and investigator blinding to mutation status when reporting individual data points, specific EYOs are not shown. Top: Spaghetti plots show individual data points. Bottom: Histograms show the associated estimated mean (SE) annual within-individual rate of change in biomarker concentrations in the two genetic groups in the period before their EAO (EYO ≤ 0) or the period after their EAO (EYO > 0). General linear mixed models were used that allowed for different slopes in the genetic groups, as well as based on where participants fell with respect to their EYO at baseline (EYO ≤ 0 versus EYO > 0). To account for the within-individual and within-family correlations, the models incorporated random intercepts/slopes at the individual level and a random intercept at the family level. Blue, NC; red, MC. Italicized numbers above the mean group slopes correspond to the significant or near-significant P values comparing the slopes to zero. Italicized numbers above the horizontal lines correspond to the significant P values comparing the slopes between the NC and MC groups within the indicated EYO range (EYO ≤ 0 versus EYO > 0).

To quantify and statistically evaluate these observations, we fitted linear random coefficients models (32) that permitted different longitudinal rates of change (that is, slopes) as a function of whether the participants’ EYO was ≤0 or >0 at baseline, thereby facilitating comparisons of the longitudinal rate of change in biomarker concentrations across the groups. Additional models including APOE genotype (E4+ versus E4) and gender were also implemented to examine the effects of these covariates. Statistically, NCs exhibited no significant longitudinal change in CSF Aβ1–42 concentrations before their EAO (EYO ≤ 0) [−3.12 (9.33) pg/ml per year, P = 0.7406] but a small (4.8%), albeit significant, decrease after their EAO (EYO > 0) [−24.96 (11.89) pg/ml per year, P = 0.0432] (Fig. 5A, bottom, and table S3A). MCs exhibited a trend toward decreases in CSF Aβ1–42, both before [−16.14 (10.04)] and after [−17.86 (12.45)] EYO 0, although these trajectories did not reach statistical significance (P = 0.1170 and P = 0.1607, respectively) (Fig. 5A, bottom, and table S3A).

As expected, concentrations of tau in NCs did not change significantly before [−3.00 (3.73) pg/ml per year, P = 0.4251] or after [−0.61 (4.70) pg/ml per year, P = 0.8970] their EAO (Fig. 5B, bottom, and table S3A). In contrast, MCs displayed a trend toward increasing tau concentrations [+6.91 (3.72), P = 0.0699] before their EAO, but a significant decrease in tau concentrations later in the disease course (EYO > 0) [−10.78 (4.70), P = 0.0270] (Fig. 5B, bottom, and table S3A). Similar patterns of longitudinal decreases at later stages were observed for the other markers of neuronal pathologies (ptau181 and VILIP-1) in MCs but not in NCs. Concentrations of ptau181 decreased significantly by an estimated −9.52 (3.37) pg/ml per year (P = 0.0088), and VILIP-1 by an estimated −14.61 (3.83) pg/ml per year (P < 0.0005), in MCs with baseline EYO >0 (Fig. 5, B to D, bottom, and table S3A). However, the changes before their EAO were not statistically significant [ptau181, +0.588 (2.81), P = 0.8356; VILIP-1, −0.885 (3.09) P = 0.7766] (table S3A). These markers did not change longitudinally in NCs at either time (EYO ≤ 0 or EYO > 0; all P > 0.4231) (table S3A). Virtually identical results were obtained when models were adjusted for gender and APOE genotype (table S3B). Although the magnitude of change in neuronal injury markers between longitudinal samples was relatively small [ranges (pg/ml): tau, −52 to +49; ptau181, −41 to +21; VILIP-1, −54 to +21], the coefficients of variation (% CVs) between longitudinal samples from a given individual were significantly greater than those within each sample analyzed in duplicate (tau, P = 0.0004; ptau181, P < 0.0001; VILIP-1, P = 0.0069; fig. S1). In contrast, there was no significant difference between the within-sample and between-sample % CVs for Aβ1–42 (P = 0.3241) (fig. S1). These data provide support for the idea that the longitudinal changes we observed in tau, ptau181, and VILIP-1 are not due to assay imprecision.

To put these longitudinal biomarker changes into the context of cognitive performance in this subcohort, we evaluated a composite of three psychometric test measures that were recently reported to be sensitive to cognitive decline in asymptomatic MCs (CDR 0) in DIAN (33). These tests include Logical Memory IA—Immediate Recall and Logical Memory IIA—Delayed Recall, both tests of episodic memory (34), and Digit Symbol, a measure of speeded visual spatial processing (35). Cognitive performance in NCs across the range of sampled EYOs did not change significantly before [z = 0.0259 (0.0608), P = 0.6721] or after [z = 0.0636 (0.0778), P = 0.4179] their EAO (fig. S2). In contrast, MCs who were after their EAO (EYO > 0) exhibited lower mean baseline scores (worse performance) than those who were before their EAO (EYO ≤ 0) [EYO > 0, z = −0.9305 (0.2247); EYO ≤ 0, z = 0.2018 (0.1902), P = 0.0004]. However, performance was quite variable in MCs who were close to their EAO (EYO ~−5 to 0), with some performing within the range of NCs and others already impaired to the level of those past their EAO (fig. S2). Furthermore, whereas the mean estimated change in this composite was not significantly different from zero in MCs whose baseline assessment was before their expected age of symptom onset (EYO ≤ 0) [z = 0.0144 (0.0601), P = 0.8119], performance decreased in MCs whose baseline assessment was after their expected age of symptom onset (EYO > 0) [z = −0.2815 (0.0647), P < 0.0001]. Virtually identical results were obtained when analyses were adjusted for gender and APOE4 status (table S3B).

DISCUSSION

Our understanding of AD has evolved greatly over the past two decades with the advent of fluid and imaging biomarkers that permit the in vivo detection of underlying disease pathologies. Data from clinicopathological and biomarker studies have converged to support the existence of a long preclinical stage during which AD pathologies develop before the appearance of cognitive symptoms. As a consequence, individuals in this preclinical stage are receiving intense focus as a targeted population for AD secondary prevention trials. In such trials, biomarkers are being used for subject enrollment, for proof of therapeutic target engagement, and as surrogates for assessing downstream effects on disease pathology. Thus, it is critical to elucidate the trajectories of biomarker changes during the natural course of the disease, especially during this dynamic preclinical phase.

Studies of families harboring known ADAD mutations provide a unique and powerful means by which to investigate AD biomarker changes that take place as the disease progresses from its initial asymptomatic/preclinical phase through its symptomatic phase characterized by cognitive decline and eventual dementia. These individuals have long been excluded from observational studies and clinical trials because of the genetic mechanism of their disease. However, despite potential differences in the mechanisms by which Aβ accumulates to form amyloid in the brain (hypothesized to be due to an increase in Aβ1–42 production in ADAD compared to more complex and still poorly understood mechanisms in LOAD) (36), the hallmark pathology of amyloid plaques and neurofibrillary tangles and associated neuronal loss and brain atrophy is similar between the two groups. Yet, individuals with ADAD often exhibit pathologies not frequently observed in LOAD including “cotton wool” plaques, severe cerebral amyloid angiopathy, and Lewy bodies (37), and these differences may affect biomarker profiles. In general, ADAD pathologies develop much earlier in life and to a greater extent than in LOAD, clinically manifesting as faster cognitive decline, additional nonmemory and noncognitive neurological symptoms (for example, apraxia, aphasia, myoclonus, and spastic paraparesis), and higher mortality (3843). However, despite these differences, investigation of the asymptomatic period in ADAD overcomes the three biggest challenges facing similar studies in LOAD: (i) knowing whether future dementia will develop (a virtual certainty in ADAD MCs), (ii) the unknown AAO (relatively predictable estimates within ADAD families), and (iii) possible confounds due to age-related comorbidities (relative lack of other diseases in the younger ADAD individuals). Comparing MCs with NCs (as controls) of similar ages within the same families also minimizes possible genetic influences beyond the disease-causing mutations themselves.

The EYO construct in DIAN permits evaluation of biomarker concentrations as a function of where along the disease trajectory an individual falls, independent of the actual age of dementia onset of their parent. With this strength in mind, the present study demonstrates fluid biomarker profiles consistent with the presence of AD pathological abnormalities 10 to 20 years before the EAO in MCs and before significant impairments in cognitive performance as measured by CDR-SB, MMSE, and a psychometric composite recently reported to be a sensitive measure of cognition in MCs before the onset of dementia (33), in agreement with our initial report from an earlier, smaller group of DIAN participants (22) and a young cohort of individuals from families with the PSEN1 E280A mutation (the Colombian kindred) (44). Although we cannot formally rule out possible aging-related reductions in CSF production and/or turnover (45) as a contributor to the alterations in CSF analytes in MCs, the young age of the DIAN cohort and the lack of change in NCs make this an unlikely possibility. Cross-sectional analyses suggest that concentrations of CSF Aβ1–42 in asymptomatic MCs drop as individuals approach their parental AAO, becoming significantly lower than those in NCs by ~10 years before their estimated age of dementia onset. Consistent with what has been reported in LOAD (1014, 24, 46, 47) and in cognitively normal individuals at risk for LOAD by virtue of their advanced age (10, 24, 47), low concentrations of CSF Aβ1–42 were associated with PIB positivity, despite the overproduction of Aβ42 in these MCs (1014, 24, 46, 47). However, also similar to what is observed in LOAD, the relationship is not perfect; there are several members of the DIAN cohort, including NCs, who exhibit relative low concentrations of CSF Aβ1–42 in the absence of detectable PIB retention in the selected cortical regions. Whether this apparent discrepancy reflects a true pathobiological process (for example, concentrations of CSF Aβ1–42 dropping before Aβ becoming detectable by PIB), simple biological variability across individuals, or a more technical limitation of the fluid assays and/or imaging protocols is the subject of ongoing investigation. In general, the “normal” pattern (high CSF Aβ42, PIB negativity) was observed more often for those who were farthest from their EAO (that is, earlier in the disease course), whereas the AD pattern (low CSF Aβ42, PIB positivity) was more often observed in those closer to or who had surpassed their EAO (that is, later in the course of the disease). Evaluation of the relationship between longitudinal changes in CSF Aβ42 and longitudinal PIB measures within a given individual will lead to a better understanding of the actual timing of these biomarker changes early in the disease process and whether such trajectories differ between individuals with ADAD compared to LOAD. Although the trajectories estimated by cross-sectional baseline data suggested higher concentrations of CSF Aβ1–42 in MCs compared to those in NCs even earlier in the disease process (~10 to 25 years before their EAO), the statistics did not bear this out, likely due to the small number of participants at these extreme age points. Such CSF changes appear to parallel those observed in Down syndrome in which the eventual development of AD occurs because of the presence of three copies of the APP gene on chromosome 21. Elevated concentrations of CSF Aβ1–42 are observed very early in life in Down syndrome (48), whereas concentrations are reduced at later ages (49). Higher concentrations of CSF Aβ1–42 have also been reported in early adulthood (ages 18 to 26 years) in the Colombian ADAD kindred (44).

Although MCs exhibited elevated concentrations of plasma Aβ1–42 throughout the course of the disease (consistent with the known stimulatory effect of these mutations on Aβ1–42 production) (50), in contrast to CSF Aβ1–42, concentrations in plasma did not appear to change as a function of EYO. In addition, plasma Aβ1–42 (and Aβ1–40) concentrations were not related to cortical PIB retention, suggesting that elevated plasma Aβ1–42 may be a systemic effect of ADAD mutations and does not reflect underlying cortical Aβ pathology in the brain.

Consistent with what we observed in the initial DIAN cohort (22), cross-sectional analyses suggest that CSF tau and ptau181 (and their respective ratios with Aβ1–42) begin to increase in MCs in the presymptomatic period (~15 years before their EAO) and continue to increase further with disease progression as has been reported in cross-sectional studies of LOAD (8, 51, 52) and ADAD (53). MCs also displayed elevations in CSF concentrations of the new marker VILIP-1, similar to what has been observed in LOAD (54, 55). VILIP-1 is a neuron-specific intracellular calcium sensor protein that is not a component of neurofibrillary tangles and, thus, when measured in fluids, is considered to be a marker of neuronal cell injury and/or death. Concentrations of CSF VILIP-1 are positively correlated with tau (55), and both increase in response to stroke and acute traumatic brain injury (56). When combined with Aβ1–42, VILIP-1 is a strong predictor of cognitive decline in cognitively normal elders (55) and in persons with very mild symptomatic AD (57), performing as well as or slightly better than the tau/Aβ1–42 ratio. The observation of elevations in CSF VILIP-1 in MCs at least 15 years before their EAO, with concentrations even higher in individuals who are closer to their EAO, suggests a robust phase of neuronal injury and/or death that begins before the onset of cognitive symptoms. Concomitant statistical changes in VILIP-1 in plasma in MCs versus NCs were not observed, although concentrations appeared to be higher in MCs than in NCs, thus warranting further investigation. Although CSF VILIP-1 is not specific for AD, it could be used as a neurodegeneration biomarker outcome in AD clinical trials, especially those targeting tau processing or tangle formation that require assessment of tau-independent markers of neuronal cell injury or death.

In general, biomarker trajectories are typically based on the premise that cross-sectional age-related and/or clinical stage–related changes represent changes that would occur over time with disease progression in single individuals. Indeed, a wealth of cross-sectional studies in LOAD has provided the basis for the proposed patterns of reductions in Aβ1–42 and elevations in tau and ptau that are currently accepted in the field (8, 9). The modeling of cross-sectional baseline data in the current DIAN cohort supports this idea. However, the premise that once tau or other neurodegeneration markers are elevated they continue to increase or stay the same as degeneration progresses is not supported by our longitudinal data. Analyses of within-person change in LOAD have demonstrated little or no change in biomarker concentrations over relatively short time periods (6 months to 2 years) (5862). Whether such findings reflect true biology, the short interval assessed or methodologic shortcomings in fluid collection procedures and/or analysis of longitudinal samples remain to be determined. A closer review of the data, however, reveals between-subject variability in the patterns of tau changes that may be biologically relevant (6366). For example, longitudinal increases in tau have been observed in persons with LOAD who had low tau at baseline (presumably early in the course of the disease), but no difference or even decreases in tau have been observed in those with high baseline concentrations (presumed to be later in the disease process) (65, 67). Consistent with this finding, statistical modeling in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort revealed similar reductions in tau in individuals with AD dementia, but not in those exhibiting mild cognitive impairment (68). Reductions in CSF ptau181 have also been reported in demented persons who were followed for ~3 years (69).

The current observations of increases in tau in MCs early in their disease course (EYO ≤ 0) but decreases in those at later (typically symptomatic) stages (EYO > 0) lend strong support to the general concept that biomarker trajectories may differ as a function of where an individual falls in the neuropathological cascade and, specifically, that markers of the acute phase of neuronal injury increase in early, preclinical stages of AD but then later fall from their peak concentrations as cellular neurodegeneration slows. Consistent with this pattern, a previous report of a single asymptomatic ADAD (APP V717I) MC showed substantial increases in tau and ptau181 over a 4.5-year period very early in the disease process (between 19 and 14 years before their EAO) (70), whereas a longitudinal decrease (or a lack of increase) over 5 years in concentrations of CSF ptau181 has been observed in a small Japanese cohort (n = 4) of symptomatic PSEN1 MCs (71).

The present study is not without weaknesses. Because ADAD and LOAD are similar but not identical in their underlying pathologies, symptoms, and clinical course, the generalizability of the present findings to the more common “sporadic” form remains to be determined. Also, the number of participants in the longitudinal cohort is small, as is the number of repeat CSF samples. Future analyses in the growing DIAN cohort, with greater numbers of longitudinal samples collected per person, will allow us to evaluate the validity and robustness of these biomarker changes. Although the consistency in the patterns of the three different neurodegenerative markers (tau, ptau181, and VILIP-1) lends strong support to the validity of the trajectories, collection of additional samples will permit more elaborate modeling, such as one that does not assume linear changes over time. The small number of participants in the longitudinal cohort is especially problematic for assessing cognitive change over time because evaluation of only two time points does not permit an accurate characterization of the rate of cognitive decline (72). Analyses over longer follow-up times and that account for different levels of cognitive impairment among individuals will permit appropriate modeling of potential nonlinear patterns of cognitive decline. Such data will also allow conclusions to be drawn regarding the temporal relationship between cognitive change and fluid biomarker trajectories over the course of the disease, an evaluation that is not possible in the present small cohort. Another limitation is that we did not examine mutation-specific effects. Evaluation of a larger cohort will permit assessment of mutation-specific effects, as well as possible influences of APOE genotype on biomarker patterns.

In conclusion, the current data suggest a model of disease pathogenesis in the asymptomatic period of ADAD in which MCs display elevated concentrations of Aβ1–42 in the plasma and perhaps in CSF very early in the presymptomatic phase. Subsequently, Aβ1–42 begins to aggregate and accumulate in the brain until a critical threshold is reached, at which time concentrations in the CSF decrease as Aβ1–42 is sequestered in Aβ plaques in brain parenchyma, as has been reported in transgenic mouse models of ADAD (73, 74). Concentrations of Aβ1–42 in the plasma remain high in MCs over the course of the disease, likely due to the sustained contribution of peripheral overexpression of Aβ1–42 in these individuals (50). As plaques continue to develop during the long asymptomatic phase, tangles form, and neurons begin to undergo a robust phase of injury and cell death, as evidenced by increases in the CSF concentrations of tau, ptau181, and VILIP-1. Neurodegeneration continues as the disease progresses through its symptomatic phase, but its rate may be slower relative to earlier stages. This may result in a decrease in absolute concentrations of tau, ptau181, and VILIP-1 in CSF over time in symptomatic individuals.

Although this general model is consistent with data obtained from cross-sectional studies in LOAD (8, 9, 75, 76) and suggests a common pathophysiology for AD because of mutations and the much more common sporadic form, the current longitudinal data suggest a potential modification to the proposed model to incorporate an eventual slowing down of the rate of neuronal injury and death such that the release of these proteins from cells slows relative to its peak release, resulting in lower absolute concentrations in the CSF over time. However, it is also possible that early increases in these markers followed by later decreases may reflect differences in the size of neuronal populations undergoing acute injury/death during the different disease stages (for example, larger populations early followed by smaller populations later). Early elevations may also be due to acute neuronal cell death, whereas apparent later reductions reflect the death of a smaller number of neurons that remain. Finally, it is also possible that the early elevations in CSF tau and VILIP-1 are due to cellular stress that leads to an increase in the normal release of these proteins, and that this stress decreases later in the disease. None of these scenarios necessarily contradicts the findings of accelerations of rates of brain atrophy with disease progression in mild cognitive impairment and LOAD (77, 78) and in ADAD (79, 80), but instead speaks to the possible relationship between the appearance of fluid indicators of cell injury and death and the subsequent structural sequelae of such processes, namely, tissue shrinkage. If corroborated in additional cohorts, this pattern of marker change will likely have an impact on the definition of a positive neurodegenerative biomarker outcome in clinical trials, especially during the symptomatic phase. For example, depending on where a person falls along the pathological cascade, a slowing of the course of neuronal injury/death may be indicated by a slowing of the rate of increase in neurodegenerative markers in individuals who are early in the disease process, but perhaps a stabilization or even a slowing or reversal of the downward trajectory later in the disease, potentially reflected as an increase in these markers. This is an important issue to consider as trials move forward.

MATERIALS AND METHODS

Study design

The objectives of this study were to characterize the patterns of fluid biomarker evidence of amyloid and neuronal pathologies in a large research cohort of individuals from families known to carry ADAD mutations and to test the hypothesis that the degree of biomarker abnormality increases over time with disease progression. Cross-sectional baseline data for four analytes in plasma and five analytes in CSF were obtained by immunoassay and compared in MCs (n = 146) and NCs (n = 96) as a function of the participants’ expected age of symptom onset. Longitudinal data from a subset of individuals (n = 37) were obtained for four analytes in CSF to evaluate within-person change in biomarker concentrations over time.

Cohort

Participants at 50% risk of carrying an ADAD mutation in one of three genes (APP, PSEN1, and PSEN2) were enrolled in the DIAN [National Institute on Aging (NIA) U19 AG032438; J.C.M., principal investigator] at 1 of 11 performance sites (http://www.dian-info.org, clinicaltrials.gov number NCT00869817) (81). All study procedures were approved by the Washington University Human Research Protection Office and the local institutional review boards of each participating site. Written informed consent (or assent with proxy consent if capacity to consent was impaired) was obtained from all participants before their participation. Data in the present follow-up study were from participants with clinical, genetic, amyloid imaging (PIB PET), and fluid (CSF and plasma) measures that were collected and passed quality control (QC) standards as of Data Freeze 4 (DF4). This cohort of 242 participants included 110 individuals described in our initial cross-sectional study (22) (n = 29 NCs and n = 62 MCs with CSF, and n = 34 NCs and n = 76 MCs with plasma). No longitudinal data were presented in our earlier study.

Clinical evaluation

Participants underwent an extensive clinical evaluation, which included family history of AD, personal medical history, and physical and neurologic examination. Cognitive status was determined with the CDR in accordance with standard protocols and criteria (8284). CDR 0 indicates cognitive normality [referenced as asymptomatic (AS)], whereas CDRs of 0.5, 1, 2, or 3 correspond to cognitive impairments that are considered to be very mild, mild, moderate, or severe, respectively [referenced as symptomatic (S)]. A clinical diagnosis of AD in individuals who were CDR 0.5 or greater was based on the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association criteria (85). The CDR-SB and scores on the MMSE (23) were also evaluated with standardized protocols. DIAN investigators involved in all clinical, cognitive, and other assessments were without knowledge of the mutation status of the participant. No research data, including genetic status, were provided to participants as part of the DIAN study, and special care was taken to prevent unblinding of participants to their mutation status (actual or perceived) in the presentation of all data (as is required by DIAN). Participants wishing to know their status were offered private genetic counseling and testing at no cost and no disclosure to any other entity. The parental AAO was determined by a semistructured interview with family members to estimate the age of the first progressive cognitive decline. The age at clinical onset of ADAD is similar between generations (86) and is affected mostly by mutation type and background family genetics (87).

Genetic analysis

DNA sequencing for familial AD mutations and APOE genotyping was performed by the DIAN Genetics Core at Washington University. Ambient blood samples were shipped from the DIAN performance sites to both the National Cell Repository for Alzheimer’s Disease (NCRAD) and the DIAN Genetics Core, and DNA was extracted at both sites using standard procedures. DNA sequencing of APP, PSEN1, and PSEN2 was performed by DIAN Genetics Core personnel, using Sanger sequencing methods on an ABI 3130xl, to determine the presence or absence of a disease-causing mutation (88). APOE genotyping was performed with an ABI predesigned real-time TaqMan assay “rs7412 and rs429358” according to the manufacturer’s protocol (Applied Biosystems). DNA fingerprinting was performed with the Cell ID kit, using short tandem repeat (STR) analysis of 10 specific loci in the human genome, 9 STR loci, and Amelogenin for gender identification (Promega #G9500) to confirm that DNA samples obtained by NCRAD and the DIAN Genetics Core were from the same individual. DNA sequencing, fingerprinting, and genotyping were performed on DNA from NCRAD and the DIAN Genetics Core in parallel for each individual, and the data were compared for QC purposes. All individuals included in this analysis have 100% concordant data for each DNA sample and are defined as NCs or MCs.

Fluid collection

Protocols for the collection of blood (for plasma) and CSF are consistent with the biofluid protocol of the ADNI (http://www.adni-info.org/) to facilitate comparisons between biomarker measures in LOAD and ADAD. Briefly, fluids were collected at 8:00 a.m. after overnight fasting. Blood was obtained by venipuncture into polypropylene tubes. Plasma (in EDTA) was prepared by standard methods, transferred to polypropylene transfer tubes, and immediately frozen on dry ice. After blood draw, CSF (≥15 ml) was collected by standard LP (typically L4/L5) using sterile technique. CSF was collected into polypropylene tubes and immediately frozen on dry ice. Frozen samples were shipped on dry ice the same day (or batch-shipped from non-U.S. sites after storing at −80°C) to the DIAN Biomarker Core at Washington University. Frozen samples were subsequently thawed, aliquoted (300 to 500 μl) into polypropylene tubes, and stored at −84°C until analyzed.

Fluid analysis

All fluid samples were analyzed by the DIAN Biomarker Core at Washington University. CSF concentrations of Aβ1–42, tau, and ptau181 were measured by immunoassay using Luminex bead-based multiplexed xMAP technology (INNO-BIA AlzBio3, Innogenetics). The assay has intra-assay precisions of 2.5 to 6% for Aβ42, 2.2 to 6.3% for tau, and 2 to 11% for ptau181 (68, 89). Concentrations of CSF Aβ1–40 were measured by plate-based enzyme-linked immunosorbent assay (ELISA) (research prototype INNOTEST Aβ1–40, Innogenetics), as were the concentrations of Aβ1–42 (INNOTEST Aβ1–42, Innogenetics), to permit comparison of Aβ1–42 concentrations in the DIAN cohort to those obtained in the many published biomarker studies in LOAD, which used the INNOTEST kit (14, 9096). Concentrations of plasma Aβ species (Aβ1–40, Aβ1–42, Aβx–40, and Aβx–42) were measured by xMAP (INNO-BIA Plasma Aβ Forms Multiplex Assay, Innogenetics). Fluid samples were also analyzed for VILIP-1 using a two-site immunoassay implemented via a microparticle-based immunoassay system (Erenna, Singulex). This assay uses a monoclonal antibody coated on magnetic beads for the “capture” step and affinity-purified sheep antibody labeled with Alexa Fluor 647 for the detection step. The assay has an intra-assay precision of 4.4%, an interassay precision of 6.2%, and a lower limit of quantitation of 2.1 pg/ml in CSF and 3.9 pg/ml in plasma. For all assays, values had to meet QC standards, including CVs ≤25%, kit “controls” within the expected range as defined by the manufacturer, and measurement consistency of two common CSF samples that were included on each plate. Two-hundred thirty-seven plasma samples and 206 CSF samples passed analytical QC and, therefore, contributed to the current data set.

Longitudinal CSF samples collected over 2 to 3 years after baseline were available from a subset of participants (n = 37) at the time of DF4. For longitudinal analyses, samples from the same individuals were reanalyzed for Aβ1–42 (by AlzBio3 only), tau, ptau181, and VILIP-1 after being loaded onto the same assay plate to compare within-person change in analyte concentrations over time.

PET PIB image analysis

Participants were evaluated for fibrillar brain Aβ using PET with the amyloid imaging agent PIB (97) within 3 months of clinical evaluation and fluid collection. All images underwent QC inspection and preprocessing at a DIAN centralized site. Magnetic resonance imaging (MRI) was performed on qualified 3-T scanners at each site with initial and ongoing QC and matching between site scanners performed according to the ADNI protocol and QC (see http://www.adni-info.org). The volumetric scan consisted of an 8- to 10-min three-dimensional (3D) T1-weighted image (that is, MP-RAGE) with 1.0 × 1.0 × 1.2–mm resolution. Aβ imaging was performed with [11C]PIB-PET scans, acquired with a 30-min dynamic (4 × 5–min frames), 3D acquisition, beginning 40 min after a bolus injection of 15 mCi of PIB. The T1-weighted MRI scans were processed through the FreeSurfer image analysis suite version 5.1 using Dell PowerEdge 1950 servers with Intel Xeon processors running CentOS 5.5 Linux. FreeSurfer involves cortical reconstruction and volumetric segmentation (http://surfer.nmr.mgh.harvard.edu/). The processing routine includes segmentation of the subcortical white matter and deep gray matter volumetric structures, extraction of the cortical surfaces, and parcellation of cortical regions. PET frame-to-frame motion correction and PET-MRI alignment were accomplished through standard registration techniques written with in-house software (98). PET images were then transformed into each individual participant’s MRI space. For each FreeSurfer region of interest, SUVRs were calculated using a brainstem (pons) gray matter reference. To minimize the impact of partial volume effects on the PET signal, a regional spread function–based approach for partial volume correction was used for all regional measurements (99). The mean cortical SUVR was calculated as the summary of regions within the prefrontal cortex, gyrus rectus, lateral temporal, and precuneus regions. As a preliminary analysis in this ADAD cohort, a k-means clustering algorithm (100) implemented in R (http://www.R-project.org/) was used to estimate a cutoff point for PIB positivity. Three clusters were defined, and the threshold between the first and second cluster, 0.85, was considered the cutoff for the current study.

Longitudinal cognitive performance measures

Within-individual performance over time on a composite of three psychometric measures shown to be sensitive to cognitive decline in asymptomatic MCs (CDR 0) in DIAN (33) was evaluated in the longitudinal cohort (n = 36). These standardized tests included measures of episodic memory (Logical Memory IA—Immediate Recall and Logical Memory IIA—Delayed Recall from the Wechsler Memory Scale—Revised) (34) and a measure of speeded visual spatial processing (Digit Symbol Coding from the Wechsler Adult Intelligence Scale—Revised) (35). Z scores for each individual measure were derived from the longitudinal sample and averaged to form a composite. Higher Z scores are indicative of better performance.

Statistical analyses

The EYO for each DIAN participant was calculated as the participant’s age at clinical evaluation minus the reported parental AAO. For example, if a participant’s age was 35 years and the reported parental AAO was 45, then the EYO for this participant would be −10. Participant characteristics of the cohort at DF4, as well as at baseline for the subset of participants with longitudinal data, were summarized as mean ± SD for continuous characteristics and as n (column percent) for categorical characteristics. Continuous participant characteristics were compared between groups with general linear mixed models, dichotomous characteristics were compared with mixed-effects logistic regression models (101), and family mutations were compared with a mixed-effects multinomial logistic regression model (102). All of the above models incorporated a random intercept at the family level to account for within-family dependencies. Pairwise testing was conducted only after an omnibus test suggested significant (P < 0.05) differences between groups.

In general, a biomarker “trajectory” is defined as the pattern of biomarker change over time. In DIAN cross-sectional analyses, participants’ EYO served as a proxy for time. To explore the trajectories of the biomarkers over the range of EYOs, nonparametric curves were fit to the cross-sectional (baseline) data using a locally weighted regression method (LOESS) (103). General linear mixed models, with linear, quadratic, or cubic trends where appropriate, were then used to estimate the trajectories of the biomarkers (and CDR-SB and MMSE) over EYO and subsequently test for differences in mean values between the participant groups (NC and MC) at 5-year intervals along the EYO distribution, using approximate t tests. Comparisons between the participant group biomarker concentrations and trajectories were also made through approximate t or F tests on appropriate interactive effects in these models. Additionally, the trajectories in both participant groups were tested relative to zero with approximate t tests for linear trajectories, or with approximate F tests for nonlinear trajectories. To account for the within-family dependencies of the biomarkers, these models incorporated random intercepts at the family level, which also allowed differential variances across the participant groups (NC and MC). Given that the current study aimed to generate scientific hypotheses that will be critically tested in the future, values were not adjusted for multiple comparisons.

Longitudinal analyses used general linear mixed models with random intercepts/slopes (32) at the subject level and with random intercepts at the family level to quantify and statistically evaluate the within-person rate of change in biomarker concentrations and psychometric test performance. The fixed effects of these models included the participant groups (NC and MC) and where participants fell with respect to their EYO (EYO ≤ 0 versus EYO > 0) at baseline, as well as their interactions with time. Consequently, different slopes across participant groups before and after baseline EYO 0 were estimated, thus facilitating the comparisons of the longitudinal annual rate of change across the two groups. All general linear mixed models in the longitudinal analyses were estimated using restricted maximum likelihood estimation, with the approximate F test denominator degrees of freedom based on the method of Kenward and Roger (104). To protect participant confidentiality regarding mutation status and investigator blinding (as required by the DIAN protocol), longitudinal individual data points were plotted with reference to relative EYO, not specific EYO. P values reported for the Pearson or Spearman correlation coefficients were computed on the basis of 2000 bootstrap replications using the percentile-t method (104). All statistical analyses were performed with SAS version 9.3 (SAS Institute Inc.), and statistical significance was defined as P < 0.05.

Supplementary Materials

www.sciencetranslationalmedicine.org/cgi/content/full/6/226/226ra30/DC1

Table S1. Comparison of estimated mean plasma biomarker concentrations between NCs and MCs at defined EYO.

Table S2. Comparison of estimated mean CSF biomarker concentrations and cognitive performance between NCs and MCs at defined EYO.

Table S3. Estimated mean (SE) annual within-individual rate of change in CSF biomarkers and a psychometric composite score in NCs and MCs in (A) unadjusted analyses and (B) analyses adjusted for gender and APOE genotype.

Fig. S1. Mean (SE) % CVs in analyte measures between longitudinal samples obtained from a given individual (black) compared to within a given sample run in duplicate (white).

Fig. S2. Evaluation of psychometric test performance in NCs and MCs in the longitudinal cohort.

References and Notes

  1. Acknowledgments: We acknowledge the technical and administrative support of M. Amos, T. Blazey, D. Carrell, S. Chakraverty, S. Fague, E. Macy, K. Moulder, A. Oliver, A. Shah, and J. Hassenstab regarding psychometrics, and the contributions of the DIAN research and support staff at each of the participating sites. We acknowledge the altruism of the DIAN participants and their families. Funding: Data collection and sharing for this project was supported by DIAN (U19 AG032438; to R.J.B., T.L.S.B., V.D.B., N.J.C., A.M.F., B.G., A.M.G., D.M.H., M.S.J., D.M., R.N.M., C.L.M., R.M., J.C.M., J.M.R., M.N.R., S.S., P.R.S., R.A.S., and C.X.) and a cooperative agreement grant (NCRAD U24 AG21886) funded by the National Institute on Aging (NIA), the DIAN Pharma Consortium [Alzheimer’s Immunotherapy Program (Janssen Alzheimer Immunotherapy and Pfizer Inc. Alliance), Biogen Idec Inc., Eisai Inc., Elan Pharmaceuticals Inc., Eli Lilly and Company, En Vivo Pharmaceuticals, Genentech Inc., F. Hoffman–La Roche Ltd., Mithridion Inc., Novartis International AG, Pfizer Inc., and Sanofi-Aventis Groupe] (to R.J.B., T.L.S.B., V.D.B., A.M.F., A.M.G., and C.X.), and in part by R01 EB009352 (to D.M.), P30 NS048056 (to D.M.), P50AG016750 (to J.M.R.), the National Institute for Health Research Queen Square Dementia Biomedical Research Unit (to M.N.R.), and the JO & JR Wicking Trust grants 13026 and 20821 (to P.R.S.). This article has been reviewed by DIAN Study investigators for scientific content and consistency of data interpretation with previous DIAN Study presentations/publications. Author contributions: R.J.B., V.D.B., A.M.F., D.M.H., and J.C.M. contributed to the conception and design of the study. Data analysis was performed by T.L.S.B., A.M.F., M.S.J., J.H.L., and C.X. Data were interpreted by R.J.B., T.L.S.B., N.J.C., A.M.F., A.M.G., D.M.H., M.S.J., J.H.L., J.C.M., and C.X. Data were collected by R.J.B., T.L.S.B., A.M.F., B.G., A.M.G., J.H.L., R.N.M., C.L.M., R.M., J.M.R., M.N.R., S.S., P.R.S., and R.A.S. Data were assembled by A.M.F., M.S.J., D.M., and C.X. A.M.F. drafted the article, and R.J.B., T.L.S.B., V.D.B., N.J.C., B.G., A.M.G., D.M.H., M.S.J., D.M., R.N.M., C.L.M., R.M., J.C.M., J.M.R., M.N.R., S.S., P.R.S., R.A.S., and C.X. provided critical revision for important intellectual content. R.J.B., B.G., R.N.M., C.L.M., R.M., J.M.R., M.N.R., S.S., P.R.S., and R.A.S. provided study materials or patients. Statistical expertise was provided by M.S.J. and C.X. V.D.B. and J.C.M. obtained funding. Competing interests: A.M.F. consults for Roche and Eli Lilly. R.J.B. receives research funding from AstraZeneca, Merck, and Eli Lilly. He is a scientific advisor and co-founder of C2N Diagnostics, a scientific advisory board member for EnVivo, and a consultant for Merck, Eisai, Novartis, Medtronic Scientific, and Genentech. He has been an invited speaker for Elan and AstraZeneca. A.M.G. reports research funding from Pfizer, Genentech, and iPierian; royalties from Taconic for the tau mutation patent; and receiving industry-funded travel and honoraria from Pfizer for a lecture in 2010. She also has provided expert testimony and received compensation from Howrey and Associates, Finnegan HC, and Dickstein Shapiro in 2011. T.L.S.B. consults for Eli Lilly and receives research funding from Avid Radiopharmaceuticals. She has also provided expert testimony and received compensation from Kujawski & Associates in 2011. B.G. consults for Bayer. R.N.M. reports consulting for Alzhyme Pty Ltd. and is on the scientific advisory board of Hollywood Private Hospital and on the board for the Cooperative Research Centre for Mental Health. C.L.M. reports consulting for Eli Lilly and Prana Biotechnology. J.M.R. reports consulting for Takeda Pharmaceuticals and StemCells Inc. M.N.R. reports being a member of the Data Monitoring Committee for Servier DMC Phase 2B AD Study S38093 and the Bapineuzumab Independent Safety Monitoring Committee for Janssen Al/Pfizer. S.S. reports providing consultation to Janssen AI, Avid-Lilly, Baxter, Pfizer, Athena, Merck, Bristol-Myers Squibb, GE, Roche, and Sanofi. He received honoraria from Athena Diagnostics. His hospital received research support for clinical trials from Janssen AI, Bristol-Myers Squibb, Biogen, Pfizer, Merck, Genentech, GE, Roche, Functional Neuromodulation, NIA Alzheimer’s Disease Neuroimaging Initiative, and the Alzheimer’s Association. P.R.S. reports receiving speaking honoraria from Janssen Pharmaceuticals. R.A.S. reports consulting for Bayer, Biogen-IDEC, Bristol-Myers Squibb, Eisai, Janssen Alzheimer Immunotherapy, Pfizer, Roche, and Avid Radiopharmaceuticals. D.M. reports consulting for Avid Radiopharmaceuticals. J.H.L. reports being named on patents related to the use of VILIP-1. These are being managed by Washington University in accordance with University policy. D.M.H. reports consulting for Bristol-Myers Squibb, AstraZeneca, and Genentech, and is on the scientific advisory board of C2N Diagnostics. His research grant support is from the NIH, Ellison Medical Foundation, Cure Alzheimer’s Fund, AstraZeneca, C2N Diagnostics, and Integrated Diagnostics. J.C.M. has or is currently participating in clinical trials of anti-dementia drugs sponsored by Janssen Immunotherapy, Eli Lilly and Company, and Pfizer. He has served as a consultant for or has received speaking honoraria from Eisai, Esteve, Janssen Alzheimer Immunotherapy Program/Elan, GlaxoSmithKline, Novartis, and Pfizer. He receives research support from Eli Lilly/Avid Radiopharmaceuticals. The other authors declare no competing interests. Data and materials availability: J.H.L. is a co-inventor on patent 11/630582 (2005) (Markers for brain damage) and patent 60957132 (2008) (Alzheimer’s diagnosis). Data and tissue generated by DIAN are available to qualified investigators according to DIAN’s data and tissue sharing policies (http://www.dian-info.org/resourcedb/). The DIAN Genetics Core follows the NIH guidelines on the sharing of genetic material and associated data as stated in the Alzheimer’s Disease Genetics Sharing Plan (http://www.nia.nih.gov/research/dn/alzheimers-disease-genetics-sharing-plan).
View Abstract

Navigate This Article