Different AT(N) profiles and clinical progression classified by two different N markers using total tau and neurofilament light chain in cerebrospinal fluid

Participants

J-ADNI was initiated to discover the fluid and imaging biomarkers of AD using a harmonised protocol with ADNI.12 13 Briefly, volunteer participants aged between 60 and 84 years were recruited from 38 clinical sites in Japan. Cognitively unimpaired (CU) subjects, subjects with MCI, and patients with AD dementia (ADD) were enrolled into J-ADNI using criteria consistent with those of ADNI.13 Their clinical and neuropsychological data were obtained from the National Bioscience Database Center (https://humandbs.biosciencedbc.jp/en/hum0043-v1).

Out of 715 volunteers assessed for eligibility, 537 met the criteria and were enrolled. Out of 537 participants recruited in J-ADNI (CU, 154; MCI, 234; ADD, 149), 4 withdrew their consent. Of the 533 remaining participants, 194 (CU, 53; MCI, 86; ADD, 55) underwent lumbar puncture. The incidence of postdural puncture headache was 2.6%, and that of severe postdural puncture headache that required hospitalisation was 0.7%. All these 194 participants were analysed using AD core biomarkers including Aβ42, tau phosphorylated at threonine 181 (p-tau181), and t-tau. Due to sample availability, CSF NfL was measured in 177 participants (CU, 46; MCI, 82; ADD, 49). At 12 months, longitudinal changes in CSF biomarkers classified by AT(N) profiles were analysed in 126 participants (CU, 38; MCI, 56; ADD, 32) (online supplemental figure 1).

Lumbar puncture and biochemical analysis

CSF was collected by lumbar puncture, transferred into polypropylene tubes followed by freezing and shipped to the J-ADNI Biomarker Core at Niigata University. CSF was aliquoted at a volume of 0.5 mL and stored at ‒80℃ until the assay. The CSF concentrations of Aβ42, p-tau181, and t-tau were examined using on AlzBio3 kit (Fujirebio, Ghent, Belgium), and that of NfL was measured using R-PLEX Human Neurofilament L Antibody Set (Meso Scale Discovery, Rockville, MD). All analyses were conducted in duplicate by experienced laboratory personnel blinded to the clinical diagnosis. The intra-assay and interassay coefficients of variation were <20% for all assays. The laboratory at Niigata University participates in the Alzheimer’s Association external quality control programme for CSF biomarkers.14

We previously used CSF Aβ42<333 pg/mL as the cut-off value for Aβ positivity.12 15 Thereafter, we have established a protocol for AD core biomarker measurements unified with the ADNI Biomarker Core (PI: Leslie M. Shaw, PhD). We used this unified protocol for remeasuring the CSF biomarkers. Subsequently, we conducted the area under the receiver operating characteristic curve analysis (PET Aβ negative (PET Aβ−, n=47) vs positive (PET Aβ+, n=53); CU with PET Aβ− (n=31) vs ADD with PET Aβ+ (n=22); CU (n=53) vs ADD (n=56)), and calculated the optimal cut-off values according to Youden’s index (online supplemental figures 2 and 3). Furthermore, we used Gaussian mixture models (GMMs) for calculating the cut-off value of CSF biomarkers (n=194), excluding NfL, which was unsuitable for GMMs because of the unimodal distribution (online supplemental figure 2).

PET image acquisition and clinical evaluation

All PET images underwent the J-ADNI PET quality control process as previously described.16 Cognitive performance was assessed using the Mini-Mental State Examination (MMSE), Alzheimer’s Disease Assessment Scale-Cognitive Subscale (ADAS-Cog), and the sum of boxes of the Clinical Dementia Rating (CDR-SB). Instrumental activities of daily living were assessed using the Functional Assessment Questionnaire (FAQ). In this study, when the CDR changed from 0 or 0.5 to ≧1 during a follow-up, the patient was considered to have progressed to dementia.

Statistical analysis

Data were analysed statistically using GraphPad Prism (V.8.2.0; GraphPad Software, La Jolla, California, USA) and the software R. For continuous variables, we used the Mann-Whitney U test for comparing two groups and the Kruskal-Wallis test for comparing multiple groups, followed by Dunn’s multiple-comparison test. For categorical variables, groups were compared using the χ² test. The correlation between two data sets was assessed using Spearman’s rank-correlation coefficient. For the longitudinal analyses of changes in CSF biomarker, we compared slopes with zero by linear regression model analyses. The covariates included age, sex and education years. For the longitudinal analyses of clinical score changes, independent variables including combined AT(N) groups, follow-up time, age, sex, education years and the interactions between AT(N) groups and follow-up time were examined using the linear mixed model (LMM). Additionally, random slopes and random intercepts of the follow-up time within subjects served as the random factors for the longitudinal analyses of the clinical scores. P values were adjusted by false discovery rate to avoid type I error.

Demographics of participants

At baseline, CSF samples were collected from 194 participants with CU (n=53, 27.3%), MCI (n=86, 44.3%) and ADD (n=55, 28.4%). Of the 194 participants, 100 (51.5%) were analysed by Aβ PET imaging, and half of them (53.0%) were Aβ-positive (online supplemental table 1). Due to sample availability, 177 (91.2%) of the 194 participants had CSF NfL measurements at the baseline, and 126 (64.9%) underwent follow-up lumbar puncture after 12 months.

Cross-sectional analysis of CSF biomarkers

The correlations between baseline characteristics and CSF biomarkers were analysed. Both the MCI and ADD groups showed significantly lower CSF Aβ42 level and higher p-tau181, t-tau and NfL levels than the CU group. Additionally, the CSF Aβ42 level was significantly lower in the ADD group than in the MCI group (online supplemental figure 4A). In all groups, age showed a significant positive correlation with p-tau181, t-tau, and NfL level except for Aβ42 (online supplemental figure 4B). Years of education also positively correlated with CSF Aβ42 level but not with p-tau181, t-tau, nor NfL level (online supplemental figure 4C). In addition, males showed significantly higher CSF NfL levels than females (online supplemental figure 4D). Both APOE ε4 heterozygous and homozygous carriers showed significantly lower CSF Aβ42 levels and higher p-tau181, t-tau and NfL levels than non-carriers (online supplemental figure 4E).

Next, correlations among CSF biomarkers were analysed. We found that Aβ42 level moderately negatively correlated with p-tau181, t-tau, and NfL levels. As expected, p-tau181 and t-tau levels were highly correlated (r=0.7923, p<0.0001). NfL level showed moderately positive correlations with p-tau181 (r=0.2487, p=0.0008) and t-tau levels (r=0.4907, p<0.0001) (online supplemental figure 5).

AT(N) classification at baseline

We used CSF Aβ42 as the A marker, p-tau181 as the T marker, and t-tau or NfL as the N marker. AT(N)_tau and AT(N)_NfL were defined using t-tau and NfL as the N marker, respectively. We classified the participants into eight AT(N) categories.

The cut-off value was compared by different methods. When comparing clinical status (CU vs ADD) with PET status (PET Aβ− vs PET Aβ+), the cut-off values were lower for Aβ42 and higher for p-tau181, t-tau and NfL (table 1). When the PET Aβ status and clinical status were combined (CU with PET Aβ− vs ADD with PET Aβ+), the cut-off values were intermediate between the PET Aβ status and the clinical status only, and close to GMM-calculated cut-off values (table 1). Thus, hereafter, the cut-off values used in this study were CSF Aβ42<359.6 pg/mL (A+), p-tau181 >30.6 pg/mL (T+), t-tau >105.3 pg/mL (N+) and NfL >2650 pg/mL (N+).

We showed the demographic and clinical variables among the eight AT(N) biomarker categories in the AT(N)_tau classification (table 2, upper half). The proportion of CU decreased from 33% to 60% in the A− groups to 0%–19% in the A+ groups, whereas that of ADD increased from 0% to 25% in the A− groups to 40%–50% in the A+ groups.

Next, we determined the characteristics of the eight AT(N) profiles in the AT(N)_NfL classification (table 2, lower half). The proportion of CU decreased from 31% to 69% in the A− groups to 0%–35% in the A+groups, whereas that of ADD increased from 0% in the A− groups to 25%–62% in the A+ groups.

To determine the frequency of biological AD, we classified 177 subjects by using the AT system comprising CSF Aβ42 and p-tau181. The subjects were then classified into A−T− (n=55, 31.1%), A−T+ (n=7, 4.0%), A+T− (n=41, 23.2%) and A+T+ (n=74, 41.8%) (figure 1). A−T− accounted for 69.9% in CU, whereas A+T− and A+T+ accounted for 68.3% in MCI. In ADD, A+T− was 36.7% and A+T+ was 61.2%.

Request permission

AT(N) classification: comparison between t-tau and NfL

We compared the frequencies of AT(N) categories between AT(N)_tau and AT(N)_NfL. In AT(N)_tau, the most common was A+T+(N)+ (n=64, 36.2%), followed by A−T−(N)− (n=52, 29.4%) and A+T−(N)− (n=37, 20.9%) (figure 1). Considering the high correlation between t-tau and p-tau181 (online supplemental figure 5), CSF t-tau may not be a fully independent marker of neurodegeneration in the AD continuum.

In AT(N)_NfL, the frequencies of the A−T−(N)−, A+T−(N)−, and A+T+(N)+ categories decreased to 22.0% (n=39), 11.3% (n=20), and 23.7% (n=42) compared with AT(N)_tau, respectively (figure 1). Thus, the subsets of participants in the A−T− and A+T− categories with neurodegeneration (A−T−(N)_NfL + and A+T−(N)_NfL+) were classified into (N)− in AT(N)_tau (figure 1). Supporting this finding, the subsets of participants in the A−T−(N)_tau− and A+T−(N)_tau− categories showed elevated NfL levels (online supplemental figure 6). In contrast, a subset of participants in the A+T+ category with undetectable neurodegeneration (A+T+(N)_NfL−) showed elevated t-tau levels (online supplemental figure 6); thus, they were classified into (N)+ in AT(N)_tau (figure 1).

Longitudinal changes of AT(N) profiles

In 126 participants with follow-up CSF examination at 12 months, changes in the levels of most of the biomarkers were not statistically significant. After 12 months, the p-tau181 level significantly elevated in the A−T−(N)− category by both AT(N) classifications, in A+T−(N)− by AT(N)_tau classification, and in A+T−(N)+ by AT(N)_NfL classification (online supplemental table 3, online supplemental figure 7).

We assessed the longitudinal changes of the AT(N) profiles at 12 months. The AD continuum biologically progressed, and the progression rate differed between AT(N) profiles at the baseline (figure 2). In the AT(N)_tau classification, the progression rate was 2.1% (1 of 47) among A− groups. A+T−(N)− progressed to A+T+(N)− and A+T+(N)+ in five and two participants, respectively. All four participants with A+T−(N)+ progressed to A+T+(N)+. Thus, the progression rate of these A+T− to A+T+ was 42.3%. One participant with A+T+(N)− progressed to A+T+(N)+ (14.3%) (figure 2A).

Request permission

In the AT(N)_NfL classification, the progression rate was 2.1% (1 of 47) in the A− groups. The progression rate from A+T− to A+T+ was 38.5%. Five A+T+(N)− participants progressed to A+T+(N)+ (20.0%) (figure 2B). Hence, participants with the A− profile rarely progressed to A+ within 12 months. Conversely, approximately 40% of participants with A+T− progressed to A+T+ and 10%–20% of participants with A+T+(N)− progressed to A+T+(N)+ within 12 months in either the AT(N)_tau or AT(N)_NfL classification (figure 2). Notably, longitudinal changes of AT(N) profiles were different in A+T−(N)+ and A+T+(N)− categories between the AT(N)_tau and AT(N)_NfL classifications (figure 2).

Longitudinal change of cognitive functions

Owing to the small sample size of some of the AT(N) categories, we categorised eight AT(N) profiles into three groups, namely, the normal biomarker (A−T−(N)−), AD continuum (A+T−/+(N)−/+) and non-AD pathological change (A−T−/+(N)−/+) groups. At the baseline, the AD continuum group showed significantly lower MMSE and higher ADAS-Cog, CDR-SB and FAQ scores than the normal biomarker group (figure 3, online supplemental table 2). In the AT(N)Nfl classification, the AD continuum group showed significantly lower MMSE and higher ADAS-cog, CDR-SB and FAQ scores than the non-AD pathological change group. No such significant differences were observed in the AT(N)_tau classification (figure 3).

Request permission

We conducted LMM analysis to evaluate cognitive decline assessed by four clinical measures (MMSE, ADAS-Cog13, CDR-SB and FAQ) during the follow-up period up to 36 months. All the clinical measures in the AD continuum and non-AD pathological change groups declined faster than in the normal biomarker group, except for the CDR-SB of the non-AD pathological change group in AT(N)_NfL classification (table 3, figure 4).

Request permission

Clinical conversion into dementia

Of 139 participants, 57 (41.0%) clinically converted into dementia during 36 months of follow-up. The subjects who converted to dementia exhibited significantly higher levels of t-tau and NfL at the baseline than the non-converters (t-tau, p<0.001; NfL, p=0.0033).

Cox proportional hazard analysis showed that the AD continuum and non-AD pathological change groups converted into dementia more frequently than the normal biomarker group in the AT(N)_tau classification (figure 5A). In the AT(N)_NfL classification, only the AD continuum group converted into dementia more frequently than the normal biomarker group (figure 5B). Discordance of prognosis in the non-AD pathological change group between the AT(N)_tau and AT(N)_NfL classifications suggests that CSF t-tau elevation without Aβ42 reduction (A−(N)_tau+) may be related to a higher rate of conversion to dementia; conversely, no such relationship was found in the case of CSF NfL elevation without Aβ42 reduction.

Request permission