Amyloid-β levels and cognitive trajectories in non-demented pTau181-positive subjects without amyloidopathy

Abstract

Phosphorylated Tau181 (pTau181) in CSF and recently in plasma has been associated with Alzheimer’s disease. In the absence of amyloidopathy, individuals with increased total Tau levels and/or temporal lobe atrophy experience no or only mild cognitive decline compared with biomarker-negative controls, leading to the proposal to categorize this constellation as suspected non-Alzheimer's disease pathophysiology (SNAP). We investigated whether the characteristics of SNAP also applied to individuals with increased CSF-pTau181 without amyloidopathy.

In this long-term observational study, 285 non-demented individuals, including 76 individuals with subjective cognitive impairment and 209 individuals with mild cognitive impairment, were classified based on their CSF levels of pTau181 (T), total Tau (N), amyloid-β₄₂ (Aβ₄₂) and Aβ₄₂/Aβ₄₀ ratio (A) into A+T+N±, A+T–N±, A–T+N±, and A–T–N–. The longitudinal analysis included 154 subjects with a follow-up of more than 12 months who were followed to a median of 4.6 years (interquartile range = 4.3 years). We employed linear mixed models on psychometric tests and region of interest analysis of structural MRI data.

Cognitive decline and hippocampal atrophy rate were significantly higher in A+T+N± compared to A–T+N±, whereas there was no difference between A–T+N± and A–T–N–. Furthermore, there was no significant difference between A–T+N± and controls in dementia risk [hazard ratio 0.3, 95% confidence interval (0.1, 1.9)]. However, A–T+N± and A–T–N– could be distinguished based on their Aβ₄₂ and Aβ₄₀ levels. Both Aβ₄₀ and Aβ₄₂ levels were significantly increased in A–T+N± compared to controls.

Long term follow-up of A–T+N± individuals revealed no evidence that this biomarker constellation was associated with dementia or more severe hippocampal atrophy rates compared to controls. However, because of the positive association of pTau181 with Aβ in the A–T+N± group, a link to the pathophysiology of Alzheimer’s disease cannot be excluded in this case. We propose to refer to these individuals in the SNAP group as ‘pTau and Aβ surge with subtle deterioration’ (PASSED).

The investigation of the circumstances of simultaneous elevation of pTau and Aβ might provide a deeper insight into the process under which Aβ becomes pathological.

In a long-term observational study, Oberstein et al. find that the biomarker constellation of increased pTau181, Aβ40 and Aβ42 in CSF is not associated with dementia. Further investigation of this biomarker profile may provide insights into the circumstances under which Aβ becomes pathological.

Introduction

Surrogate markers for the neuropathological hallmarks of Alzheimer’s disease, amyloid plaques (A) and the neurofibrillary tangles (T), are widely used in proposals for a biological definition of Alzheimer’s disease.^1–3 Examples for well established biomarkers in CSF for amyloidopathy (A+) are lowered levels of amyloid-β₄₂ (Aβ₄₂) or the lowered ratio of Aβ₄₂ to Aβ₄₀ (Aβ₄₂/Aβ₄₀) as well as high levels of phosphorylated Tau at threonine 181 (pTau) for tauopathy (T+).^1–6 Alternatively, PET tracers for the detection of cortical aggregation of amyloid (A+) or Tau (T+) can be used.^1–3 The classification of the National Institute on Aging and Alzheimer’s Association (NIA-AA) working group from 2018 also included biomarkers for neurodegeneration (N) in their scheme, i.e. elevated levels of total Tau (tTau) in CSF, atrophy in structural MRI and/or fluorodeoxyglucose-(¹⁸F)-hypometabolism in PET.³ The advantage of a biological definition of Alzheimer’s disease is that the biomarkers are changed years before the stage of dementia so that potential treatment can be started before massive neuronal loss has occurred, i.e. in stages with subjective (SCI) or mild cognitive impairment (MCI).⁷ Various studies have so far shown that individuals in whom all three categories indicate Alzheimer’s disease (A+T+N+) have a higher probability of developing dementia in comparison to individuals who are A–, T– and/or N–.^8–14 However, the parallel use of different biomarker categories not only increases the diagnostic reliability in the previously mentioned cases, but also makes it challenging to interpret findings in which not all parameters are congruently changed. Especially in individuals with tauopathy in the absence of amyloidopathy, i.e. A–T+N±, the relation to Alzheimer’s disease is unclear. The elevation of CSF pTau181 has been described as specific for Alzheimer’s disease, at least when compared to Lewy body, frontotemporal, and vascular dementia.^4,6,15 Furthermore, different reports indicated that elevated pTau181 in blood might be a cost-efficient Alzheimer’s disease-specific marker for cognitive decline and neurodegeneration.^15–17 However, neurofibrillary tangles have been described in post-mortem brain without the presence of neuritic amyloid plaques, which led to the hypothesis that tauopathy in the absence of amyloidopathy indicates a pathological condition other than Alzheimer’s disease, a suspected non-Alzheimer's disease pathophysiology (SNAP).^18,19 SNAP has been reported to be accompanied with mesial temporal lobe atrophy and memory deficits similar to Alzheimer’s disease but was associated with a slower cognitive decline.¹⁸ In the living, SNAP is suspected not to be associated with Alzheimer’s disease due to the lack of amyloid pathology in terms of decrease of Aβ₄₂ or Aβ₄₂/Aβ₄₀ ratio or positivity in amyloid PET.²⁰ The role of pTau181 in SNAP is unclear, since the original definition did not differentiate between pTau and tTau. In this study we planned to investigate whether an increase in pTau181 in CSF without evidence of decreased Aβ₄₂ levels or Aβ₄₂/Aβ₄₀ ratio in non-demented individuals is compatible with the described properties of SNAP. We aimed to compare (i) the relative frequencies of the A+T+N±, A+T–N±, A–T+N± and A–T–N– groups to ascertain whether the composition of our cohort differs from that of other studies; (ii) the conversion rate to dementia; (iii) the domain specific cognitive decline between the groups; and (iv) the annualized atrophy rates of the hippocampi and mesial temporal lobes.

Materials and methods

Study population

The cohort was recruited at the memory department of the Clinic for Psychiatry and Psychotherapy located at the FAU Erlangen-Nuremberg during the period from April 2010 until November 2021. Participants of the Erlangen cohort were required to have a lumbar puncture with a complete dataset available for Aβ₄₂, the ratio between Aβ₄₂ and Aβ₄₀ (Aβ₄₂/Aβ₄₀), pTau181, and tTau in CSF, a structural brain scan and a neuropsychological assessment within 6 months. The severity of concomitant depressive symptoms was monitored by the Beck’s depression inventory II (BDI-II) or the Geriatric Depression Scale (GDS). The donors of the samples were enrolled in the study during the initial diagnosis, which is why none of them received treatment with acetylcholinesterase inhibitors or memantine at the time of sample acquisition. In a multidisciplinary board, the cases were classified based on clinical and psychometric testing and imaging. Exclusion criteria were a history of stroke within the last 6 months, hallucinations, inflammatory disease of the CNS or age younger than 50 years. Classification into SCI, MCI, and dementia was not based solely on scores on the Mini-Mental State Examination (MMSE). Instead, the complete neuropsychological testing was considered, especially in the borderline areas of the MMSE. Total Tau is a non-specific marker of neurodegeneration and neuronal injury. In the present study, this marker was only useful to reliably define the controls (A–T–N– compared to A–T–N+). Once the Aβ₄₂/Aβ₄₀ ratio, Aβ₄₂ or pTau181 showed pathological levels, no further differentiation was necessary using tTau levels since it does not distinguish whether the neuronal injury is due to Alzheimer’s disease or another possible comorbid condition.³ For the longitudinal analysis, individuals were required to have at least one follow-up visit more than 12 months from baseline. Participants with focal brain lesions or defects on MRI or substantial T₂-weighted white matter hyperintensities, i.e. Fazekas 2 or 3, were excluded from the region of interest (ROI) analysis of structural MRI data.²¹ At each visit, participants were encouraged to make an appointment for the next visit and telephone contacts were initiated at approximately 48-month intervals to schedule appointments. The range of the follow-up period was 12–144 months. The median (Mdn) follow-up for all individuals was 1.0 year, the first quartile (Q1) was 0.0 years and the third quartile (Q3) was 4.7 years. The main reasons for known non-participation and dropout surveyed were ‘too much of a burden’ (52%), ‘no one to accompany them’ (6.5%), ‘too ill’ (3.2%) and death (3.2%). The study protocol was approved by the clinical ethics committee of the University of Erlangen-Nuremberg. Patients and their authorized legal representatives provided written informed consent after receiving a complete description of the study.

Neuropsychological assessment

All participants with memory complaints of the Erlangen cohort were assessed with the German version of the CERAD neuropsychological battery plus (CERAD-NB+).²² We excluded the Boston Naming Test and Constructional Praxis from the analyses of cognitive decline, as the majority of the A–T–N– participants achieved maximum scores and we therefore deemed them too easy. The proportion of missing tests varied between 6.8% MMSE and 33.6% of phonemic fluency. A total of 643 observations were available from 285 subjects, of whom 154 had a follow-up period of more than 12 months.

MRI acquisition and brain volumetry

MRI of the Erlangen cohort were obtained with a 3 T MR scanner (Magnetom Tim Trio, Siemens Healthineers AG). For the longitudinal voxel-based morphometric (VBM) analyses, only individuals who received at least one T₁-weightened MRI at baseline and 12 months of follow-up were included. We applied the VBM-workflow for longitudinal data optimized for large changes of the Computational Anatomic Toolbox (CAT12 v. 12.8; Jena University Hospital, Jena, Germany; http://dbm.neuro.uni-jena.de/cat/; http://www.neuro.uni-jena.de/cat12/CAT12-Manual.pdf). Details on the image acquisition and VBM-workflow are provided in the Supplementary material.

CSF ELISA

The concentrations of Aβ₄₀ and Aβ₄₂ in the CSF were assessed either with commercially available ELISAs from IBL International or in the case of Aβ₄₂, also from Fujirebio Europe (Innotest® formerly Innogenetics). tTau and pTau181 in CSF were measured with two ELISAs from Fujirebio Europe/Innogenetics (Innotest®). Cut-off values of the ELISAs are given in Supplementary Table 1. Cut-off value of the Aβ₄₂/Aβ₄₀ ratio was 0.05. The cut-offs were calculated by maximization of the Youden Index, i.e. at the data-points where the sum of the sensitivity and the specificity of the given assay are the largest. The details for Aβ₄₂ and the Aβ₄₂/Aβ₄₀ ratio are given elsewhere.²³ The same approach was taken for pTau181 and tTau. For pTau181 and tTau, the coefficients of variation were in the range of 5%. In this study, the assessment of the presence of neurodegeneration (N) was solely based on tTau levels in CSF, as a uniform standard for setting the cut-off in structural MRI and FDG-PET has not yet been established and we wanted to avoid circularity as the regional brain volume was measured as a dependent variable.

Statistics

The descriptive statistics of the psychometric tests in this study are presented using age-, sex-, and education-adjusted z-scores.^22,24

Shapiro–Wilk’s test was used to test for normality. Homogeneity of variance was assessed by Levene’s test. Differences between the selected ATN groups were assessed with Pearson’s χ² test for categorial variables (sex, Aβ₄₂-ELISA#2) or with the Kruskal–Wallis H-test followed by Dunn’s multiple comparison test in case a significant effect was observed for ordinal or non-normally distributed data (MMSE, BDI-II, GDS, Fazekas). For continuous variables, we used the t-test, analysis of variance (ANOVA) or the Brown–Forsythe test for data with inhomogeneous variances followed Tukey B or Dunn Bonferroni correction in case a significant effect was observed (e.g. age in years, education in years, CERAD-NB+ z-scores, pTau-, tTau-, Aβ₄₂-, Aβ₄₀-levels (pg/ml), Aβ₄₂/Aβ₄₀ ratio).

The annualized per cent change (ΔROI) in hippocampal and mesial temporal lobe volume (V_i in cm³) based on the MRI scans at different time points (t_i in months) was computed as follows:

$$\mathit{\Delta }ROI=[(V_{i\mathit{T}1}-V_{i\mathit{T}0})\times 1200]/[V_{i\mathit{T}0}\times (t_1-t_0)]$$ (1)

We used a one-way analysis of covariance (ANCOVA) to identify main effects of the selected ATN groups on the ROI volume and annualized atrophy rates while controlling for total intracranial volume (TIV), age, time of follow-up and education.

Linear mixed-effect models (LMEM) were used to assess whether the change in individual CERAD-NB+ subtest scores over time differed between the A+T+N±, the A–T+N± and the A+T–N ± groups compared with the A–T–N– group as reference. The dummy coded groups and disease stage, i.e. SCI versus MCI, time from baseline in years, the interaction term between time from baseline and groups, and the interaction term between time from baseline and disease stage were included as fixed effects. The z-scores of the individual tests of the CERAD-NB+ were used as dependent variables. Models were specified with a random intercept and slope at the subject level with an unstructured covariance matrix.

We performed Cox proportional hazard analyses corrected for age at baseline, MMSE, years of education, sex and disease stage stratified by the A+T+N±, the A–T+N±, the A+T–N± and the A–T–N– groups to assess the effect of these categories on progression to dementia. Survival was defined as the period between entry into the study and progression to dementia, which was clinically defined as the occurrence of impairment of daily life activities and an MMSE score of ≤24. The proportional hazards assumption was evaluated using Schoenfeld residuals.

In the case of the analysis of the trajectories of psychometric tests, missing data were replaced by multiple imputation, as implemented in SPSS 28.0, unless the absence was due to participant withdrawal. Otherwise, missing data were handled using the inherent maximum likelihood estimation of the Cox proportional hazard models and the linear-mixed effect models.

Data analysis was performed using the SPSS statistical package (version 28.0; SPSS, Chicago, IL, USA).

Data availability

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to restrictions e.g. their containing information that could compromise the privacy of research participants.

Results

Two hundred and eighty-five non-demented participants from the Erlangen memory clinic were enrolled in this study. A flow chart of the study selection is shown in Supplementary Fig. 1. At baseline, the subjects’ mean age was 66.5 ± 9.1 years and median MMSE score was 28 with a range from 23 to 30. The cohort comprised 177 (62.1%) males and 108 (37.9%) females. The relative frequencies of the different ATN profiles of all individuals of the Erlangen cohort is given in Fig. 1. The largest group was the biomarker negative control group with 104 (36.5%) individuals, followed by 87 (30.5%) in the A+T+N+ group, 39 (13.7%) in the A–T+N+ group and 26 (9.1%) in the A–T+N– group. The relative frequencies of the other biomarker groups were each below 5%. The A+T+N ± group was significantly older than the A–T+N ± group [P = 0.016, M_Diff = 4.485, 95% confidence interval (CI) (0.51, 8.46)] and the A–T–N– group (P = 0.001, M_Diff = 8.354, 95% CI (4.87, 11.84)], respectively (Table 1). There were no significant differences in sex or education between the A+T+N±, the A–T+N±, the A+T–N± and the A–T–N– groups, which were selected for the follow-up analyses. The baseline demographics and normalized subtest results of the CERAD-NB+ are shown in detail in Table 1 and Supplementary Table2, respectively. The median follow-up time for participants with at least one visit at 12 months was shortest for the A+T+N ± group, at 40 months (Q1, Q3 = 25, 71 months), with no significant difference between the remaining groups: 57 (40, 78) months in the A–T–N+, 60 (39, 89) months in the A+T–N±, 71 (39, 93) months in the A–T–N– and 85 (41, 101) months in the A–T+N ± group. Baseline characteristics of those who were not followed up after more than 12 months and those who were included in the longitudinal analyses are provided in Supplementary Table 3.

Figure 1.

Relative proportions of CSF-biomarker profiles. The pie chart illustrates the distribution of CSF-biomarker profiles, i.e. the ratio of Aβ₄₂/Aβ₄₀ (R), Aβ₄₂ (Aβ₄₂), pTau181 (T) and total Tau (N) levels, of non-demented individuals (n = 285). A– = normal Aβ₄₂/Aβ₄₀ and Aβ₄₂ levels; A+ = pathological Aβ₄₂/Aβ₄₀ or Aβ₄₂ levels.

Table 1.

Demographic and clinical characteristics of non-demented individuals grouped by biomarker profiles based on the ATN classification

	A–T–N–		A+T–N±		A+T+N±		A–T+N±		A–T–N+
	n	%	n	%	n	%	n	%	n	%	P-value (female)
Total (female)	104 (37)	37 (34)	16 (5)	6 (5)	88 (37)	31 (34)	65 (25)	23 (23)	12 (4)	4 (4)	0.862
MCI (female)	70 (29)	34 (34)	9 (4)	4 (5)	74 (33)	35 (39)	47 (16)	23 (19)	9 (3)	4 (4)	0.809
SCI (female)	34 (8)	45 (35)	7 (1)	9 (4)	14 (4)	18 (17)	18 (9)	24 (39)	3 (1)	4 (4)	0.294
Follow-up (female)	56 (25)	36 (40)	11 (4)	7 (7)	51 (23)	33 (38)	30 (6)	20 (10)	6 (3)	4 (5)	0.173
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	P-value
Age (years)	63	8	66	9	71	8	67	9	65	9	<0.001
Education (years)	14	3	14	3	14	3	14	3	13	2	0.572
MMSE	28	2	28	2	27	2	28	2	28	2	<0.001
BDI-II	15	9	11	5	10	9	17	14			0.138
GDS	4	3	2	2	3	3	4	4	3	3	0.680
Aβ42 (pg/ml)a	1010	240	674	348	637	156	1250	445	1030	281	<0.001
Aβ42 (pg/ml)b	1336	420	578	18	729	201	1693	354	1422	262	<0.001
Aβ42/40 ratio	0.08	0.02	0.05	0.02	0.04	0.02	0.09	0.07	0.07	0.03	<0.001
pTau (pg/ml)a	38	7	17	51	99	35	63	35	44	5	<0.001
pTau (pg/ml)b	38	12	34	10	116	40	74	13	55	2	<0.001
Total Tau (pg/ml)a	199	53	238	126	766	473	399	194	433	139	<0.001
Total Tau (pg/ml)b	207	59	226	117	619	231	340	91	356	34	<0.001
GM (ml)	575	73	631	37	530	71	550	139	484	88	0.154

GM = grey matter volume.

^a ELISA#1 for Aβ42, pTau, total Tau.

^b ELISA#2 for Aβ42, pTau, total Tau.

Table 2.

Linear mixed model estimates for the comparison of individual CERAD-NB+ subtest z-scores between the selected ATN groups compared with the A–T–N– group as reference

Outcome variable		Parameter	Estimate	P -value	95% CI
MMSE	ΔBaselinea	A–T+N±	–0.1	0.618	–0.7	0.4
		A+T+N±	–0.6	0.007	–1.1	–0.2
		A+T–N±	–0.2	0.543	–1.0	0.5
	ΔSlopeb	A–T+N±	0.1	0.491	–0.1	0.3
		A+T+N±	–0.4	<0.001	–0.6	–0.2
		A+T–N±	–0.4	0.014	–0.7	–0.1
Word List Learning	ΔBaseline	A–T+N±	0.3	0.510	–0.5	1.0
		A+T+N±	–0.8	0.022	–1.4	–0.1
		A+T–N±	–0.2	0.646	–1.2	0.7
	ΔSlope	A–T+N±	0.1	0.640	–0.2	0.3
		A+T+N±	–0.2	0.044	–0.4	0.0
		A+T–N±	–0.3	0.106	–0.6	0.1
Word List Delayed Recall	ΔBaseline	A–T+N±	0.2	0.536	–0.5	1.0
		A+T+N±	–0.7	0.013	–1.3	–0.2
		A+T–N±	–0.7	0.144	–1.6	0.2
	ΔSlope	A–T+N±	0.0	0.782	–0.2	0.2
		A+T+N±	–0.2	0.042	–0.4	0.0
		A+T–N±	–0.1	0.435	–0.4	0.2
Constructional Praxis Delayed Recall	ΔBaseline	A–T+N±	–0.5	0.171	–1.3	0.2
		A+T+ N±	–0.9	0.009	–1.6	–0.2
		A+T–N±	–0.6	0.305	–1.8	0.6
	ΔSlope	A–T+N±	0.1	0.330	–0.1	0.2
		A+T+N±	–0.2	0.077	–0.3	0.0
		A+T–N±	–0.2	0.348	–0.5	0.2
TMT-Part A	ΔBaseline	A–T+N±	0.1	0.804	–0.7	0.9
		A+T+N±	–0.2	0.604	–0.9	0.5
		A+T–N±	0.5	0.423	–0.7	1.6
	ΔSlope	A–T+N±	–0.1	0.511	–0.2	0.1
		A+T+N±	–0.2	0.029	–0.4	0.0
		A+T–N±	–0.1	0.381	–0.4	0.2
TMT-Part B	ΔBaseline	A–T+N±	0.2	0.530	–0.5	0.9
		A+T+N±	–0.5	0.081	–1.1	0.1
		A+T–N±	0.8	0.100	–0.2	1.8
	ΔSlope	A–T+N±	0.0	0.909	–0.1	0.1
		A+T+N±	0.0	0.471	–0.2	0.1
		A+T–N±	–0.1	0.454	–0.3	0.1
Semantic fluency (animals)	ΔBaseline	A–T+N±	0.3	0.324	–0.3	0.9
		A+T+N±	0.0	0.925	–0.5	0.6
		A+T–N±	0.0	0.986	–0.9	0.9
	ΔSlope	A–T+N±	–0.1	0.261	–0.2	0.1
		A+T+N±	–0.2	0.016	–0.4	0.0
		A+T–N±	–0.1	0.507	–0.3	0.1
Phonemic fluency (s-words)	ΔBaseline	A–T+N±	0.1	0.692	–0.5	0.7
		A+T+N±	0.3	0.237	–0.2	0.7
		A+T–N±	0.2	0.644	–0.6	0.9
	ΔSlope	A–T+N±	0.0	0.801	–0.1	0.1
		A+T+N±	–0.1	0.023	–0.2	0.0
		A+T–N±	–0.1	0.656	–0.3	0.2

CERAD-NB+ = The Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) neuropsychological battery plus (NB+); TMT = Trail Making Test. Bold font indicates statistical significance (P < 0.05).

^a ΔBaseline = Difference at baseline compared to A–T–N–.

^b ΔSlope = Difference in longitudinal slope compared to A–T–N–.

Since the CSF-levels of Aβ₄₂ and the Aβ₄₂/Aβ₄₀ ratio were used for the classification into A+ and A–, it is not surprising that these were lower in the A+T+N ± and A+T–N ± groups compared to the A–T+N ± and the A–T–N– groups as assessed with two individual ELISAs (Table 1). In addition, the A–T + N ± group showed significantly higher Aβ₄₂ levels compared to the A–T–N– group, P_ELISA#1 = 0.008, M_Diff = 254.769, 95% CI (60.87, 448.67) and X²_Elisa#2 (2, 70), P_Elisa#2 = 0.006 (Fig. 2A). Concomitantly, the Aβ₄₀ levels were significantly increased in the A–T+N ± groups [P < 0.001, M_Diff = 5517.820, 95% CI (3345.36, 7690.28)] and the A+T+N ± groups [P < 0.001, M_Diff = 5771.363, 95% CI (3773.82, 7768.91)] compared to the A–T–N– group (Fig. 2B and Supplementary Fig. 2). As the Aβ₄₀ levels of the A– groups were strongly positively associated with the Aβ₄₂ levels (Fig. 2C), the Aβ₄₂/Aβ₄₀ ratio showed no significant difference between the A–T+N ± and A–T–N– groups, t(159) = 0.361, P = 0.718. The levels of pTau181 were strongly associated with the Aβ₄₀ levels in both the A– and the A+ groups (Fig. 2D), whereas age only had a very weak association with the Aβ₄₀ level, r(275)= 0.163, P = 0.007. Scatter plots for the biomarker pairs Aβ₄₂/Aβ₄₀ ratio versus pTau181, Aβ₄₂/Aβ₄₀ ratio versus tTau, and pTau181 versus tTau are shown in Supplementary Fig. 2. To exclude that a concomitant disease such as vascular dementia, cerebral amyloid angiopathy and frontotemporal dementia may exert an influence on the comparison of Aβ₄₀ levels between the A–T–N– and the A–T+N ± group, we repeated the analysis in a subgroup of 23 patients with only subjective cognitive deficit, no deterioration in neuropsychological testing over at least 2 years and no abnormalities on MRI except for mild microangiopathy. This subgroup also showed a significant increase of Aβ₄₀-levels in the A–T+N ± group, M = 18891 pg/ml, standard deviation (SD) = 4780 pg/ml, compared with the A–T–N– group, M = 14299 pg/ml, SD = 4789 pg/ml, t(21) = −2.281, P = 0.033.

Figure 2.

CSF-Aβ₄₂ and CSF-Aβ₄₀ levels were positively associated with pTau181 in the A– groups. CSF-Aβ₄₂ levels were significantly higher in the A–T+N± group of non-demented individuals compared to the CSF-biomarker negative control group (A–T–N–) at baseline as assessed with two different ELISAs, P_ELISA#1 = 0.008 and P_ELISA#2 = 0.006 (A). CSF-Aβ₄₀ concentrations were significantly increased in the A–T+N± (P < 0.001), the A–T–N+ (P = 0.05) and the A+T+N± (P < 0.001) group compared to the A–T–N– group (B). In the A– group, the Aβ₄₂ and Aβ₄₀ levels were positively associated, r_A–T–N– (29) = 0.801, P < 0.001 and r_A–T+N± (47) = 0.861, P < 0.001 (C). The pTau181 levels were positively associated with the levels of Aβ₄₀ in the A– and the A+ groups, r_A–|ELISA#1 (92) = 0.726, P < 0.001, r_A–|ELISA#2 (84) = 0.591, P < 0.001, r_A+|ELISA#1 (51) = 0.679, P < 0.001 and r_A+|ELISA#2 (48) = 0.364, P = 0.011 (D). A and B show boxplots with superimposed individual data points; scatter plots are shown in C and D.

No significant difference in the severity of depressive symptoms was found between the A–T+N± group and controls as assessed by BDI-II or GDS (Table 1). The incidence of strokes and the extent of the microangiopathy at baseline did not provide any indication that vascular brain damage was more frequent in the A–T+N± group (Supplementary Table 4).

To assess differences between the selected ATN groups in both brain atrophy at baseline and annualized atrophy rate, One-way ANCOVAs were conducted on hippocampal volumes (H) and mesial temporal lobe volumes (MTL) adjusting for age at baseline, years of education, TIV, and the interval between baseline enrolment and the most recent follow-up. H and MTL at baseline and their annualized atrophy rates (ΔH and ΔMTL, respectively) showed no difference between the A–T+N± and the A–T–N– group in pairwise comparison of adjusted means (Supplementary Fig. 3 and Supplementary Table 5). The A+T+N± group showed a significantly increased ΔH [P = 0.016, M_Diff = 2.222, 95% CI (0.434; 4.009)] and a significantly decreased H at baseline [P = 0.018, M_Diff = –0.519, 95% CI (−0.947, –0.092)] compared with the A–T+N± group. Imaging data from at least two time points 12 months apart were available from 71 individuals with a mean follow-up of 4.3 years. Descriptive information and unadjusted and adjusted means are given in Supplementary Table 5.

The cross-sectional and longitudinal associations of the selected ATN groups with items of the CERAD-NB+ were examined using LMEM with the A–T–N– profile as reference group. The longitudinal trajectories and the performance at baseline of the A–T+N± group showed no significant difference compared to the A–T–N– group on the eight CERAD-NB+ items assessing memory, language, executive function or psychomotor speed (Table 2). The A+T+N± group performed worse on MMSE, Word List Learning and Recall and Constructional Praxis Recall at baseline and had significantly worse longitudinal trajectories for all subtests compared to the A–T–N– group, with the exception of Constructional Praxis Delayed Recall and Trail Making Test Part B (Table 2). The A+T–N± group had significantly worse longitudinal trajectories on MMSE (Table 2). The NIA-AA research framework criteria allow grouping solely based on Aβ₄₂ levels without considering the Aβ₄₂/Aβ₄₀ ratio (R). The results of this additional LMEM-analysis are shown in Supplementary Table 6. In addition, we repeated the analysis excluding 12 individuals, who had borderline values, i.e. ±5% of the ELISA cut-off value, because the dispersion of the measured values in these individuals may lead to incorrect group assignment (Supplementary Table 7). The LMEM analyses indicated only marginally changed trajectories after the exclusion of these individuals (Supplementary Table 7).

Over the course of follow-up, 38 of the 154 non-demented individuals (25%) with follow-up data were diagnosed with onset dementia. We used Kaplan-Meier curves for a preliminary analysis of risk of dementia between the selected ATN groups (Fig. 3A). The A+T+N± group had a significantly higher risk of dementia compared to the risk of the A–T–N– group in univariate and multivariate Cox regression models adjusted for age, years of education, sex, disease stage and baseline MMSE-Score, P = 0.001, adjusted hazard ratio (HR) 6.6 (95% CI: 2.1, 20.7) (Fig. 3B). Non-demented individuals from the A+T–N± and the A–T+N± showed no significantly different onset of dementia compared to the A–T–N– group. The A–T+N± had the lowest hazard ratio of all groups, adjusted HR 0.3 (95% CI 0.0, 1.6). However, when the Aβ₄₂/Aβ₄₀ ratio was not considered in the grouping, the Aβ₄₂–T+N± group showed a significantly higher dementia risk than the Aβ₄₂–T–N– group (Supplementary Fig. 4). In contrast, the exclusion of individuals with borderline values in the ELISAs had no significant influence on the risk of dementia (Supplementary Fig. 5). The overall model and individual factors adhered to the proportional hazard’s assumption. A total of seven A–T+N± individuals developed dementia during follow-up, of which three were vascular in origin, one was Alzheimer’s disease, one was mixed type Alzheimer’s disease and two were due to frontotemporal dementia.

Figure 3.

Non-demented A–T+N± individuals showed no increased risk of dementia onset compared to controls. The Kaplan Meier curves show the risk of onset of dementia in non-demented subjects by group (A). Comparison of hazard ratios of dementia onset between groups A–T+N±, A+T–N± and A+T+N± versus group A–T–N– in a univariate and a multivariate Cox regression model with age, education, sex, disease stage (SCI versus MCI) and MMSE at baseline as covariates (B).

Discussion

In this study, 23% of the cohort were non-demented individuals with CSF-pTau181 elevation without concomitant amyloidopathy, who did not differ from those of the biomarker negative reference group in terms of their trajectories of cognitive decline or annualized hippocampal atrophy rate and showed the lowest hazard ratio in the multivariate Cox regression model during the follow-up. The relative frequencies of the A–T+N± and A–T–N+ group in this study were in accordance with previous reports about the ATN classification, in which the percentage of non-demented subjects with SNAP ranged from 10 to 34%.^8,9,11–14 The relative frequency of the A+T+N± group (31%) was above the reported range from 1.6 to 23% in previous studies.^8,9,11–14 This could be due to the fact that this study did not only rely on the Aβ₄₂ levels but also on the Aβ₄₂/Aβ₄₀ ratio to detect amyloidopathy. Fifty-six per cent of the A+ individuals showed a decreased Aβ₄₂/Aβ₄₀ ratio in combination with a normal Aβ₄₂ level. A change in the Aβ₄₂/Aβ₄₀ ratio has previously been reported to precede both the amyloid positivity in PET and the decrease of Aβ₄₂ levels in Alzheimer’s disease, which is why we consider the determination of Aβ₄₂/Aβ₄₀ to be essential for the identification of individuals with SNAP.^25–27 Regarding the surrogate marker hippocampal atrophy for neurodegeneration, this study showed a significantly lower annualized atrophy rate only for the A+T+N± group but not for the A–T+N± group compared to the A–T–N– control group. An absence of longitudinal hippocampal atrophy in SNAP has previously been reported by Burnham et al.^20,26 In contrast to our findings, which did not indicate a significant difference in hippocampal or MTL volume at baseline for the A–T+N± group, MTL atrophy has previously been reported to be a non-specific feature of SNAP, as the atrophy patterns show spatial overlap with Alzheimer’s disease and other neurodegenerative disorders like hippocampal sclerosis, limbic-predominant age-related TDP-43 encephalopathy (LATE), and argyrophilic grain disease.²⁰ The use of different detection methods, in this case CSF-total Tau instead of FDG-hypometabolism or atrophy in MRI, could affect the composition of the cohort. Especially in studies which only determined Aβ₄₂ levels and not the Aβ₄₂/Aβ₄₀ ratio, individuals with Alzheimer’s disease could have been falsely assigned to the SNAP group, which might lead to an overestimation of brain atrophy or cognitive decline in this group. In addition to the different measurement methods, the heterogeneity in the relative frequencies of each ATN group between studies could also be due to different proportions of patients with SCI or MCI. This study cohort was largely composed of patients with MCI and had a higher proportion of A+T+N± and lower of A+T–N± compared with the studies that consisted solely or largely of SCI patients.^8,11,28

The survival models suggested that A+ individuals progressed more frequently to dementia than A– individuals in this study, which is in accordance with previous reports.^26,29 The performance at baseline and the cognitive trajectories of the A–T+N± and the A–T–N– group did not show a significant difference between memory, language, praxis and executive function as assessed with items of the CERAD-NB+. The A+T+N± group showed a greater decline than the A–T–N– group in all items of the CERAD-NB+ examined; the A+T–N± group showed a worse trajectory only on the MMSE compared to the A–T+N± and the A–T–N– groups. The findings that A+ individuals show steeper cognitive decline and faster conversion than A–T+N± group to dementia align with previous reports comparing preclinical Alzheimer’s disease and preclinical Alzheimer’s pathology with SNAP.^25–27 Similar to our findings for the A–T+N± group, previous reports indicated that SNAP does not appear to have a distinct cognitive profile from Alzheimer’s disease.²⁰

In this study, the psychometric trajectories from items of the CERAD-NB+, the hippocampus and mesial temporal lobes volumes and the hazard ratios of dementia did not indicate that there were more individuals suffering from a dementing disease in the A–T+N± group than in the A–T–N– group. However, the cross-sectional comparison showed that the elevation of pTau181 in the A–T+N± group was not independent from Aβ, as both Aβ₄₀ and Aβ₄₂ were congruently elevated compared to the A–T–N– group. Whether these changes indicate a pathological or a non-pathological condition is unclear. At least with respect to Alzheimer’s disease, previous reports showed that normalizing Aβ₄₂ or pTau181 to Aβ₄₀ improved their performance as biomarkers for Alzheimer’s disease.^25,27,30,31 Therefore, a general increase in Aβ and Tau variants could be due to an increased production rate rather than to pathological process. Alternatively, this stage might represent an early stage of Alzheimer’s disease with presumably slower disease progression, years before interferences with a person’s daily life and activities occur. As described in the Aβ cascade hypothesis, the aggregation of Aβ can be promoted by the increased production of Aβ, insufficient clearance of Aβ, or a shift in production towards more amyloidogenic Aβ peptide variants. In a model in which increased production or reduced clearance of all Aβ are the cause of Alzheimer’s disease, an increase of both Aβ₄₂ and Aβ₄₀ should be observed first, before the more amyloidogenic Aβ₄₂ is sequestered by the developing Aβ plaques. Consequently, higher Aβ₄₀ levels would persist longer, which could explain a better sensitivity of the Aβ₄₂/Aβ₄₀ ratio compared to the measurement of Aβ₄₂ alone or amyloid PET. In line with this model we detected higher CSF-Aβ₄₀ levels in the A+T+N± and the A+T–N± groups compared to the A–T–N– group, which aligns with previous reports of elevated Aβ₄₀ levels being associated with Alzheimer’s disease.^32,33 Given the high proportion of MCI in this study and the consistent proportion of A–T+N± in both SCI and MCI, it seems unlikely that this is merely a very early stage of Alzheimer’s disease with no differences in disease progression compared with the canonical Alzheimer’s pathological cascade.³⁴ Longitudinal studies of CSF biomarker concentrations would be useful for clarification but have been hampered by short follow-up time, lack of Aβ₄₂/Aβ₄₀ ratio detection, and/or small sample size.^35,36 Detection of the corresponding biomarkers in blood might be a cost-effective alternative for this purpose in future studies. Furthermore, the increase of Aβ₄₂ and Aβ₄₀ in A–T+N± group might indicate a pathological process apart from Alzheimer’s disease, as previous reports indicated a variety of other diseases like FTD and non-neurodegenerative disease like cerebrovascular disease, multiple sclerosis and encephalitis to be associated with changes in Aβ levels.^37–39 The identification of factors associated with the onset of Alzheimer’s disease could help to determine whether the A–T+N± CSF-biomarker profile is part of the spectrum of Alzheimer’s disease.

There are several limitations to the study. The study population was not chosen randomly from the community and was generally well educated, which precludes extrapolation of study results to the general population. Furthermore, the use of cut-off values, which is inherent with the biomarker based ATN classification, results in a loss of information and can limit the comparability with other cohorts, whereas in this study, exclusion of individuals with borderline values did not significantly alter the psychometric trajectories. Third, the CERAD-NB+ is standardized for the detection of Alzheimer’s disease. Many symptoms of Alzheimer’s disease overlap with those of other neurodegenerative disease, however it cannot be excluded that early stages of other neurodegenerative diseases were not adequately detected. When interpreting the results, it should be noted that the follow-up intervals varied between subjects. In this study, the number of A+T–N± and A–T–N+ individuals was too small to draw conclusions about the psychometric trajectories of these groups. The strong association between tTau and pTau181 in this study implies that determination of N using tTau levels is unlikely to provide novel aspects of neuronal injury. The assessment of neurodegeneration biomarkers such as neurofilament light chain protein, hippocampal atrophy, or FDG-PET hypometabolism for grouping, as well as the combination of CSF biomarkers and imaging, could provide more comprehensive insight into N and its prognostic value. Finally, in the estimation of annualized brain loss between groups, it is not possible to identify relevant factors influencing within-group effects.

Using CSF-pTau181, total Tau, Aβ₄₂ and ratio of Aβ₄₂/Aβ₄₀ as parameters for tauopathy, neurodegeneration and amyloidopathy, we found no indication in our long-term study that the A–T+N± individuals suffered from a dementing disorder. However, the positive association between pTau181 and Aβ in this group might help to elucidate the factors necessary for the onset of Aβ deposition. We suggest to refer to individuals in the SNAP group with increased pTau and Aβ and mild cognitive decline as pTau and Aβ surge with subtle deterioration (PASSED).

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Competing interests

P.L. received consultation and/or lecture honoraria from IBL International, Fujirebio Europe, AJ Roboscreen, Biogen and Roche. We acknowledge support from Deutsche Forschungsgemeinschaft (DFG) and Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) within the funding program Open Access Publishing. The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Supplementary material

Supplementary material is available at Brain online.