Tau, Not Amyloid Plaques, Tracks Synaptic Damage in Alzheimer’s Disease

CSF synaptic proteins rise sharply when tau pathology appears, but amyloid plaques alone leave them nearly untouched, according to a two-cohort study published in Annals of Clinical and Translational Neurology. The finding, drawn from 109 patients at the Geneva Memory Center and replicated in 787 participants of the Alzheimer’s Disease Neuroimaging Initiative, suggests that the strongest biological correlate of cognitive decline in Alzheimer’s disease tracks tau-related processes far more closely than amyloid burden itself.

Alzheimer’s disease is neuropathologically defined by extracellular amyloid-β plaques and intracellular neurofibrillary tangles built from hyperphosphorylated tau. Both begin accumulating years or even decades before symptoms appear. Yet the burden of plaques and tangles shows only partial correlations with cognitive impairment, while synapse loss is consistently the strongest correlate in both in vivo imaging and neuropathological studies. That gap has left researchers with an unresolved question: which of the two hallmark proteins actually drives synaptic injury in living patients?

To answer it, the Geneva team turned to three cerebrospinal fluid markers of synaptic integrity. Neurogranin is a postsynaptic protein involved in synaptic plasticity that reflects dendritic spine pathology. SNAP-25 is a presynaptic protein essential for vesicle fusion and neurotransmitter release. GAP-43, highly expressed during neuronal growth and synaptogenesis, is enriched in brain regions most vulnerable to Alzheimer’s pathology. Because presynaptic and postsynaptic compartments show distinct vulnerability in the disease, measuring proteins from both sides of the synapse captures complementary aspects of synaptic damage.

Participants were classified into biological groups based on amyloid and tau status, prioritizing PET imaging over CSF measures when both were available. The Geneva cohort comprised 52 amyloid-negative, tau-negative individuals; 30 who were amyloid-positive but tau-negative; and 27 positive for both proteins. No amyloid-negative, tau-positive individuals were identified in Geneva, though such a group appeared in the ADNI validation sample. Cognitive stages ranged from cognitively unimpaired volunteers and people with subjective cognitive decline to patients with mild cognitive impairment and dementia.

The group comparisons delivered a strikingly consistent message. Neurogranin levels were essentially identical in the amyloid-positive, tau-negative group (144 ± 59 pg/mL) and the double-negative group (153 ± 59 pg/mL), but jumped to 248 ± 98 pg/mL once tau pathology was present. SNAP-25 followed the same pattern, and GAP-43 rose from roughly 2900 to 3100 pg/mL in the tau-negative groups to 4750 ± 1751 pg/mL in the amyloid-positive, tau-positive group. All three differences were highly significant. In ADNI, where an amyloid-negative, tau-positive group was available, GAP-43 concentrations in that group (7536 ± 2784 pg/mL) matched those of the double-positive group (7307 ± 3153 pg/mL), while both tau-negative groups sat near 4000 pg/mL.

Correlation analyses reinforced the picture. Synaptic biomarkers showed no significant relationship with CSF Aβ42 in either cohort, only weak-to-moderate associations with amyloid PET burden, and strong correlations with CSF p-tau181 — ρ = 0.83 for neurogranin, 0.80 for SNAP-25, and 0.78 for GAP-43 in Geneva, and ρ = 0.76 for GAP-43 in ADNI. Tau PET signal correlated moderately with neurogranin and GAP-43 but not significantly with SNAP-25, and the tau PET subgroup was small, with only 48 Geneva participants scanned.

Multiple linear regression then disentangled the independent contributions of the two proteins. When CSF Aβ42 and CSF p-tau181 were entered simultaneously as predictors of GAP-43, both showed positive associations, but the tau coefficient was far larger (β = 0.94 versus β = 0.31), and the model accounted for 81% of the variance. When amyloid PET Centiloid values replaced CSF Aβ42, the direction flipped: higher amyloid PET burden was associated with lower GAP-43 (β = −0.31), while p-tau181 remained strongly positive (β = 1.02). Because lower CSF Aβ42 and higher amyloid PET both index greater amyloid pathology, the two results are directionally concordant — greater amyloid burden, once tau is accounted for, accompanies lower synaptic protein concentrations. Comparable models for neurogranin and SNAP-25 produced the same pattern.

Mediation analysis added a cross-sectional clue about sequence. The direct pathway from amyloid PET burden to GAP-43 was negative, while the indirect pathway through p-tau181 was positive and larger, producing an overall positive association — a statistical suppression pattern replicated in ADNI. The authors interpret this as consistent with a model in which amyloid-β contributes to synaptic alterations largely through its relationship with tau rather than through a direct effect on synapses. They also caution that the inverse amyloid PET association may reflect a stage at which soluble, synaptotoxic Aβ species have been sequestered into less biologically active aggregates, and that both CSF Aβ42 and amyloid PET tend to plateau once amyloid pathology is established, limiting their ability to capture downstream processes.

The interpretation of tau depends heavily on how it is measured. CSF p-tau181 captures soluble, phosphorylated tau species that rise early in response to amyloid pathology, whereas tau PET reflects aggregated, more advanced pathology. Under the 2024 revised Alzheimer’s Association criteria, soluble phosphorylated tau species are classified as Core 1 biomarkers that become abnormal early, while tau PET is a Core 2 biomarker of aggregated tau. Accordingly, the authors state that CSF p-tau181 in their study, and the tau status derived from it in participants without tau PET, should be regarded as an index of soluble tau-related processes rather than a direct measure of neurofibrillary tangle burden. The strong p-tau181 association may partly reflect a common amyloid-related neuronal response, and the more modest tau PET associations suggest that aggregated tau explains a smaller proportion of the variance in synaptic biomarker levels.

Several limitations temper the conclusions. The cross-sectional design cannot establish temporal ordering, and longitudinal studies would be needed to characterize the sequence of changes. The number of participants with tau PET was modest, some overfitting cannot be excluded in the adjusted PET-subset models, and because p-tau181 and the synaptic proteins are all neuronally derived and released into CSF, part of their strong association could reflect shared neuronal secretion and CSF dynamics rather than a tau-specific effect on synapses. No formal correction for multiple testing was applied, though the authors note that all principal associations with p-tau181 would remain significant after Bonferroni correction, while results with p values between 0.01 and 0.05 should be regarded as exploratory. Information on copathologies such as α-synuclein was unavailable, and PET imaging was available in only a subset of Geneva participants. Even so, the convergence across two independent cohorts, three synaptic markers, and multiple biomarker modalities points in one direction: synaptic biomarkers in Alzheimer’s disease are primarily associated with tau-related processes and may serve as indicators of tau-related neuronal injury, with amyloid-β acting predominantly through tau-related mechanisms.

Subject of Research: The independent associations of amyloid-β and tau pathology with CSF synaptic biomarkers across the Alzheimer's disease continuum

Article Title: The Association of Amyloid‐β and Tau With Synaptic Dysfunction in Alzheimer's Disease

Article References: Roveta, F., Mendes, A. J., Garibotto, V., Boccalini, C., Lathuiliere, A., Rainero, I., Rubino, E., Zetterberg, H., Ashton, N. J., Ribaldi, F., Frisoni, G. B., & for the Alzheimer's Disease Neuroimaging Initiative (2026). The Association of Amyloid‐β and Tau With Synaptic Dysfunction in Alzheimer's Disease. Annals of Clinical and Translational Neurology, Article acn3.70558. https://doi.org/10.1002/acn3.70558

Image Credits: AI Generated

DOI: 10.1002/acn3.70558

Keywords: Alzheimer's disease, tau pathology, amyloid-β, synaptic biomarkers, GAP-43, neurogranin, SNAP-25, CSF p-tau181, amyloid PET, tau PET, Geneva Memory Center, ADNI