Introduction

Alzheimer’s disease (AD) is the most common neurodegenerative disease, which is characterized by the accumulation of extracellular amyloid-β (Aβ) plaques and intracellular tau neurofibrillary tangles (NFTs) in the brain. Synaptic dysfunction may lead to neurodegeneration and dementia1 and is recognized as the strongest neurobiological correlates of cognitive impairment in AD2. Recent development of PET ligands that binds synaptic vesicle glycoprotein 2 A (SV2A) has enabled in vivo visualization of synaptic density loss in AD as previously shown in post-mortem studies3. Emerging SV2A PET studies have consistently demonstrated a highly patterned distribution of synaptic loss in AD4,5, suggesting that synaptic loss may propagate along specific pathways. However, the specific mechanisms linking localized synaptic damages to patterned large-scale alterations throughout the brain in AD remain unclear.

Human brain is a highly structured network with complex connections, which facilitates the efficient interregional interactions and transport of molecules essential for metabolism and function6,7. This intrinsic topological architecture renders interconnected regions vulnerable to disease-related damage in a network-dependent manner8. Supporting this, brain areas within the same networks tend to exhibit shared gene expression profiles9,10, morphological features11, and maturational coupling12, thus leading to a similar condition promoting pathological states. The concept of network-based selective vulnerability suggests that pathology propagates not only through direct pathological protein transmission but also through the shared susceptibility of interconnected regions13. Recent evidence shows that the synaptome network, constructed by a set of molecular and morphological features of synapses, closely resembles the functional network14. Additionally, connectivity-related gene expression profiles have been found to converge on synaptic signaling15. These findings indicate that interconnected regions exhibit similar synaptic properties and synapse-related gene expression, which may contribute to their vulnerability to disease-related synaptic loss. Moreover, in the context of AD, Aβ-induced synaptic damage triggers Ca²⁺ influx and mitochondrial dysfunction16. Due to the mobility of mitochondria in axons, ongoing Ca²⁺ influx leads to progressive mitochondrial damage that extends beyond the initial injury sites, spreading through the axonal connection and causing distal synaptic damage17, as synaptic terminals have high energy demand18. Based on previous evidence, we speculate that the synaptic loss pattern in AD may be constrained by network architecture across interconnected regions, while this pattern has yet to be explicitly elucidated.

Aβ and tau are considered as the major contributors to synaptic loss in AD. Evidence implies that oligomeric Aβ induces excitotoxicity at the pre-synaptic terminal through interactions with various cell membrane receptors19. Detached pathological tau mislocates to synaptic compartments and induces toxicity to the synapses, particularly to synaptic vesicle proteins20,21. Inverse relationships between SV2A PET and amyloid22 as well as tau PET23,24,25 have been reported in hippocampus and medial temporal cortex, reinforcing the impact of AD pathologies on synaptic loss. A recent hypothesis17 integrates previous evidence and proposes that at the early stage of AD, the synaptic damage caused by the initial deposition of Aβ can be efficiently cleared by microglia and restricted to local brain regions26,27. However, as Aβ load increases, axons interact with multiple plaques, resulting in the development of pathological tau at multiple sites. This ultimately leads to more severe and widespread synaptic loss rather than the localized loss surrounding individual plaques28. Furthermore, recent evidence has demonstrated that hyperphosphorylated tau (p-tau) leads to early synaptic damage preceding aggregated fibrillar tau29,30. Therefore, both Aβ and pathological tau may propagate synapse loss to distant brain regions19. However, direct evidence linking the Aβ and tau pathology to the distribution of synaptic loss remains lacking.

Recent advances in combining functional MRI and tau PET techniques have elucidated that AD pathology accumulation follows specific macroscale brain networks31, suggesting that the pathology originates at the vulnerable core regions, and spread in a network-dependent manner. Specifically, interconnected regions share similar tau burden32. Aβ, tau, gray matter atrophy has been found to propagate following the specific connectivity profile33,34,35,36,37. Given the close link between AD pathology and synaptic degeneration, and in particular the ability of tau to propagate along functional networks and directly damage synapses, large-scale patterns of synaptic loss in AD may likewise be constrained by the brain’s intrinsic network architecture in a similar manner as tau. Therefore, in the current study, we leveraged [18 F]SynVesT-1 SV2A-PET from 54 cognitively unimpaired amyloid-negative (CU Aβ − ) controls and 91 amyloid-positive (Aβ + ) participants, together with the normative brain connectome from 58 CU controls without Aβ or tau pathology to address the following open questions: (1) whether the pattern of synaptic loss is shaped by brain network connectome; (2) whether synaptic loss in AD distributes following the connectivity patterns to heterogeneous focal epicenters; and (3) whether the network-constrained distribution of synaptic loss is influenced by Aβ and p-tau.

Here we show a highly patterned spatial distribution of synaptic loss in AD, which is modestly constrained by the brain’s network architecture. Despite inter-individual and subtype-level heterogeneity, the observed patterns of synaptic loss consistently follow the functional connectivity profiles to synaptic loss epicenters. Moreover, plasma p-tau levels significantly modulate the network-based distribution of synaptic degeneration. These findings provide a network-level perspective on synaptic vulnerability in AD and may advance future efforts to predict individual patterns of synaptic vulnerability based on network features.

More at link.