A biphasic astrocytic PTGDS trajectory marks a metabolic vulnerability stage in prodromal Alzheimer’s disease

Background: Alzheimer's disease shows prolonged prodromal stability before accelerating decline, yet molecular markers resolving this heterogeneity are limited. We asked whether astrocytic prostaglandin D2 synthase (PTGDS) marks a statistically resolved metabolic inflection during this transition. Methods: We analysed 1.24 million single nuclei retained after quality control from the Seattle Alzheimer's Disease Brain Cell Atlas middle temporal gyrus dataset (84 donors), ordered along a continuous pseudo-progression score, using donor-level quadratic and segmented regression with donor bootstrap resampling and leave-one-donor-out refitting. Findings were then examined in independent bulk brain proteomics (ROSMAP and Banner cohorts), in cerebrospinal fluid proteomics from the Alzheimer's Disease Neuroimaging Initiative on two orthogonal platforms, in a reversible zebrafish model of mild cognitive impairment, and in an amyloid-beta mouse model. Results: PTGDS followed a biphasic astrocytic trajectory with a statistically resolved donor-level inflection (quadratic term −2.27, p = 0.006; vertex at pseudo-progression score 0.47, bootstrap 95% confidence interval 0.32 to 0.53). The same directional change was reproduced in ROSMAP and Banner brain proteomics (Alzheimer's disease versus control p = 3.4 × 10−3), and the biphasic pattern reconciles previously conflicting cerebrospinal fluid reports as stage-dependent. In cerebrospinal fluid, downstream neurofilament light tracked cognitive decline strongly and lipocalin-2 weakly, whereas PTGDS itself was tissue-restricted and not a stand-alone predictor. Limitations: The atlas provides pseudo-time rather than longitudinal sampling, so causality is not established. Coverage is restricted to one cortical region, donor coverage across the pseudo-progression range is uneven with a single donor in the lowest bin, lipocalin-2 transcript detection was sparse, zebrafish validation covered only the early compensatory phase, and the pharmacological arm was not tested after the inflection. Conclusions: These data position astrocytic PTGDS as a candidate stage marker, not a causal driver, of the compensatory-to-vulnerable shift in prodromal Alzheimer's disease, with potential relevance to stage-stratified patient selection. We propose, but do not establish, that attenuation of PTGDS after the inflection accompanies lipocalin-2-linked inflammation and suppression of nerve growth factor receptor.