Serum lipid patterns tracked with Alzheimer group status in ADNI-1, but the signal was narrow
A cross-sectional analysis from ADNI-1 found a limited but measurable relationship between circulating lipid biology and Alzheimer-related status. Investigators examined 88 serum lipid measures in 175 participants and focused on their associations with clinical-amyloid group classification, cerebrospinal fluid biomarkers, cognition, and sex hormone-binding globulin, or SHBG. The analysis produced one standout lipid: PC ae C38:1, the only measure linked to ordinal group score after false discovery rate correction, with β = 0.231 and q = 0.021.
The finding is not broad, but it is biologically interesting. A phosphatidylcholine signal surviving multiple testing correction suggests that membrane lipid composition, lipid trafficking, or downstream inflammatory metabolism may vary along the Alzheimer-amyloid continuum even when the rest of the lipid panel remains unremarkable.
One phosphatidylcholine carried the corrected group signal
The clinical-amyloid grouping analysis produced a second corrected result at the omnibus level: the categorical group test also met the adjusted threshold, with q = 0.043. In practical terms, that means the aggregate lipid profile differed across the clinical-amyloid categories even though most individual analytes did not hold up under correction. PC ae C38:1 was the lone lipid with a statistically secure relationship to the ordinal group score, making it the clearest blood-based lipid marker in the analysis.
This kind of result matters because Alzheimer pathology rarely behaves like a single-marker disease. Lipids participate in neuronal membrane integrity, myelin structure, synaptic signaling, and amyloid processing. A phosphatidylcholine association does not prove causality, but it does fit a plausible pathophysiologic pattern in which systemic lipid handling mirrors central nervous system vulnerability.
The effect size was modest, and the study design does not support clinical use as a stand-alone biomarker. Still, the fact that one lipid survived correction in a relatively small cohort makes the signal worth tracking in larger, independent samples.
Aβ42 associations changed when the model changed
The most instructive part of the analysis may be the instability of the cerebrospinal fluid Aβ42 results. Before group adjustment, 20 lipid-CSF Aβ42 associations were identified. After adjustment for group status in the original linear models, none remained. When investigators switched to rank-based models with categorical adjustment, two Aβ42 associations emerged again.
This pattern points to model sensitivity rather than a clean, uniform biological effect. In a cross-sectional setting, group assignment and biomarker status can sit on top of each other, making it difficult to separate direct lipid-biomarker relationships from associations that are really being carried by disease stage. Rank-based methods can sometimes recover associations that linear models miss when distributions are skewed or outliers exert disproportionate influence. Here, the change in method altered the answer.
The bedside implication is straightforward: apparent lipid-Aβ42 links should be interpreted as provisional until they replicate across analytic frameworks. The biomarker relationship is not robust enough to be treated as settled.
SHBG was the most connected hormone-related variable
Sex hormone-binding globulin showed a broader relationship with the serum lipid panel than the Alzheimer grouping variables did. Under conventional inference, SHBG was associated with 35 lipids; under HC3 inference, 32 lipids remained associated. The persistence across both inferential approaches suggests that the SHBG-lipid connection was not a statistical artifact of one variance assumption alone.
SHBG sits at an endocrine-metabolic intersection, and its association with a wide set of lipid measures is biologically plausible. Hormone transport, insulin sensitivity, hepatic metabolism, and lipid synthesis are tightly linked systems. A broad SHBG signal may therefore reflect systemic metabolic state rather than a disease-specific Alzheimer pathway. In older adults, those systems often move together, which makes SHBG a useful marker of metabolic context even when it does not provide a direct diagnostic clue for dementia.
No lipid module met the corrected threshold for the group score or the group-adjusted Aβ42 models. That absence is important. It suggests there was no stable network-level lipid signature strong enough to survive stringent correction in this dataset, despite the individual associations that appeared in unadjusted or alternative model structures.
What this means for biomarker work in Alzheimer disease
This study adds to a growing body of work asking whether peripheral lipids can help map Alzheimer biology before or alongside clinical decline. The answer here is cautious rather than expansive. There was evidence of association, but it was concentrated in one phosphatidylcholine for group score and in a wide SHBG-lipid relationship, while Aβ42 results shifted depending on the statistical approach.
That pattern is exactly what clinicians and biomarker researchers should expect from an early-stage observational dataset. Cross-sectional designs can reveal alignment between blood chemistry and disease state, but they cannot determine whether the lipid changes precede amyloid pathology, follow it, or simply travel alongside the metabolic changes that often accompany aging and neurodegeneration. The study also cannot separate medication effects, diet, comorbidity burden, or other unmeasured influences that commonly shape circulating lipids in older adults.
Even so, the analysis carries a useful message for future work: lipid biology may be informative, but only a subset of measures may matter after correction, and analytic choices can determine whether a signal survives. Larger cohorts, longitudinal sampling, and replication across platforms will be necessary before any of these serum lipids can be considered stable Alzheimer biomarkers.
For now, the clearest result is narrow. PC ae C38:1 tracked with the clinical-amyloid ordinal score, SHBG tracked with many lipids, and the Aβ42 signal changed shape as the model changed. The data support association, not diagnosis, and they leave the central question of temporal direction untouched.