BGPT: Paper Review: Genetic predisposition, Aβ misfolding in blood plasma, and Alzheimer’s disease

Fuel Your Discoveries

Quick Explanation Copied

Core finding (skeptical):

In a German community-based cohort (ESTHER; n=728 with baseline plasma Aβ misfolding measured), blood Aβ misfolding showed much stronger discrimination for incident Alzheimer’s disease within 14 years (AUC ≈ 0.84) than an AD polygenic risk score (AUC ≈ 0.59) and than APOE ε4 (AUC ≈ 0.63)—with genetic predictors also relating to misfolding in the full sample but not in non-demented participants. (Reported in the paper.)

Long Explanation

Paper Review (science-first): Genetic predisposition, Aβ misfolding in blood plasma, and Alzheimer’s disease

DOI: 10.1038/s41398-021-01380-0 • Translational Psychiatry • Received Nov 2020; accepted Apr 2021 • Nested case-control within ESTHER; baseline blood plasma Aβ misfolding measured; incident AD/VD over 14 years.

Key question: Do genetics (APOE, AD PRS, Aβ PRS) explain future Alzheimer’s risk directly, or do they mainly relate to an intermediate blood biomarker (Aβ misfolding) that predicts disease better?

Visual: predictive performance (reported AUCs)

Plotly reproduces the paper’s reported AUC values for AD diagnosis within 14 years (AUC ≈ 0.84 for Aβ misfolding).

Visual: genetic predictors → Aβ misfolding (odds ratios)

Paper’s adjusted logistic regression results for association with baseline Aβ misfolding (full sample): AD PRS per SD and APOE ε4 were associated; Aβ PRS was not.

Visual: “intermediate biomarker vs genetics” hypothesis check

The paper reports: (i) genetics (AD PRS, APOE ε4) relate to misfolding in the full analyzed sample, but (ii) misfolding relates much more strongly to incident AD than genetics. This is consistent with misfolding acting closer to the phenotype than upstream common-variant risk.

Methods & design (what is actually measured)

Population & follow-up

Participants came from the ESTHER study (community-based German cohort; baseline age 50–75, recruited 2000–2002), and this analysis used a nested case-control subset with baseline genetic data and baseline plasma Aβ misfolding, followed for up to 14 years for incident dementia diagnoses.

Biomarker: Aβ misfolding in blood plasma (intermediate readout)

The marker quantifies Aβ secondary-structure change using an immuno-infrared sensor readout; the paper uses a validated spectral threshold (the paper states <1642 cm⁻1 as misfolding positive/high misfolding) and references prior technical validation (blinding, matrix effects, antibody characterization, and validation performance).

Genetics: APOE and two PRSs

The paper computes two PRSs: an AD polygenic risk score derived from IGAP stage-1 summary statistics (excluding SNPs in/near APOE during PRS construction) and an Aβ 42–based polygenic risk score derived from Deming et al.’s CSF Aβ GWAS summary statistics.

Core results (only what is explicitly reported)

Genetic associations with blood Aβ misfolding

In the full analyzed sample, the paper reports: AD PRS per SD increase is associated with Aβ misfolding (OR 1.25, 95% CI 1.03–1.51), and APOE ε4 presence is associated with Aβ misfolding (OR 1.61, 95% CI 1.04–2.49). The Aβ PRS shows no association (OR 1.05, 95% CI 0.85–1.29). The same associations are reported as absent in the subgroup with no dementia diagnosis during follow-up.

Prediction of incident Alzheimer’s disease (within 14 years)

The paper reports AUC values for AD diagnosis within 14 years: Aβ misfolding AUC 0.84 (95% CI 0.78–0.90), APOE ε4 AUC 0.63 (0.55–0.70), AD PRS AUC 0.59 (0.51–0.68), and Aβ PRS AUC 0.55 (0.46–0.63). The paper also reports that Aβ misfolding predicts AD more accurately than the genetic markers (and that combined models including misfolding outperform genetic-only models, with stated ROC contrast p-values).

Prediction of vascular dementia (VD)

The paper reports minimal/absent predictive ability for VD: AUCs near chance for genetic markers and misfolding (e.g., Aβ misfolding AUC ~0.54 with CI 0.46–0.63).

Critical appraisal (skeptical, mechanism-aware)

1) Assay-PRS construct mismatch

The Aβ PRS is based on CSF Aβ42 levels GWAS, while the misfolding assay is a structural/secondary-structure readout in blood plasma. This mismatch plausibly weakens direct correlation between the PRS and misfolding status—even if both reflect Alzheimer-related biology. The paper explicitly raises this as a plausible explanation for the lack of Aβ PRS ↔ misfolding association.

2) Predictive strength ≠ mechanistic proof

An AUC boost indicates discrimination, not necessarily causation or full biological locality. It remains possible that Aβ misfolding is a late surrogate for undetected neuropathology, or that common variants primarily modulate pathways that indirectly influence misfolding via intermediate steps not fully captured here. The paper itself frames misfolding as potentially important for initiation vs risk prediction, but that conceptual mapping is hypothesis-level, not definitive causal inference.

3) Subgroup nulls may reflect timing, power, or selection

The reported absence of genetic ↔ misfolding associations among participants without dementia diagnoses could be due to (a) earlier onset in high-risk individuals shifting misfolding-positive participants into the dementia groups, (b) limited statistical power, or (c) misclassification/heterogeneous diagnosis in real-world practice. The authors suggest an “earlier misfolding in high genetic risk” timing explanation, but that remains speculative without longitudinal repeated biomarker trajectories.

4) Clinical endpoint definition risk

Dementia diagnoses are clinical and made by multiple practitioners in a community cohort; the paper flags potential misdiagnosis/underdiagnosis and heterogeneity vs specialized standards. That matters because ROC analyses are only as good as the reference label.

5) Conflict of interest & assay patenting

The paper states the immuno-infrared sensor was applied for a patent by Klaus Gerwert and Andreas Nabers (WO 2015121339 A1). This does not invalidate the results, but it increases the importance of independent replication and rigorous analytic transparency.

What would most disprove or materially change the conclusions?

(i) Independent cohorts showing that Aβ misfolding (same assay/threshold or harmonized equivalent) fails to discriminate incident AD once genetics and other confounders are accounted for; (ii) longitudinal evidence that Aβ misfolding adds predictive value beyond other blood/CSF/PET measures at the same timepoint; (iii) consistency across ancestries (the paper notes limited generalizability, implying that population stratification and PRS transferability could matter).

Data used (from the paper text provided)

Component	What the paper reports	Where it matters
Sample	n=728 analyzed; 114 Aβ misfolding positive; 59 AD, 54 VD, 34 mixed dementia; 581 no dementia at 14y	Event rate + power
PRS ↔ misfolding	AD PRS OR~1.25; APOE ε4 OR~1.61; Aβ PRS null	Upstream genetics to intermediate biomarker
Aβ misfolding ↔ AD	AUC~0.84 (95% CI 0.78–0.90) vs genetic markers AUC~0.55–0.63	Discrimination for clinical risk
VD	AUC near chance for predictors	Specificity signal
Assay threshold	Validated spectral cutoff (<1642 cm⁻1) defines misfolding positive	Binary classification stability
Conflict	Sensor linked to patent filing	Replication importance

Author reviews (bespoke deep-dives)

Click to open independent author-focused reviews on BGPT.

Feedback:

Updated: April 19, 2026