Large rare-disease proteomic resource
The study analyses 972 CSF samples from 484 DIAN participants, combining cross-sectional and repeated observations across the ADAD continuum.
↳ Methods, Study Participants; Table 1
Abstract
BACKGROUND: Increasing evidence suggests that accurate prediction of Alzheimer's disease (AD) symptom onset requires more than amyloid- and tau-centric biomarkers such as cerebrospinal fluid (CSF) Aβ42/40, total tau and p-tau181 and plasma p-tau217. Autosomal dominant AD (ADAD), caused by pathogenic PSEN1, PSEN2 and APP mutations with predictable age at symptom onset, presents a unique opportunity to characterize the chronological changes in proteins beyond amyloid and tau and clarify them as early biomarkers of disease onset or as biomarkers related to disease staging and progression monitoring.
METHODS: We measured 972 CSF samples corresponding to 484 participants of the Dominantly Inherited Alzheimer Disease Network (DIAN) using the NULISASeq 120 CNS Disease Panel. We first benchmarked the technology against gold-standard measurements followed by the identification of proteins that were differentially abundant in relation to mutation status and symptomatology. Next, we determined the chronological emergence of protein changes in relation to the estimated years to onset (EYO). Finally, we assessed whether specific protein measures improved the prediction of EYO in the ADAD.
FINDINGS: NULISA measurements were comparable to those previously published. We demonstrated that known early alterations in CSF amyloid and tau were followed by inflammatory and neurodegenerative responses suggesting that clinical manifestation of AD happens before the inflammatory processes is fully developed. Finally, we found a multi-protein composite approach for predicting EYO that outperformed single biomarker values.
INTERPRETATION: Our results suggest that the main CSF proteomic landscape changes in ADAD are due to the presence of a pathogenic mutation and occur prior to symptom onset. Improved performance of multi-protein composite to predict EYO compared to single biomarker values highlights the added value of multiplex proteomic signatures for biomarker panel development.
Key Claims
Primary findingTentative
Our results suggest that the main CSF proteomic landscape changes in ADAD are due to the presence of a pathogenic mutation and occur prior to symptom onset.
Possible — the claim's weakest requirement is that the effect exists at all. Experimental. Every point below is traceable to a quoted passage, but this forecast has not yet been validated against known replication outcomes.
observational carrier comparisons and threshold-based chronology do not fully identify mutation-driven presymptomatic change
Abstract, Interpretation
SecondarySupported
known early alterations in CSF amyloid and tau were followed by inflammatory and neurodegenerative responses
EYO-based ordering supports the direction, but selected abnormality thresholds could shift individual timings
Abstract, Findings; Results, chronological changes
SecondaryTentative
we found a multi-protein composite approach for predicting EYO that outperformed single biomarker values.
superiority appears in the main test set but not in the held-out clinical converters
Abstract, Findings; Results, multiprotein models
SecondaryTentative
These results suggest that mutation in these three genes drive changes that are similar regardless of the affected ADAD gene.
small gene subgroups and weak or negative effect-size correlations do not establish shared profiles
Results, gene-stratified comparison; Figure 1D
SecondarySupported
Overall, we showed that NULISA measurements strongly correlate to the established immunoassay and IP-MS methods in the DIAN study
most analytes show strong cross-platform correlations, although SNAP25 is only moderately correlated
Results, benchmarking section; Figure 1A
Derived from the full evaluation — not a separate score.
Strengths · Limitations
Strengths
Orthogonal assay benchmarking and drift handling
NULISA measurements are compared with Lumipulse, IP-MS, SMC, and other immunoassays, and suspected Aβ freeze/thaw drift is explicitly investigated and handled in the chronology analysis.
↳ Methods, Protein Measurements, Benchmarking, and Chronological Changes; Results, Figure 1A
Substantive engagement with prior findings
The Discussion distinguishes replications from non-replications and offers platform or sample-selection explanations for discordant SNAP25, NPTX, and YWHAZ findings.
↳ Discussion, replication and non-replication paragraphs
Limitations
Converter results undercut predictive superiority
In 24 clinical converters, the multi-protein model’s MAE was 7.88, numerically worse than p-tau217, p-tau181, PiB PET, and age alone. This qualification is not carried into the abstract or the Discussion’s robust-superiority language.
↳ Results, converter paragraph; Abstract, Interpretation; Discussion, prediction paragraph
Chronology depends on selected thresholds
The temporal sequence depends on NC-derived 90th or 10th percentiles chosen for group separation and a 50% abnormality probability threshold, without reported sensitivity analyses around those choices.
↳ Methods, Chronological Changes of Protein Abundance
Cross-gene similarity lacks statistical support
The claim of shared dynamics across genes is difficult to reconcile with effect-size correlations of −0.09 for PSEN2 and −0.22 for APP, alongside an unexplained negative PSEN1 correlation.
↳ Results, gene-stratified comparison; Figure 1D
Nabu Commentary
The study’s contribution rests on the scale of the DIAN CSF resource, broad assay benchmarking, and integration of chronology with EYO modelling. Methodological Rigour is moderated by threshold-dependent chronology, unquantified exclusions, and unmodelled family dependence despite otherwise competent mixed modelling and internal validation. The principal interpretive problem is that the converter analysis does not reproduce the main test set’s multi-protein advantage, yet the Discussion describes robust superiority. The paper therefore functions most convincingly as a descriptive biomarker resource and platform benchmark rather than a validated prediction tool.
Check another paper
Author Note / Field Feedback
Nabu’s assessment, alongside the field’s view.
Are you an author of this paper?
Dimension breakdown
Study Quality
Sound3.3
Confidence highContribution
3.7
The study combines multiplex CSF profiling, assay benchmarking, temporal ordering, and EYO modelling in a large rare-disease cohort. It meaningfully extends prior ADAD proteomics, although much of the biological ordering confirms existing biomarker patterns.
Introduction; Results, Figures 1–2; Discussion
Early alterations in tau and amyloid were followed by inflammatory and neurodegenerative responses.
Methodological Rigour
3.2
The design includes orthogonal assay benchmarking, mixed models for repeated measurements, training-set feature selection, and held-out testing. Confidence is reduced by data-driven abnormality thresholds, unquantified QC exclusions, family dependence, and the converter analysis’s failure to confirm multi-protein superiority.
Methods, Quality Control, Chronological Changes, and Improvement of EYO Prediction; Results, Supplementary Figure 5
Although the multi-protein model showed numerically higher MAE than single protein models in converters
Reporting
3.3
The benchmark-to-chronology-to-prediction sequence is readily followable. The abstract and Discussion nevertheless generalise multi-protein superiority beyond the main test set despite contrary numerical results in converters.
Abstract, Findings and Interpretation; Results, converter paragraph; Discussion, prediction paragraph
robustly predicted time-to-symptom onset outperforming traditional single-marker approaches and amyloid imaging
Positioning
2.8
The paper engages relevant ADAD and sporadic-AD literature and reports several non-replications. Its limitations treatment is incomplete because the acknowledged low-powered converter result is followed by an unqualified statement of robust predictive superiority.
Discussion, non-replication, prediction, and limitations paragraphs
confirming the limited statistical power in this sub-cohort
Lower confidence on Methodological Rigour, Positioning — domain match limited.
Reliability
Caveats4 of 4 checks
Internal CoherenceNoted
Several reported numbers and interpretations are internally inconsistent, especially in the converter and gene-stratified analyses. These issues qualify central claims but do not establish wholesale unreliability of the dataset.
Research ConductClean
Ethics approval, consent, observational-study registration, and a defined DIAN data-request route are reported. No article-grounded research-conduct concern was identified.
Flags: 3 declared / 5 total
Reference IntegrityClean
60 of 60 checkable references verified
61 references in manuscript 1 have no canonical index record — counted, but not index-checkable 4 references confirmed by manual review
Post-Publication RecordClean
No retraction notice found in Retraction Watch.
Last checked: 08 Oct 2026
Sources: Retraction Watch ✓
Readiness
Where this paper’s evidence sits on the path from initial observation to real-world use.
Mechanistic
The multi-protein model and CHIT1 timing are research-stage findings rather than clinically calibrated tools. Relative quantification, absent external validation, and weak converter performance require substantial further work.
Results, Figure 2 and Supplementary Figure 5; Discussion, limitations paragraph
does not provide absolute quantifications
AI-generated, human-governed. Something look off? Contact us to request a review.