Comparators Shape Plasma Proteomic Signals

Evidence

Two recent primary studies used plasma proteomics to search for diagnostic signals, but they tested different parts of the discovery process. One asked how sample preparation changes protein coverage and candidate findings in relapsing-remitting multiple sclerosis (RRMS). The other directly compared two clinically confusable causes of low platelet counts in children: MYH9-related disease (MYH9-RD) and immune thrombocytopenia (ITP). Together they distinguish analytical visibility from diagnostic specificity; they do not establish a clinical test.

The RRMS study first compared five preparation workflows for liquid chromatography–tandem mass spectrometry. Using plasma from five healthy controls, the researchers tested SP3, iST, and ENRICH-iST on raw plasma, plus SP3 after depletion of 14 abundant proteins and SP3 after enrichment of plasma extracellular vesicles. Raw-plasma SP3 quantified a mean of 391 proteins. Depletion followed by SP3 quantified 646, while extracellular-vesicle enrichment followed by SP3 quantified 923. The deeper workflows were not interchangeable: depleted plasma uniquely quantified 152 proteins and the extracellular-vesicle workflow uniquely quantified 161.

The investigators then applied the two deepest workflows to plasma from 15 women with RRMS at diagnosis, before treatment, and five sex- and age-matched healthy controls. Depleted plasma yielded 54 regulated proteins and extracellular-vesicle-enriched plasma yielded 35. Only four appeared in both lists: A2M, IGKV3-15, APOA2, and HYOU1. The authors highlighted von Willebrand factor (VWF), which was elevated in the RRMS group, as a potential diagnostic biomarker. The result was a discovery-stage candidate, not a validated marker. The full text also notes that distinguishing an RRMS-specific signal from proteins associated with other inflammatory or neurological conditions remains difficult.

The newer pediatric study built a more direct differential-diagnosis comparison. It used data-independent-acquisition plasma proteomics in 10 children with MYH9-RD, 10 with ITP, and 14 healthy controls. The analysis quantified 2,533 proteins and reported 91 proteins specific to the MYH9-RD comparison and 124 specific to the ITP comparison. The abstract describes different immune and metabolic patterns, but it does not provide effect sizes for individual proteins.

The researchers combined protein-interaction networks with Boruta feature selection and Random Forest algorithms to derive preliminary candidate panels. The MYH9-RD panel contained VWF, CETP, APOA1, and POSTN; the ITP panel contained NGAL, PAI-1, and CPB2. The abstract reports excellent apparent discrimination while explicitly calling the study exploratory and requiring rigorous external validation. No sensitivity, specificity, confidence interval, locked threshold, or external-cohort result is available in the ingested abstract, so those performance details cannot be inferred.

The studies therefore contribute complementary evidence. The RRMS work shows that preprocessing can change both proteome depth and the list of disease-associated proteins. The MYH9-RD/ITP work shows the value of including a disease comparator that resembles the target condition clinically. Neither study demonstrates that its proposed proteins retain diagnostic performance across laboratories or independent populations.

Analysis — Visibility Before Specificity

The cross-study inference is that plasma biomarker development has at least two separate validity layers. First, an analytical workflow determines which part of the plasma proteome becomes visible. In the RRMS study, depletion and extracellular-vesicle enrichment produced different protein counts and largely different regulated-protein lists. Second, the comparison design determines what a visible difference can distinguish. Healthy controls can reveal case-control separation, whereas the MYH9-RD study also included ITP, the clinically challenging alternative named in its diagnostic question. This is analysis, not a framework prospectively tested by either team.

VWF illustrates why the layers should not be collapsed. It appeared as an elevated RRMS candidate in one study and as one component of the preliminary MYH9-RD panel in the other. That recurrence does not replicate a disease-specific marker; instead, it may reflect a broader hemostatic or inflammatory signal that requires context. The studies used different diseases, cohorts, analytical strategies, and endpoints. A stronger validation sequence would therefore lock the preparation method, protein panel, and decision threshold, then test them blindly in independent cohorts containing realistic disease mimics. Only that design could show whether added proteome depth translates into stable differential-diagnostic information.

Limitations

The RRMS disease analysis was small and single-center, with 15 cases and five healthy controls, all women. The same five controls were used during workflow testing. Reported regulated proteins were identified with an unadjusted significance threshold, increasing the chance of unstable discoveries, and the comparator group did not include inflammatory or neurological disorders that can resemble RRMS. Deeper coverage also does not by itself mean better diagnostic accuracy.

The MYH9-RD/ITP study was available in the source pack only at abstract depth. Each disease group contained 10 participants, with 14 healthy controls. The abstract does not describe all preprocessing, missing-data handling, feature-selection safeguards, cross-validation structure, or whether model evaluation was separated from feature discovery. It reports apparent discrimination but no numerical performance estimates. The candidate panels are therefore hypothesis-generating and cannot be treated as validated classifiers.

Finally, these studies do not replicate one another. They examined different diseases and used different proteomic and statistical workflows. Their VWF findings cannot be combined into one effect estimate, and the recurrence of a protein name does not establish shared mechanism or diagnostic usefulness. Larger, independently recruited cohorts and prespecified analytical pipelines would be needed to determine which signals are reproducible, specific, and transportable.