# A novel multiplexed biomarker panel reveals key molecular pathways in progressive multiple sclerosis.

- **DOI:** [10.1093/braincomms/fcag142](https://doi.org/10.1093/braincomms/fcag142)
- **Rating:** Limited (2.5 out of 5)
- **Verdict:** Longitudinal multiplex proteomics nominates KITLG and plasma cytokines as progression markers in multiple sclerosis, but internal model selection and reporting inconsistencies limit confidence.
- **Framework version:** v5.9
- **Pipeline version:** v0.4
- **Evaluation date:** 2026-10-08
- **Evaluation page:** https://www.nabu.science/paper/10.1093%2Fbraincomms%2Ffcag142

## Key claims

1. Among RRMS patients, low baseline CSF concentrations of KIT ligand (KITLG) predicted both conversion to SPMS and future PIRMA events.
   - Support: Tentative
   - Type: Primary finding
   - Basis: small overlapping outcome groups, an internally derived cut-off, event-poor adjusted models, and no external validation
   - Location: Abstract; Results, Fig. 4
2. We identified distinct proteomic signatures in CSF of both RRMS and iPMS patients compared with controls, with no significant differences in plasma.
   - Support: Supported
   - Type: Secondary
   - Basis: FDR-controlled adjusted CSF comparisons support the direction, while the small plasma sample limits confidence in the null result
   - Location: Abstract; Results, Fig. 1–2
3. A model selection analysis identified a two-protein logistic regression model including interleukin 1 beta and interleukin 36 gamma as the best-performing combination (area under the curve = 0.990).
   - Support: Tentative
   - Type: Secondary
   - Basis: the apparent AUC follows selection in a very small cohort without cross-validation and duplicates another model’s reported statistics
   - Location: Abstract; Results, Fig. 5C

## Strengths

- **Long follow-up links baseline biomarkers** Baseline samples from untreated RRMS patients are connected to a median nine years of follow-up, including SPMS conversion and sustained PIRMA outcomes.
  - Dimension: contribution
  - Location: Methods, Patients and Samples; Results, Fig. 4
- **Standardized sampling and multiplicity control** Samples followed the BioMS-EU handling protocol, and differential-expression analyses used age- and sex-adjusted models with FDR correction.
  - Dimension: methodological rigour
  - Location: Methods, NULISA Proteomic Analysis and Statistical Analysis
- **Prior proteomic evidence is engaged** The Discussion compares the findings with prior CSF signatures and a large plasma proteomic study that reported broader alterations, rather than presenting the results in isolation.
  - Dimension: positioning
  - Location: Discussion, first two paragraphs

## Limitations

- **Potentially duplicated headline model statistics** The IL1B–IL36G SPMS model and the CCL27–NGF–SPP1 PIRMA model report exactly the same AUC, confidence interval, and P-value, despite using different proteins and outcomes.
  - Dimension: reporting
  - Location: Results, Fig. 5C and Fig. 5E
- **Model selection precedes performance evaluation** Proteins were filtered by outcome correlation and models were dredged by AICc, while validation was absent for plasma models and was not shown to encompass feature selection for the CSF models.
  - Dimension: methodological rigour
  - Location: Methods, Statistical Analysis; Results, Fig. 3 and Fig. 5
- **Progression evidence lacks independent validation** The prognostic analyses use few events, internally selected cut-offs, and no strong clinical-only comparator or external cohort. PIRMA and SPMS conversion also overlap substantially.
  - Dimension: methodological rigour
  - Location: Results, Fig. 4–5; Discussion, limitations

## Reliability checks

Caveats (4 of 4 checks).

- **Results check:** Noted
  - A model selection analysis identified a two-protein logistic regression model including interleukin 1 beta and interleukin 36 gamma as the best-performing combination (area under the curve = 0.990): "To evaluate their predictive potential, we performed a model selection analysis, which identified a two-protein logistic regression model including IL1B and IL36G as the best-performing combination (AUC = 0.990, 95% CI: 0.961–1.000, P < 0.001), slightly outperforming the individual proteins" (Results, Plasma concentrations of two proteins are associated with conversion to SPMS, Fig. 5C) vs "From these seven altered proteins, we could identify a prediction model (Best Model) for PIRMA consisting of CCL27, NGF and SPP1 outperforming any of the individual proteins alone (AUC = 0.990, 95% CI: 0.961–1.000, P < 0.001)" (Results, Plasma concentrations of two proteins are associated with conversion to SPMS, Fig. 5E). The SPMS-conversion and PIRMA models report identical AUCs, confidence limits and P-value bounds despite different predictors and outcomes, but those equalities alone do not establish a transcription error. → Effect on result unclear
  - Minor reporting notes (2):
    - Descriptive: "Following a 10-fold CV, the model retained strong performance (CV AUC = 0.91)" (Results, Top DEPs may assist in differentiating between RRMS and iPMS at baseline, MS-versus-controls classifier) vs "achieved an AUC of 0.934 and remained strong after CV (AUC = 0.916)." (Discussion, MS-versus-controls classifier paragraph). The MS-versus-controls cross-validation AUC is 0.91 in Results and 0.916 in Discussion, a difference of 0.006, without information identifying the intended estimate. → Effect on result unclear
    - Descriptive: "which, after 10-fold CV, presented an AUC of 0.799." (Results, Top DEPs may assist in differentiating between RRMS and iPMS at baseline, RRMS-versus-iPMS classifier) vs "also showed a strong predictive performance (AUC = 0.871; CV AUC = 0.823)." (Discussion, RRMS-versus-iPMS classifier paragraph). The RRMS-versus-iPMS cross-validation AUC is 0.799 in Results and 0.823 in Discussion, a difference of 0.024, without information identifying the intended estimate. → Effect on result unclear
- **Reporting and transparency:** Noted. The Discussion introduces an unreported IKBKG result and a log-rank analysis not described in Methods or Results. Funding, conflict-of-interest and code-availability statements are absent from the supplied text; the conventional data-on-request statement is not itself a reliability concern.
  - Conflict of interest: not stated · Ethics approval: declared · Data availability: declared · Code availability: not stated
- **Reference integrity:** Clean
  - 50 references in manuscript
  - 50 of 50 checkable references found in an index
  - References not verified automatically are resolved manually within 24 hours.

## Post-publication record

- **Status:** Clean
- **Summary:** No retraction or expression of concern found in Crossref (includes Retraction Watch).
- **Last checked:** 2026-10-17
