Intelligence

Analysis

When three spectral-flow signatures vanish after FDR: a cytometry interpretation lesson

An exploratory pediatric exercise sub-study observed three nominal immune-metacluster interactions among 20. None survived multiplicity adjustment; rigorous interpretation is the result.

A high-dimensional cytometry experiment can generate a beautiful map and an intriguing biological story before the statistics have finished their work. In a small immunological sub-study of exercise during pediatric cancer treatment, three immune-cell clusters appeared to change differently between study groups. None of those three findings survived adjustment for the number of clusters examined. That reversal—not the attractive plot—is the most valuable lesson for clinical and translational cytometrists.

Yanguas-Casás and colleagues reported this exploratory ancillary analysis of the randomized FORTEe exercise trial in Frontiers in Immunology on 13 November 2026. Paired blood samples were available for 29 pediatric patients with solid tumours: 14 in the exercise arm and 15 controls. The supervised intervention ran for approximately 8–10 weeks. These 29 participants are a study subset, not a separate adequately powered randomized trial of immune endpoints.

Original CFCM conceptual four-panel graphic showing a longitudinal 29-patient spectral cytometry design, 20 examined metaclusters, three nominal signals and zero surviving false-discovery-rate correction.
Original CFCM conceptual diagram. Educational synthesis; not clinical validation.

What was actually measured?

Peripheral blood mononuclear cells were cryopreserved, stained and measured on a Cytek Aurora full-spectrum cytometer. The study describes routine instrument controls, single-stained unmixing controls, fluorescence QC, viability and doublet exclusion, and reference material across experimental batches. High-dimensional analysis assigned cells to 20 immune metaclusters and used visualization to inspect lineage and differentiation signatures. Those maps are descriptive summaries of multivariate data, not physical maps of cell migration and not substitutes for gating and marker-by-marker examination.

For each group, the authors compared baseline and post-intervention measurements. A group-by-time interaction in a two-way mixed ANOVA asks whether the change over time differs between intervention and control arms, not merely whether two colours on a low-dimensional map are visually separated. Because multiple clusters were tested, the investigators also applied Benjamini–Hochberg false-discovery-rate (FDR) correction.

Three plausible phenotypes—and a statistical limit

Three of the 20 metaclusters reached nominal p values below 0.05: MC-01 (p = 0.014), a myeloid antigen-presenting population with dendritic-like features; MC-02 (p = 0.031), a mature B-cell-like group; and MC-04 (p = 0.030), a differentiated CD8 TEMRA-like population. The specific marker signatures are biologically plausible, but the study explicitly reports that none of these interactions remained statistically significant after FDR correction. Therefore the correct description is three exploratory nominal signals, not three validated exercise-induced immunological effects.

No significant group-by-time difference was detected in the principal circulating T-, B-, NK- or monocyte distributions. Absence of statistical significance is not evidence of equivalence: the sample was small, heterogeneous and exposed to clinical treatment effects. Conversely, nominal cluster-level trends are not evidence that exercise improves antitumour immune surveillance or changes clinical outcomes. Functional, clinical and tumour-site measurements would be necessary for those interpretations.

Why high-dimensional workflows are vulnerable

Clustering opens many analytical degrees of freedom: preprocessing, cells retained, quality-control cutoffs, marker scaling, number of clusters, annotation, donor selection and batch correction. A signal can depend on one of these choices even if the phenotypic label makes biological sense. When the dataset contains paired longitudinal observations, correlations between repeated samples must also be respected; pooling all events as independent observations can manufacture precision. Report the participant, sample and cell-event denominators separately.

Immunophenotyping after cryopreservation introduces additional risk. Cell viability, treatment timing, reference controls and relative proportions can bias which cells remain analyzable. Moreover, a rise in a relative-frequency cluster need not mean its absolute cell count has increased. In this study, the absence of absolute blood counts prevents converting all percentage changes into absolute circulating changes.

How a better replication might look

Specify the principal immunological endpoint and multiplicity family in advance; power the analysis on biological participants rather than cellular events; apply standardized acquisition, unmixing and quality controls; predefine acceptable viability and event counts; test robustness of clustering and marker assignments; and include an independent replication cohort. If the proposed mechanism involves cytotoxicity or trafficking, add orthogonal assays rather than inferring function from a differentiated T-cell phenotype alone.

The broader conclusion is constructive. High-dimensional spectral cytometry can reveal specific hypotheses that broad lineage counts miss. Its credibility increases when a well-characterized cluster is followed by transparent uncertainty, multiplicity correction and independent confirmation.

Daily Learning Pill: A cell cluster with a convincing phenotype and a nominal p value is a candidate hypothesis until the multiplicity plan and independent validation support it.

Source: Yanguas-Casás N et al., Frontiers in Immunology (13 November 2026), DOI 10.3389/fimmu.2026.1993392. Original research from an exploratory trial subset, not validated clinical biomarker performance. Illustration created specifically by CFCM; no published figure is copied.