Abstract
The variability analysis of antibody arrays highlights the importance to consider variability components and their relative contributions when designing and evaluating studies for biomarker discoveries with exploratory, high-throughput and multiplexed methods.
Introduction
The search for reliable biomarkers—molecular indicators of disease—is often hindered by the inherent variability of biological systems and technical measurements. In proteomics, protein levels in human blood are influenced by a complex interplay of genetics, environment, short-term biological fluctuations, and technical noise. This study utilizes a longitudinal twin model to "unmix" these sources of variance, aiming to identify stable proteins that can serve as robust biomarkers. By analyzing 154 twins over two time points, the researchers quantified the relative contributions of these factors to protein variation.
Methods and Materials
The study employed a longitudinal twin design using serum samples from 154 healthy female twins from the TwinsUK database (77 pairs: 56 monozygotic and 21 dizygotic). Samples were collected at two separate visits, months apart.
Protein profiling was performed using antibody suspension bead arrays from the Human Protein Atlas, targeting 69 different proteins. The experimental workflow involves labeling serum proteins and capturing them with antibody-coated beads, followed by detection using fluorescent reporters and laser-based flow cytometry.
To decompose the total variance, Linear Mixed-Effects Models (LMM) were applied. The variance was partitioned into five components:
- Familial: Combined genetic and shared childhood environment.
- Individual Environment: Long-term stable factors unique to the individual.
- Short-term Biological: Fluctuations between visits.
- Experimental: Technical error and residual noise.
Data was refined using Probabilistic Quotient Normalization (PQN) to mitigate technical bias, and Akaike’s Information Criterion (AIC) was used for optimal model selection.
Results and Discussion
A primary finding was the significant impact of technical noise; the median proportion of total variation explained by experimental sources was 63%. This high level of noise underscores the challenge of biomarker discovery in high-throughput arrays, where signal can easily be drowned out.
Despite this noise, approximately 25% of the variation was found to be "stable" (combined familial and individual environment). For certain proteins, familial factors accounted for up to 42% of the variance. The study identified five proteins—IL-12a, PLAT, TTR, SERPINC1, and ADAMTSL4—where more than 50% of the variance was stable biological signal, making them prime candidates for clinical diagnostic development.
Normalization via PQN improved data consistency, increasing the median Spearman correlation between technical replicates from 0.26 to 0.36. Furthermore, model selection revealed that accounting for plate-specific intensity drifts significantly improved the fit for many proteins, suggesting that technical errors often follow systematic patterns.
Conclusion
While high technical noise remains a limitation of current antibody array platforms, this variance decomposition approach successfully identifies a subset of proteins with high biological stability. These stable proteins should be prioritized in future large-scale clinical studies. Future work will focus on applying these findings to specific disease contexts, such as Metabolic Syndrome, where stable biomarkers are essential for detecting meaningful deviations from health.
Summary updated