CSF Protein Signals Stay Stable Through Repeated Freeze-Thaw Cycles in High-Throughput Assay Test

Protein signals measured in cerebrospinal fluid remain remarkably stable even after samples have been frozen and thawed up to nine times, according to a new evaluation of a high-throughput proteomic platform published in BMC Biotechnology. The study, led by Marieta Hyde and colleagues, is the first to report on both the detectability and the reproducibility of the Proximity Extension Assay with Next Generation Sequencing readout when applied to cerebrospinal fluid, a matrix widely used in neurology research but never before systematically characterized on this particular platform. The findings carry practical weight for biobanks and clinical trials, where samples routinely endure multiple freeze-thaw cycles before they ever reach a proteomics laboratory.

The technology at the center of the study is the Proximity Extension Assay, or PEA, a dual-antibody method commercialized by Olink Proteomics. Originally designed to read out up to 96 biomarkers simultaneously, the platform was later developed into a version with Next Generation Sequencing readout that can measure several thousands of protein biomarkers at once. The assay provides relative quantitation of proteins, meaning it allows researchers to compare protein abundance between different samples and focus on changes rather than absolute concentrations. In this evaluation, the team used the Olink Explore 3072 assay, which spans eight panels covering roughly 3,000 biomarkers, with sequencing performed on an Illumina NovaSeq 6000 according to the manufacturer’s protocol.

Cerebrospinal fluid itself is a clear, colorless fluid that surrounds the brain and spinal cord, consisting primarily of water along with diluted proteins, diverse small molecules, and ions. Because of its direct interface with the central nervous system, it has seen extensive use as a source of biomarkers to diagnose central nervous system diseases and to assess the effectiveness of drugs acting on the brain and spinal cord. High-throughput proteomics platforms such as Olink and the aptamer-based SOMAScan technology are increasingly being deployed to profile large numbers of proteins in this fluid simultaneously, making rigorous characterization of assay performance in this matrix an urgent need.

For their evaluation, the researchers obtained cerebrospinal fluid samples collected by lumbar puncture from six subjects through the commercial vendor PrecisionMed. Two of the donors had a diagnosis of probable amyotrophic lateral sclerosis, determined by the revised El Escorial criteria, while the other four were controls with no known health conditions. Notably, all donor samples came from individuals aged 60 and above, a demographic detail the authors highlight as an important scope limitation. Each sample was prepared with four technical replicates and processed across all eight PEA panels, with quality control parameters met using four internal controls added to every sample that monitor the immunoreaction, extension, and amplification and detection steps of the protocol.

The first major result concerns detectability, defined as the percentage of proteins whose Normalized Protein Expression values exceed the limit of detection, which is calculated for each assay and plate as typically three standard deviations above background noise estimated from negative controls. Across the four original panels, detectability was consistently above 60 percent, with the Cardiometabolic panel reaching the highest value at 90 percent and Oncology the lowest at 61 percent. The four expansion panels told a different story: detectability was much lower, falling to 58 percent for Inflammation II and as low as 22 percent for Oncology II. Overall, the expansion panels fell short of the original panels by margins reaching as high as 50 percent when comparing Cardiometabolic with Cardiometabolic II.

The authors attribute this pattern to the nature of the proteins each set of panels targets. The four original panels focus on extracellular and circulating proteins, which are naturally more abundant in cerebrospinal fluid, while the four expansion panels cover a higher number of intracellular and disease-specific proteins that appear at far lower levels in this matrix. A similar reduction in detectability for expansion panels has been observed with plasma samples in vendor documentation, but the effect proved more pronounced in cerebrospinal fluid. The reproducibility analysis showed the same divide: with the exception of the Inflammation panel, coefficient of variation values were higher for all expansion panels than for their original counterparts, which the researchers hypothesize reflects the larger number of proteins falling below the limit of detection.

Crucially, when the team recalculated variability while restricting the analysis to proteins whose values exceeded the limit of detection, the coefficients of variation dropped to less than 15 percent on average. That figure is comparable to the variation calculated from plate controls and sits well within the vendor’s recommended range. Moreover, the mean differences in variability between the original and expansion panels shrank substantially under this restriction. The practical implication is stark: many proteins, particularly on the expansion panels, are effectively undetectable in cerebrospinal fluid, and measurements below the limit of detection carry high variance that can mislead downstream analyses.

The second half of the study tackled a question that haunts nearly every biobank: what happens to protein signals when samples are repeatedly frozen and thawed? To simulate everyday laboratory operations, each aliquot was subjected to six additional freeze-thaw cycles beyond those already incurred, with cycles induced by removing samples from minus 80 degree Celsius storage and incubating them at room temperature for 20 minutes, a duration visually confirmed to fully thaw the samples, before refreezing. One cycle was conducted per day to ensure complete refreezing. In total, the analysis spanned samples with between 3 and 9 freeze-thaw cycles, drawn from 88 randomized samples plated across two 96-well plates.

The outcome was striking. Across all eight panels, the number of freeze-thaw cycles did not appear to affect protein detectability at all. Principal component analysis of 36 samples, covering six subjects each with six different cycle counts, showed that samples clustered primarily by donor rather than by the number of freeze-thaw cycles, indicating that the dominant source of variation between samples was the individual subject, not the handling history. This donor-driven clustering was unambiguous in the four primary panels, though it became less pronounced in the noisier expansion panels. When the researchers compared expression levels of 12 proteins that a previous study had implicated as affected by freeze-thaw cycling, none showed significant differences between cycle 4 and cycle 8, with all p-values exceeding 0.05, and a global differential expression test with subject as a random effect detected no protein with significant differences between the two cycles.

The authors are careful to situate their results against earlier work. A previous study by Hok-A-Hin and colleagues, using a combination of proteomic platforms including PEA with qPCR readout, had found that 85 percent and 76 percent of proteins remained stable after eight freeze-thaw cycles depending on the technology used. The present study found no significant differences at all for the overlapping proteins, and the authors point to key design differences: their experiment focused on quick sample aliquoting conditions, whereas the earlier work involved longer laboratory processing with extended room-temperature exposure; the two studies also used different readout platforms, which Olink considers separate and non-interchangeable, and four years separated the studies, raising the possibility of changes in antibody design. Given these conflicts, the authors advise researchers to remain agnostic about which specific proteins may degrade with additional cycles and to avoid treating any particular protein as safe or compromised. They also acknowledge that degradation may occur within the first several cycles, which their design could not rule out, and they recommend limiting cycles where feasible, though they note that samples from clinical trials have typically already undergone at least three. With the vendor since releasing a broader Explore HT kit targeting more than 5,000 assays that is compatible with Explore 3072, the authors believe their findings remain relevant to the newer platform as well.

Subject of Research: Technical performance and freeze-thaw stability of the Olink proximity extension assay with NGS readout in cerebrospinal fluid proteomics

Article Title: Evaluating assay performance and freeze-thaw stability in cerebrospinal fluid samples using High-Throughput Proximity Extension Assay Platform

Article References: Hyde, M., Cunningham, M., Chai, X., Peng, M., Hu, H. A., Lu, C., & Vasanthakumar, A. (2026). Evaluating assay performance and freeze-thaw stability in cerebrospinal fluid samples using High-Throughput Proximity Extension Assay Platform. BMC Biotechnology, 26(1), Article 125. https://doi.org/10.1186/s12896-026-01144-y

Image Credits: AI Generated

DOI: 10.1186/s12896-026-01144-y

Keywords: cerebrospinal fluid, proteomics, proximity extension assay, Olink Explore 3072, freeze-thaw cycles, protein detectability, coefficient of variation, biomarkers, biobanks, next generation sequencing, amyotrophic lateral sclerosis, pre-analytical sample handling