TL;DR
Long-term cryopreservation of DNA, even for just a few years, introduces a subtle but detectable bias towards hypomethylation at individual CpG sites. While global methylation levels remain largely unchanged, this site-specific degradation can skew results in sensitive downstream analyses, particularly those using machine learning. Researchers working with cryopreserved DNA should be aware of this bias and consider its potential impact on their findings.
What Problem Does This Paper Address?
DNA methylation is a crucial epigenetic mark that plays a significant role in gene regulation and is increasingly used as a biomarker for various diseases. To enable large-scale studies and robust statistical analyses, researchers often rely on biobanks that store DNA samples for extended periods, sometimes for decades. However, the impact of long-term cryopreservation on DNA methylation profiles is not fully understood, and previous studies have yielded conflicting results, reporting hypomethylation, no change, or even hypermethylation. This ambiguity poses a significant challenge, as biases introduced by sample storage could lead to inaccurate conclusions in methylation analyses, potentially affecting diagnostic and prognostic applications. This paper aims to clarify the extent and nature of DNA methylation changes associated with cryopreservation over several years.
Methods
The study employed an epigenome-wide DNA methylation profiling approach using the Illumina MethylationEPIC bead chip, which targets over 850,000 CpG sites. They analyzed DNA from two independent prospective cohorts: a discovery cohort (n=126) and a validation cohort (n=136).
Key methodological steps included:
- Sample Handling: Samples were collected and processed under different short-term storage conditions (blood/buffy coat) before DNA extraction and long-term cryopreservation as DNA.
- DNA Extraction and Methylation Assay: Standard DNA extraction protocols were followed, and methylation profiling was performed using the Illumina MethylationEPIC array.
- Data Preprocessing and Quality Control: This involved removing cross-reactive probes, applying functional normalization, removing CpGs near SNPs, and filtering out CpGs with low detection p-values.
- Statistical Analysis:
- Global Methylation: Robust linear regression was used to assess the association between mean global methylation (quantified as mean M-value) and storage duration as DNA, while controlling for covariates like age, cell type composition, and buffy coat storage duration.
- Individual CpG Methylation: Differential methylation analysis for individual CpGs was performed using linear models (limma package) to assess the change in methylation M-value per year of storage. This was done both with and without controlling for covariates. The q-value was used to control for false discovery rates.
- Genomic Distribution Analysis: The study investigated whether hypomethylated CpGs were enriched in specific genomic regions (e.g., CpG islands, shores, shelves, open sea).
Datasets
The study utilized two datasets derived from peripheral blood samples of healthy female subjects of Chinese ancestry:
- Discovery Cohort: n=126
- Validation Cohort: n=136
These cohorts were distinct in terms of age, specific short-term storage conditions before DNA extraction, and the duration of DNA storage. The study did not introduce a new benchmark dataset but rather used existing samples from biobanks.
Key Insights and Ideas
- Global vs. Individual CpG Changes: The most striking insight is the divergence between global and individual CpG methylation. While overall methylation levels appear stable, individual CpG sites show a consistent trend towards hypomethylation with longer storage. This highlights the limitations of looking only at aggregate methylation.
- Hypomethylation Bias: Cryopreservation introduces a bias towards hypomethylation at the individual CpG level. This means that over time, some methylated CpGs are more likely to appear unmethylated, potentially due to spontaneous hydrolytic deamination of 5-methylcytosine to uracil.
- CpG Island Proximity Matters: CpGs located near CpG islands (but not within them) are more prone to hypomethylation during cryopreservation. This suggests that the local genomic context influences the stability of methylation marks. CpG islands themselves are often unmethylated, which might explain their resistance to further hypomethylation.
- Implications for Machine Learning: The subtle but systematic bias could be particularly problematic for "black box" machine learning algorithms, which are sensitive to small changes and can be difficult to interpret, potentially leading to misclassification or biased predictions.
Key Results
- Global Methylation: No detectable change in mean global methylation was observed with increasing storage durations as DNA in the discovery cohort (p=0.360). In the validation cohort, there was a slight increase (p=0.030) that disappeared after controlling for covariates.
- Individual CpG Methylation:
- In the discovery cohort, 4,049 CpGs (0.5%) were hypomethylated at q-value < 0.05, while only 50 were hypermethylated.
- In the validation cohort, 63 CpGs were hypomethylated at q-value < 0.05, and only 6 were hypermethylated.
- These trends persisted even after controlling for covariates.
- Genomic Distribution: CpGs closer to CpG islands were more likely to be hypomethylated, with odds ratios indicating a significant enrichment of hypomethylated CpGs in CpG shores (North and South) and open sea regions.
Limitations and Future Work
- Effect Size: The study acknowledges that the effect size of cryopreservation on methylation is small, especially at the analyzed timescales (up to ~50 months). This might explain why previous studies with fewer tested loci or global analyses found negligible effects.
- Storage Duration: While the study analyzed up to 50 months, longer durations (decades) were not directly assessed.
- Storage Conditions: Although the study used two cohorts with different storage protocols, a comprehensive comparison of various storage conditions (temperature, media, buffers) was not the primary focus.
- Future Work: The authors emphasize the urgent need for developing more stable methods of DNA preservation. They also suggest that future computational methods should be developed to account for the effects of cryopreservation on DNA methylation profiles, especially for large-scale biobanking initiatives. Further research is needed to understand the precise mechanisms of degradation and to establish optimal storage protocols.
Summary updated