Defining the Cohort Study Question

Every design decision flows from the research question. In human cohort studies, the question determines sample size, compatible sample types, assay resolution, and analysis strategy.

Write a concise question that specifies what you are comparing. A discovery-oriented question like "which CpG loci differ between treatment responders and non-responders" requires genome-wide coverage with correction for millions of tests. A hypothesis-driven question like "is promoter methylation of gene X associated with survival" can use a targeted assay at lower cost.

Document the reference genome build (GRCh38 or T2T-CHM13), which affects probe mapping and alignment accuracy. Also specify the comparison structure upfront — a two-group comparison differs in pipeline requirements from a longitudinal or multi-factorial design.

Experimental group definitions matter for downstream analysis:

• Two-group comparison — cases versus controls with 3-5 biological replicates per group
• Longitudinal or paired design — same individuals at multiple time points, paired or mixed-effects analysis
• Multi-factorial design — multiple variables requiring interaction modeling
• Quantitative trait design — methylation correlated with a continuous phenotype, regression-based analysis

The clearer the comparison structure, the easier the assay selection and pipeline setup.

Sample Types and Preparation Guidance

Human cohort methylation studies draw from several sample types, each with distinct collection protocols, DNA quality profiles, and assay compatibility considerations.

Blood Samples

Blood is the most accessible sample type for large human cohort methylation studies. Whole blood contains a mix of cell types — neutrophils, monocytes, lymphocytes — each with a distinct methylation landscape. This cellular heterogeneity is the central technical challenge for blood-based studies and must be addressed through cell-type deconvolution during analysis.

Blood collection should use EDTA or citrate tubes. Heparin should be avoided as it inhibits bisulfite conversion and PCR. Process blood within 2-4 hours to minimize ex vivo methylation changes. Typical DNA yields from 1 mL of whole blood range from 30-50 µg, with A260/280 between 1.7 and 1.9. High-molecular-weight DNA (>20 kb) is preferred for sequencing-based assays.

Tissue Samples

Fresh-frozen tissue is the gold standard for tissue-based studies. Snap-freeze in liquid nitrogen within 30 minutes of collection and store at -80°C. OCT-embedded tissue is acceptable but may introduce contaminants requiring additional cleanup.

Cell-type heterogeneity is more pronounced in tissue than blood. A tumor biopsy contains malignant cells, stromal cells, and immune infiltrates. Laser capture microdissection or fluorescence-activated nuclei sorting can isolate specific populations when cell-type-specific resolution is needed. A typical 5-10 mg punch yields 5-15 µg of DNA.

FFPE Samples

FFPE samples are the most challenging type for methylation analysis. Formalin fixation crosslinks DNA with proteins, fragments DNA to 200-500 bp, and can chemically modify cytosines in ways that interfere with bisulfite conversion. Despite these challenges, FFPE is often the only material available for retrospective studies with long clinical follow-up.

Success depends on storage duration (samples stored less than 5 years perform better), fixation time (over-fixation beyond 24-48 hours increases crosslinking), and the extraction protocol. FFPE DNA typically yields 0.5-5 µg from 2-4 sections. RRBS with modified library protocols and methylation arrays are the most reliable options. WGBS and EM-seq are possible with FFPE-optimized kits but produce lower library complexity.

cfDNA from Plasma

Cell-free DNA from plasma is useful for longitudinal monitoring and studies where tissue biopsy is not feasible. cfDNA is highly fragmented (~166 bp) and present at very low concentrations — typically 5-50 ng per mL of plasma.

Plasma collection requires cell-stabilizing tubes (Streck Cell-Free DNA BCT) or EDTA tubes processed within 2 hours, followed by double centrifugation to remove cellular debris. Standard genomic DNA extraction methods do not efficiently recover short fragments — dedicated cfDNA kits are required.

Because yields are low, library preparation typically uses all available DNA, and PCR duplicate rates are higher than with genomic DNA. Assays compatible with low-input DNA — EM-seq, RRBS, and array-based methods — are preferred.

Figure 2. Sample types commonly used in human cohort methylation studies — blood, fresh tissue, FFPE, and plasma cfDNA with key quality metrics and assay compatibility.

Study Design and Batch Management

Study design decisions made before sample collection have a greater impact on data quality than any downstream bioinformatics correction. Two issues deserve particular attention in human cohort methylation studies: choosing the right comparison structure and managing batch effects.

Comparison Structures

Case-control studies compare methylation between affected and unaffected individuals. They are the most common design because they require only one time point per participant and can draw from existing biobanks. The limitation is that cross-sectional measurements cannot distinguish cause from consequence.

Longitudinal studies measure methylation in the same individuals at multiple time points. Within-individual comparisons reduce genetic and environmental confounding and can establish temporal ordering. The trade-offs are higher cost and longer duration.

Nested case-control designs sample cases and controls from within an existing cohort using stored biospecimens. They combine the efficiency of case-control sampling with the temporal clarity of prospective cohorts.

Batch Randomization

Batch effects are systematic technical variation from processing samples across different reagent lots, operators, dates, or sequencing lanes. Methylation data is highly sensitive to batch effects. A sample processed on plate 1 may differ systematically from one on plate 2 in ways that mimic or obscure biological signals.

The critical rule: never confound batch with phenotype. If all cases are processed in batch 1 and controls in batch 2, biological and technical differences are inseparable. Randomizing samples so each batch contains a mix of cases and controls ensures batch effects can be removed statistically.

Practical measures include processing samples in randomized order across plates, recording batch variables for every sample, using a single sequencing center when possible, and avoiding mid-project protocol changes. Order effects from plate position or storage time can also accumulate across large cohorts, so extend randomization to the processing order within each batch.

Power and Replicate Planning

Statistical power in human cohort methylation studies depends on the expected effect size (Δβ — methylation difference between groups), methylation variance at each locus, sample size, and the significance threshold after multiple-testing correction.

For arrays, a 10-20% methylation difference (Δβ = 0.1-0.2) requires approximately 100-200 samples per group for 80% power at p < 1 × 10⁻⁷. A Δβ of 1-5%, typical for complex trait EWAS, needs 3,000-10,000 participants depending on per-CpG variance.

For sequencing-based methods, coverage depth partially compensates for sample-level variation. Most providers recommend 3-5 biological replicates per group for differential methylation analysis. ENCODE guidelines similarly recommend 2-3 replicates as a minimum with 4-5 preferred.

Cell-type-specific analysis can improve power. Purified cell populations reduce within-group methylation variance, effectively increasing statistical power without increasing sample size.

Recommended Replicate Minimums

Selecting the Right Methylation Assay

The assay choice converts the biological question into a technical workflow. For human cohort methylation studies, four assay categories cover the spectrum from genome-wide discovery to targeted validation.

Whole-Genome Bisulfite Sequencing

WGBS provides single-base resolution across the entire genome, capturing CpG, CHG, and CHH methylation contexts. At standard 30× coverage, each human sample requires approximately 1.5-2 billion reads, making it the most comprehensive and most expensive methylation assay per sample.

WGBS is the method of choice for discovery studies in small-to-medium cohorts (up to 100 samples total). It is particularly valuable when studying non-CpG methylation or when single-CpG resolution for DMR boundary detection is required.

Enzymatic Methyl-Seq

EM-seq replaces bisulfite conversion with enzymatic conversion using TET2 and APOBEC. The enzymatic approach is gentler on DNA, preserving library complexity and producing more uniform GC-rich coverage than WGBS.

EM-seq is the preferred option for low-input samples (below 100 ng), partially degraded DNA, or when GC bias is a concern. EM-seq libraries consistently show higher mapping rates and better coverage uniformity than WGBS libraries from the same input.

Reduced Representation Bisulfite Sequencing

RRBS uses MspI digestion to enrich for CpG-dense regions before bisulfite conversion, targeting approximately 5-10% of CpG sites with strong enrichment in CpG islands and promoters.

RRBS requires only 10-100 ng of input DNA and generates 10-20 million reads per sample, making it significantly more cost-effective than WGBS or EM-seq for large cohorts. The trade-off is limited coverage — it systematically misses intergenic and gene-body CpG sites. RRBS suits large studies (200+ samples) focused on promoter or CpG island methylation.

Methylation Arrays

Array-based profiling — typically the Illumina Infinium MethylationEPIC platform (935K CpG sites) — interrogates a fixed set of CpGs across gene regions, CpG islands, enhancers, and open chromatin. Arrays are the dominant EWAS platform because they are reproducible, require simple data processing, and scale cost-effectively.

For studies with 500+ samples, arrays offer the lowest per-sample cost among genome-scale methods and have mature analysis pipelines. The trade-off is genomic coverage — arrays assay about 3% of human CpG sites.

Targeted Bisulfite Sequencing

Targeted bisulfite sequencing focuses depth on pre-selected regions identified through prior discovery experiments. It provides very high coverage (500-1,000× or more) at specific loci, enabling detection of subtle methylation differences and allelic patterns.

Targeted approaches are the most cost-effective per sample when the regions of interest are known, making them practical for validation cohorts of hundreds to thousands of samples. Common enrichment methods include PCR-based amplification and hybrid capture.

Figure 3. Coverage, cost, and scalability trade-offs across five DNA methylation assay methods for human cohort studies.

Bioinformatics for Cohort Methylation Data

The bioinformatics pipeline for human cohort methylation data extends beyond standard alignment and methylation calling to address two challenges specific to cohort studies: cell-type heterogeneity and batch effects.

Cell-Type Deconvolution

Blood and tissue samples contain mixtures of cell types with distinct methylation profiles. Differences in cell composition between groups — common in case-control studies where disease correlates with immune cell proportions — can produce false-positive differential methylation signals.

Deconvolution methods estimate the proportion of each cell type per sample, then adjust the statistical model accordingly. Reference-based methods like Houseman's algorithm and CIBERSORT require external reference methylation profiles. Reference-free methods estimate latent cell-type components when reference profiles are not available. Including cell-type estimates as covariates is now standard in blood-based EWAS.

Batch Effect Detection and Correction

Batch effects should be assessed early in the pipeline. Principal component analysis of the most variable probes often reveals batch structure in early PCs. Visualization by processing date, plate, or operator can identify systematic shifts.

ComBat is the most widely used batch correction method for methylation data. It models batch-specific location and scale parameters to align batch means and variances. For arrays, the ChAMP and minfi Bioconductor packages include ComBat in their standard pipelines. Alternative approaches include modeling batch as a random effect in linear mixed models (limma, lme4) or using control-probe adjustment on array platforms.

Differential Methylation Analysis

For sequencing-based methods, differential methylation can be tested at single-CpG resolution (DMP analysis) or across genomic regions (DMR analysis). DMR analysis often has higher statistical power by sharing information across neighboring CpG sites. Tools including MethylKit, DSS, bsseq, and RnBeads support both approaches.

For array-based data, standard pipelines use limma for linear modeling and minfi for preprocessing. The significance threshold for EWAS is p < 1 × 10⁻⁷, corresponding to Bonferroni correction for approximately 850,000 tests.

Standard Pipeline Deliverables

A complete cohort methylation pipeline should deliver QC reports, normalized data, cell-type proportion estimates, differential methylation results with annotation, and publication-quality figures. Most providers include these as standard with optional modules for pathway enrichment and multi-omics integration.

FAQ

1) What is the minimum sample size for a human cohort DNA methylation study?

The minimum sample size depends on the expected effect size and the assay platform. For discovery-phase studies using arrays, a minimum of 100-200 samples per group is typically needed to detect moderate methylation differences. For sequencing-based studies, 3-5 biological replicates per group can suffice for discovery. Pilot studies can start with fewer samples but should acknowledge the statistical limitations.

2) Can I use FFPE samples for whole-genome methylation sequencing?

FFPE samples can be used for WGBS and EM-seq, but library complexity is lower than with fresh DNA, and the success rate depends on sample age, fixation conditions, and DNA quality. RRBS and methylation arrays are more reliable choices for FFPE samples. If WGBS or EM-seq is necessary, use FFPE-optimized library preparation kits and expect lower unique mapping rates.

3) How do I control for batch effects in a multi-year cohort study?

Randomize samples across batches so that each batch contains a mix of all experimental groups. Record batch variables (processing date, operator, reagent lot, sequencing lane) for every sample. Use statistical correction methods such as ComBat or linear mixed models during data analysis. Including reference control samples across all batches provides a direct measure of batch-to-batch technical variation.

4) What is the practical difference between WGBS and methylation arrays for cohort studies?

WGBS provides genome-wide coverage at single-base resolution but costs significantly more per sample and generates substantially more data per sample. Arrays are cost-effective at scale, have simpler data processing pipelines, and are the dominant platform for large EWAS. The choice depends on whether genome-wide discovery coverage is needed or whether the assay can focus on known regulatory regions.

5) Do I need to correct for cell-type composition in blood-based methylation studies?

Yes, cell-type composition adjustment is considered essential in blood-based methylation studies. Differences in immune cell proportions between comparison groups are common and can produce spurious differential methylation signals if not accounted for. Cell-type deconvolution should be included in the standard analysis pipeline for any blood-based cohort methylation study.

References

1. Tsai PC, Bell JT. "Power and sample size estimation for epigenome-wide association scans to detect differential DNA methylation." International Journal of Epidemiology. 2015; 44(4): 1429-1441.
2. Jiang Y, Chen J, Chen W. "Controlling Batch Effect in Epigenome-Wide Association Study." Methods in Molecular Biology. 2022; 2432: 73-84.
3. Liu X, Pang Y, Shan J, et al. "Beyond the base pairs: comparative genome-wide DNA methylation profiling across sequencing technologies." Briefings in Bioinformatics. 2024; 25(5): bbae440.
4. Guanzon D, Ross JP, Ma C, Berry O, Liew YJ. "Comparing methylation levels assayed in GC-rich regions with current and emerging methods." BMC Genomics. 2024; 25: 741.

Services mentioned in this article are provided for research use only and are not intended for clinical diagnosis, treatment, or personal health assessment.