Evaluation of the HumanMethylationEPIC v2.0 Bead Chip using low quality and quantity DNA samples

doi:10.21203/rs.3.rs-5917328/v1

Download PDF

Research Article

Evaluation of the HumanMethylationEPIC v2.0 Bead Chip using low quality and quantity DNA samples

https://doi.org/10.21203/rs.3.rs-5917328/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 21 Aug, 2025

Read the published version in Biological Procedures Online →

You are reading this latest preprint version

Background: The HumanMethylationEPIC v2.0 BeadChip (EPIC v2.0) microarray is a widely used tool for genome-wide DNA methylation (DNAm) analysis, designed for high-quality human DNA with a recommended input of 250 ng. However, in clinical and forensic settings, DNA samples may be of low quality and/or quantity (highly fragmented and/or available in low amounts). This study assessed the performance of the EPIC v2.0 on DNA samples with various combinations of average DNA fragment size (350, 230, 165, and 95 bp) and DNA input amount (100, 50, 20, and 10 ng), compared to a control sample analyzed under optimal conditions (high-quality DNA and 250 ng DNA input).

Results: The best performance was obtained for samples with average DNA fragment size of 350 bp and 100 ng DNA input (~90% probe detection rate, r = 0.995, and median absolute beta value differences |Δβ| = 0.012 when compared with the control sample). Samples with lower average DNA fragment sizes and DNA input amount performed worse, with the lowest probe detection rate (~43%), r = 0.946, and the highest |Δβ| (0.038). Samples with average DNA fragment sizes of 95 bp and those with 165 bp at 10 ng DNA input failed to pass sample quality control (QC). CpG sites with intermediate DNAm values (β = 0.1-0.9) showed higher |Δβ| than the extreme DNAm values (β= 0-0.1, and β = 0.9-1). Finally, we assessed an application of DNAm by performing epigenetic age analysis, and observed mean absolute errors (MAEs) below 10 years for 350 bp samples across four epigenetic clocks.

Conclusions: Both DNA fragment size and DNA input amounts affect DNAm analysis on the EPIC v2.0, with the investigated DNA fragment size having a greater impact than the investigated DNA input amount. DNAm measurements were achieved with the EPIC v2.0 microarray down to an average DNA fragment size of 165 bp and a 20 ng DNA input. Highly fragmented DNA (95 bp) did not result in usable DNAm analysis as all samples failed QC. Overall, our study demonstrates the potential and limitations of EPIC v2.0 microarray with low quality and quantity DNA samples.

DNA methylation

EPIC v2.0 microarray

fragmented DNA

low-input DNA

clinical epigenetics

forensic genetics

epigenetic age

degraded DNA

DNA methylation (DNAm) is an epigenetic modification that regulates gene expression and is implicated in various biological processes and diseases [1, 2]. DNAm is mainly found on cytosine bases that are followed by a guanine (CpG sites). Various technologies can measure DNAm, with whole-genome bisulfite sequencing and DNAm microarrays being the most widely used for genome-wide investigations. Microarray-based approaches offer a good compromise between genome-wide coverage and affordability [3]. Several microarrays are available for DNAm analysis, including the Agilent SurePrint Methylation Microarray, Affymetrix GeneChip Human Promoter Array, and Illumina HumanMethylationEPIC BeadChip microarray. These arrays have been used to generate data from large cohorts such as The Cancer Genome Atlas (TCGA) [4] and the Accessible Resource for Integrated Epigenomics Studies (ARIES) [5] to investigate environmental exposures [6, 7], ageing [8], and various diseases [9–11].

Illumina developed multiple versions of the Infinium Human Methylation BeadChip microarray over time. The latest version, the HumanMethylationEPIC v2.0 BeadChip (EPIC v2.0), targets ~ 935,000 CpG sites, with 77.63% homology to its predecessor (EPIC v1.0) and over 200,000 new CpGs located in enhancers and open-chromatin regions [12]. This new and improved release contains fewer probes subjected to cross-reactivity and sequence polymorphism issues and introduces probes targeting cancer somatic mutations [13]. The EPIC v2.0 microarray requires high-quality DNA samples, and high DNA quantities are recommended (250 ng input). In a clinical and forensic setting, it may be essential to be able to work on fragmented and/or low-quantity DNA material. For example, circulating free DNA (cfDNA), also known as cell-free DNA, consists of small circulating fragments of DNA (median size of 167 bp) present in plasma, saliva, urine, and seminal fluid [14]. cfDNA is released by cells due to cell death and can be considered as a marker of inflammation and tissue injury. It has been used as a noninvasive diagnostic and disease-monitoring tool in oncology [15, 16] and autoinflammatory conditions [17–19]. Additionally, cfDNA methylation patterns could be used to localize the tissue of origin [20] or to detect cancer [21]. Another source of degraded DNA in clinical diagnostics is formalin-fixed, paraffin-embedded (FFPE) tissue, where DNA is often fragmented and present in low quantities, making DNAm analysis challenging [22].

In forensic genetics, researchers aim to extract as much information as possible from minute DNA traces, including the estimated chronological age of a donor using DNAm. Illumina methylation microarrays have been explored as a potential tool for DNAm age prediction [23–25]. However, forensic traces are often exposed to harsh environmental conditions, leading to poor DNA quality and low yields. This makes the applicability of DNAm arrays in forensic casework challenging if the recommended input of 250 ng DNA must be adhered to. Also, even more degraded samples could be found in ancient DNA material, where DNA is exposed to harsher conditions for much longer periods of time. In recent years, the interest in DNAm analysis for ancient DNA has grown [26–28], with DNAm array technologies successfully applied to infer age at death and cell composition in mummified human remains from the eighteenth century, using EPIC microarray DNA inputs of 320 and 220 ng [29].

These fields would benefit from a detailed investigation into the potential and limitations of the EPIC v2.0 microarray technology when applied to DNA samples of low quality and quantity. Previously, DNA input of less than 250 ng yielded accurate results for the EPIC microarrays [13, 30], but so far there is no study systematically assessing how the combination of low DNA quality (fragmented DNA) and quantity (input amount) influences the reproducibility, precision, and applicability of the EPIC v2.0 microarray. In this study, we measured genome-wide DNAm in duplicate using the EPIC v2.0 microarray and samples with different average DNA fragment sizes (350, 230, 165, and 95 bp) and input amounts (100, 50, 20, and 10 ng). We assessed data quality, probe detection rate, reproducibility, precision, and absolute methylation (beta) value differences (|Δβ|) compared to a control DNA sample. Additionally, we assessed the impact of these parameters on a key DNAm application: DNAm age prediction.

Study design and DNA sample preparation

Figure 1 shows the design of the study. Peripheral blood from a male individual was collected in a EDTA whole blood collection tube and stored at −80°C until DNA extraction. Genomic DNA extraction was performed using the QIAamp DNA Blood Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer’s guidelines. Extracted genomic DNA was quantified using the Qubit™ dsDNA HS Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA) following the manufacturer’s recommendations. Qubit measurements were used to prepare a DNA solution containing 2,000 ng in 210 µL. This DNA sample was equally split into four different tubes. Each tube was subjected to DNA fragmentation with the Covaris S220 instrument (Covaris, Woburn, MA, USA) for time periods of 10, 14, 21, and 113 min to achieve DNA fragment sizes of ∼350, ∼230, ∼165, and ∼95 bp, respectively. To confirm the obtained fragment sizes, fragmented DNA samples were analyzed using the High Sensitivity DNA Kit for 2100 Bioanalyzer Systems (Agilent Technologies, Santa Clara, CA, USA), according to manufacturer’s instructions. Average DNA fragment sizes were calculated using the 2100 Expert Software package (Agilent Technologies, Santa Clara, CA, USA) and setting a range from 38 to 9,330 bp. Degradation Indexes (DIs) were obtained by quantifying fragmented DNA samples with Quantifiler™ Trio DNA Quantification kit (Thermo Fisher Scientific, Waltham, MA, USA) using the Applied Biosystems™ 7500 real-time PCR system (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer’s protocol. DIs were plotted against average DNA fragment size using ggplot2 R-library (version 3.4.4) [31]. From each of the fragmented DNA extracts, we prepared four different dilutions with DNA amounts of 100, 50, 20, and 10 ng in duplicate.

DNA methylation profiling

Genomic DNA samples were bisulfite-converted using the EZ DNA Methylation™ Kit (Zymo Research, Irvine, CA, USA) according to the manufacturer’s instructions, with a final elution volume of 10 µl. Genome-wide DNAm levels were measured using the Infinium MethylationEPIC kit v2.0 (Illumina, San Diego, CA, USA), following the manufacturer’s protocol. Samples were randomly distributed on the EPIC bead chips and chip positions (rows 1-8). The arrays were scanned using the iScan™ System (Illumina, San Diego, CA, USA) to obtain raw Intensity Data files (.idat).

Data analysis was carried out in the R statistical environment (version 4.3.1) [32] with packages from the Bioconductor project (version 3.18). Raw .idat files were imported into R and processed using the Sensible Step-wise Analysis of DNA Methylation BeadChip (SeSAMe) package, version 1.20.0 [33]. DNAm levels (β-values), ranging from 0 to 1, and quality metrics were obtained using the SeSAMe function openSesame(), which provides end-to-end processing that converts IDATs to DNA methylation levels. In brief, the selected pre-processing functions qualityMask(), inferInfiniumIChannel(), dyeBiasNL(), pOOBAH(), and noob() masked the probes of poor-quality design (32,896 probes), inferred the channel of Infinium-I probes, corrected for dye bias, removed probes with low-quality signal (detection p-value ≥ 0.05), and implemented a background subtraction. Quality control (QC) was performed using the standard SeSAMe QC quality metrics. More specifically, we assessed the relationship between red and green background intensities, median intensities, the ratio between red and green median intensities, probe detection rate (calculated as ), β-value distributions, and Principal Component Analysis (PCA). PCA was used to assess the overall variability of the data. Samples that did not meet the following criteria were removed from downstream analysis: i) Median signal intensity < 2000; ii) Ratio of red and green median signal intensity < 0.5 or > 2; iii) deviation from bimodal β-value distribution shape. Data from the 65 single nucleotide polymorphism (SNP) probes included in the EPIC v2.0 microarray were used to verify the identity of the sample donor and to assess the impact of degraded samples on EPIC v2.0 genotyping. All plots were generated using the ggplot2 R-library (version 3.4.4) [31].

DNAm data analysis

To investigate the impact of fragmented and low-input DNA samples (hereafter referred to as degraded samples) on reproducibility, precision, and β-value estimation of the EPIC v2.0 microarray, we calculated: 1) the Pearson correlation coefficient (r) between sample replicates, 2) r between degraded samples and control sample, and 3) the absolute differences in beta values (|∆β|) compared to the control sample and between sample replicates. To investigate the precision of the EPIC v2.0 microarray in degraded samples, we calculated within-replicate correlations, r, using the shared CpGs between replicate pairs. To assess the EPIC v2.0 microarray reproducibility we calculated pairwise correlations between degraded samples and the control sample using whole-array r on β-values with the cor() function in R. To ensure comparability of r between degraded samples, missing values were removed using the na.omit() function, resulting in a total of 202,439 common CpGs. For each combination of DNA fragment size and input amount, average DNAm values between the two replicates were used to calculate r. P-values were adjusted for multiple testing using the r function p.adjust() using the Benjamini-Hochberg method [34] to control the False Discovery Rate (FDR) at a threshold of p = 0.05. Additionally, we calculated the absolute differences in beta values (|∆β|) of the 202,439 common CpGs between the average DNAm value obtained for the degraded DNA samples that passed QC and the one of the control sample. We calculated median, interquartile range (IQR), and percentage of CpG sites with |∆β| ≥ 0.05 and 0.1 from |∆β| distributions. Additionally, we investigated |∆β| in CpGs using β-value intervals of 0.1 (0.0-0.1, 0.1-0.2, 0.2-0.3, 0.4-0.5, 0.5-0.6, 0.6-0.7, 0.7-0.8, 0.8-0.9, and 0.9-1.0). Within-replicate |∆β| statistics (median, IQR, |∆β| ≥ 0.05 and 0,1) were also calculated using the shared CpGs between the replicate pair.

Information related to the type of probe (type I and II) and genomic location of each CpG site (island, shore, shelf, and open sea) was extracted from the IlluminaHumanMethylationEPICv2anno.20a1.hg38 Bioconductor package [35]. Gene location (gene body, intergenic, and promoter regions) for each CpG was obtained from EnsDb.Hsapiens.v111 Bioconductor package [36].

Age prediction of degraded samples

To evaluate the impact of DNA fragment size and input amount on a DNAm application, we calculated DNAm-based age for all degraded samples using the function DNAmAge() from methylclock Bioconductor package (version 1.8.0) [37]. We predicted the chronological age of the samples’ donor using four different epigenetic clocks: the multi-tissue age estimator developed by Horvarth et al. [38] (Horvath clock), which contains 353 CpGs; the skinHorvath clock [39], with 391 CpGs; the BLUP clock [40], containing 319,607 CpGs; and the EN clock [40], with 514 CpGs. To assess age prediction accuracy on degraded samples, we calculated the mean absolute error (MAE) by averaging the absolute differences between predicted and chronological ages across the two sample replicates.

We investigated the performance of the Infinium MethylationEPIC v2.0 microarray (EPIC v2.0) using a combination of fragmented (average sizes of 350, 230, 165, and 95 bp) and low-input (100, 50, 20, 10 ng) DNA material. Supplementary Figure 1 shows the obtained fragment size distributions of the four DNA sample sizes. These distributions are represented by Gaussian-like curves with elongated tails extending towards longer DNA fragments (right skewed), with shorter DNA fragment samples exhibiting a narrower distribution. The obtained DIs for the DNA samples with average fragment sizes of 350, 230, 165, and 95 bp were 2.09, 3.44, 8.88, and 8008.45, respectively. Supplementary Figure 2 illustrates the relationship between the average DNA fragment size and DI.

Quality control of degraded DNA methylation data

It was key to carry out a thorough QC procedure of the Illumina EPIC v2.0 microarray data to ensure reliable results given that samples of suboptimal DNA quality and quantity were employed.

Illumina EPIC v2.0 microarray utilizes beads with multiple copies of 50 bp oligonucleotide probes targeting specific loci in the genome. Once a DNA fragment hybridizes with its complementary probe, single-base extension (SBE) incorporates labeled nucleotides (green or red), which emits a signal. The intensities of the signals from probes targeting a given CpG site provide information about its DNAm level. Non-specific binding of negative control probes and fluorescence intrinsically associated with the microarray surface generate background signal intensities, which could interfere with the analysis if not properly removed. Figure 2A shows the relationship between the two mean background signal intensities (green and red) across all samples. Degraded samples with average DNA fragment sizes of 350 bp, 230 bp, and 165 bp (except for the 10 ng input) clustered near the high-quality DNA control sample (grey dot), with background signal intensities above 218. Degraded samples with an average DNA fragment size of 95 bp and 165 bp with 10 ng DNA input had at least one background signal intensity close to zero, indicating potential issues with QC. Figure 2B shows the mean signal intensities of all probes across all samples. The control sample exhibited a mean signal intensity of approximately 5,000, which was also observed in samples with DNA fragment sizes of 350 bp and 230 bp. For samples with a DNA fragment size of 165 bp, we observed a progressive decrease in mean signal intensity with decreasing DNA amounts. The low mean background signal intensity observed for degraded samples with a DNA fragment size of 95 bp and 165 bp with 10 ng DNA input was desirable, however, the mean signal intensity was low, which indicates poor DNA hybridization. Another important QC metric is the ratio of median red to green signal intensities, that represents the relative balance between the two fluorescent dye channels (green and red) used in the microarray. A balanced ratio (close to 1) indicates good microarray performance, while an imbalanced ratio suggests potential technical issues. When we compared the ratio of median red to green signal intensities, we observed a deviation from a ratio of 1 in samples with DNA fragment sizes of 95 bp and 165 bp with a DNA input of 10 ng (Figure 2C). Additionally, we assessed the genotyping performance of the 65 SNP probes included in the EPIC v2.0 microarray for sample matching (Supplementary Table 1). When comparing the SNP genotypes of the degraded samples with the control sample, we observed an increase in discordant genotype calls with decreasing length of DNA fragment size and lower DNA input. Particularly, more than half of the SNP genotypes were discordant in samples with an average fragment size of 165 bp at 10 ng input, as well as samples with DNA fragment size of 95 bp.

We performed PCA using beta values of the 6,160 CpGs that were shared among all samples (including the samples that failed QC) (Figure 2D) to check for outliers, batch effect, and degradation patterns (DNA fragment size and input amount) in the data. We observed that the control sample and the samples with average fragment sizes higher than 95 bp (except for the combination of 10 ng input and average fragment size of 165 bp DNA) clustered together, while the other samples were scattered across the PCA plane. PC1 and PC2 represented 44.41% and 19.12% of the total variance in the data, respectively, indicating that average DNA fragment size and input amount were the primary drivers of the variance. We also examined the β-value distribution in all samples. Degraded samples that clustered together with the control sample showed the expected bimodal β-value distribution shape (Supplementary Figure 3). In contrast, degraded samples with average fragment sizes of 95 bp and 165 bp (with a DNA input of 10 ng) presented a deviation from the bimodal β-value distribution shape in both replicates. Based on these results, we excluded all DNA samples with an average DNA fragment size of 95 bp, and DNA samples with an average DNA fragment size of 165 bp and a DNA input of 10 ng. Thus, 23 out of 33 samples, including both control and degraded samples, passed QC and were included in subsequent analyses. Looking at the corresponding PCA plot considering the 202,439 common CpGs (Figure 2E), degraded samples with an average DNA fragment size of 350 bp clustered together with the control sample, whereas samples with DNA fragment size of 230 bp and 165 bp were more spread out, indicating a different degradation pattern. PC1 and PC2 represented 20.25% and 8.94% of the total variance, respectively. Finally, considering the probe detection rate, 929,306 probes (99.1%) passed filtering for the control sample, whereas a decrease in probe detection rate was observed with both shorter DNA fragment size and lower DNA input (Figure 2F). Degraded samples with an average DNA fragment size of 350 bp, whereas the lowest probe detection rate (42.63%) and the smallest number of probes passing filtering (399,760) were found for the degraded samples with DNA fragment size of 165 bp and 20 ng input amount. We also investigated the failing probes across the different DNA fragment sizes and DNA input amounts. The SeSAMe pipeline masked 32,896 poorly designed probes, which were therefore excluded from this analysis. A total of 6,066 failing probes were common between all the samples passing QC (including the control sample), however, when considering only the degraded samples, we obtained 57,910 commonly failing probes. We calculated the number of common failing probes for each DNA fragment size and input amount. A total of 71,820, 132,622, and 173,396 common failing probes were detected in samples with average DNA fragment sizes of 350, 230, and 165 bp, respectively. Regarding DNA input amounts, we obtained 67,563, 93,651, 130,661, and 166,795 shared failing probes for input amounts of 100, 50, 20, and 10 ng, respectively (Supplementary Figure 4A-B). Lists with the names of the commonly failing probes across all samples, all degraded samples, DNA fragment sizes, and DNA input amounts are shown in Supplementary File 1.

Degraded DNA methylation data correlations and absolute differences

To investigate the impact of low quality and quantity DNA samples on EPIC v2.0 microarray reproducibility, we calculated the within-replicate correlation (precision) for each combination of average DNA fragment size and input amount (Supplementary Table 2). Degraded samples with an average DNA fragment size of 350 bp and DNA input of 100 ng had within-replicate r value of 0.990 (804,732 shared CpGs). We observed a progressive decrease in r values and number of shared CpGs with shorter DNA fragment sizes and lower DNA inputs. The lowest within-replicate r value (0.851) and number of shared CpGs (259,031) was observed in degraded samples with average DNA fragment size of 165 bp and DNA input of 20 ng. All correlation tests had an adjusted p-value below 0.05. We also assessed the correlation between degraded samples and control sample (reproducibility) for each combination of average DNA fragment size and input amount. Figure 3 shows the r values obtained from 202,439 common CpGs among all samples passing QC and the control sample. First, the correlation between the control sample and sample with average DNA fragment size of 350 bp and DNA input of 100 ng was r = 0.995. For the rest of the degraded samples, we observed a progressive decrease in r values with shorter DNA fragment sizes and lower DNA inputs. The lowest correlation value, r = 0.946, was observed in sample with an average DNA fragment size of 165 bp and DNA input of 20 ng. A drop in r values (from 0.995 to 0.989) was observed in degraded samples with DNA fragment size of 350 bp when reducing the DNA input to 20 ng. Degraded samples with DNA fragment size of 165 bp showed a larger r value drop (from 0.971 to 0.946) (Supplementary Figure 5).

Furthermore, we aimed to estimate the deviation of β-values from the control sample in all degraded samples that passed QC. Quantifying the impact of sample degradation on DNAm measurement could provide a criterion for determining whether to proceed with DNAm analysis. Therefore, we calculated the absolute difference of β-values (|∆β|) between the control and degraded samples, using mean β-values between sample replicates that passed QC. A distribution of |∆β| is generated for each pair of degraded samples (Figure 4). All distributions showed a peak close to 0 and elongated tails towards higher values of |∆β|. We observed a lower median |∆β| and percentage of probes with a |∆β| ≥ 0.05 and 0.10 for samples with larger average DNA fragment sizes and higher DNA inputs (Supplementary Table 3). Samples with average DNA fragment size of 350 bp and DNA input amount of 100 ng had a median |∆β| of 0.016 and 1.95% of probes with a |∆β| ≥ 0.10. While samples with average DNA fragment size of 165 bp and DNA input amount of 20 ng had a median |∆β| of 0.051 and 31.70% of probes with a |∆β| ≥ 0.10. We further investigated |∆β| in β-value intervals of 0.1 to comprehend whether EPIC v2.0 measurements of CpG sites with intermediate β-values were more affected by low quality and quantity DNA samples. We observed higher |∆β| values for intermediate β-values intervals (0.1-0.9) compared to the extreme β-values intervals (0.0-0.1 and 0.9-1.0) in all DNA fragment sizes investigated (Supplementary Figure 6). Samples with shorter average DNA fragment sizes and lower DNA inputs exhibited higher |∆β| in intermediate β-value intervals, while extreme β-values intervals were less affected by the quality and quantity of the DNA sample. We also calculated |∆β| between sample replicates and estimated median |∆β| and percentage of probes with a |∆β| ≥ 0.05 and 0.10. We observed an increase of these parameters with shorter average DNA fragment sizes and lower DNA inputs (Supplementary Table 4). Degraded samples with average DNA fragment size of 350 bp and DNA input of 100 ng had median |∆β| of 0.018 and percentage of probes with |∆β| ≥ 0.10 of 0.068 (804,732 shared CpGs). While for the sample with average DNA fragment size of 165 bp and DNA input of 20 ng, the median |∆β| was 0.03 and 0.23% of probes had |∆β| ≥ 0.10 (259,031 shared CpGs).

Then, we investigated the type of probes and target regions that were more robust in low quality and quantity DNA samples. Illumina EPIC v2.0 microarray uses two types of oligonucleotide probes, Type I probes target methylated and unmethylated epialleles of a CpG site using two different probes, while Type II probes use a single probe able to target both methylated and unmethylated epialleles.

We calculated the proportion of Type I and II probes in probes passing filtering for each DNA sample that passed QC. The relative proportion of Type I probes that passed filtering increased in samples with shorter average DNA fragment sizes and lower DNA input (Supplementary Figure 7A), which indicated that these probes were more efficient than the Type II probes. We also assessed the genomic (CpG island, shore, shelf, and open sea regions) and gene (promoters, gene body, and intergenic regions) distribution of probes passing filtering (Supplementary Figure 7B-C). We observed an increase in the proportion of successfully typed probes targeting CpG islands and gene promoters in samples with shorter average DNA fragment sizes and lower DNA input when compared to the control sample.

Age prediction of degraded samples

Finally, we wanted to test the impact of DNA fragment sizes and DNA inputs on a DNAm application. We chose to estimate DNAm age because it is a potential analysis of interest in the medical and forensic fields. We calculated MAE for the control and degraded samples using four epigenetic clocks: BLUP (319,607 CpGs), EN (514 CpGs), Horvath (353 CpGs), and skinHorvath (391 CpGs) clocks. The control sample had an absolute error (AE) of 0.19, 1.83, 8.77, and 0.21 years for BLUP, EN, Horvath, and skinHorvath clocks, respectively. For the degraded samples, we obtained higher MAE in all samples, especially for the BLUP, EN, and skinHorvath clocks (Figure 5). We observed a general increase in MAE with shorter DNA fragment size and lower DNA input amounts in all epigenetic clocks tested, indicating an impact of sample degradation on DNAm age prediction analysis. Samples with average DNA fragment size of 350 bp had a MAE <10 years for all DNA inputs and epigenetic clocks, except for the age predicted with the Horvath clock in 20 ng DNA input samples. In contrast, degraded samples with average DNA fragment size of 165 had MAE > 10 years, except for age predicted from 20 ng DNA input samples with the BLUP clock and the 50 ng DNA input sample with the skinHorvath clock. Supplementary Table 5 shows the number of missing CpGs in each epigenetic clock across all samples that passed QC. As expected, the number of missing CpGs in the four epigenetic clocks increased with shorter DNA fragment size and lower DNA input amounts.

The HumanMethylationEPIC v2.0 BeadChip is the latest update in the Illumina DNAm microarray series. The EPIC microarray has been successfully employed in clinical and forensic contexts to study DNAm patterns using high-quality and quantity DNA samples [23, 24, 41, 42]. However, clinical and forensic DNA samples often encounter suboptimal conditions, such as fragmented and/or scarce DNA. Here, we comprehensively investigated the EPIC v2.0 microarray’s performance on human DNA samples of varying fragment sizes (350, 230, 165, and 95 bp) and input amounts (100, 50, 20, and 10 ng). Overall, the EPIC v2.0 microarray maintained high reproducibility (r > 0.94 with control sample), precision (r > 0.86 between sample replicates), and probe detection rate > 42.6% (~ 399,760 probes passing filtering) for degraded samples down to an average DNA fragment size of 165 bp with a DNA input of 20 ng. However, as expected, we observed a progressive decrease in reproducibility, precision, probe detection rate, and age prediction accuracy as DNA fragment sizes and input amounts decreased, with DNA fragment size having a greater impact than input amount. This was suggested by the larger drop in reproducibility with shorter DNA fragment size and the QC failure of highly fragmented DNA samples (95 bp).

The analysis of EPIC v2.0 microarray data necessitates upstream QC to ensure robust and reproducible results; hence, samples and probes that do not meet QC standards must be excluded from downstream analyses. Key QC metrics for sample inclusion are probe detection rates, background and median signal intensities, red-to-green median signal intensity ratios, bisulfite conversion rates, β-value distributions, sex and identity matching, and PCA [43, 44]. In our study, degraded samples with an average DNA fragment size of 95 bp failed QC since they displayed altered signal intensities and β-value distributions despite having probes that initially passed filtering criteria. We observed altered β-value distributions, particularly a shift towards intermediate β-values. Consistent with our findings, Lee et al. [45] reported similar skewing of β-value towards intermediate values in suboptimal samples, such as those from FFPE tissues and cfDNA. This trend was attributed to the low DNA input, resulting in reduced overall signal intensities. Our data showed that short DNA fragments, combined with low DNA input, may drive the shift towards intermediate β-values. This phenomenon likely reflects the limited availability of DNA molecules for probe hybridization. EPIC v2.0 microarray technology relies on beads bound to 50 bp oligonucleotides that hybridize with specific bisulfite-converted DNA regions. Technically, DNA fragments as small as 95 bp should be successfully targeted by these probes. However, the initial step of the Infinium array protocol, bisulfite conversion, is a harsh chemical reaction necessary to distinguish methylated from unmethylated cytosines, which causes substantial DNA damage. We propose that the failure of 95 bp DNA samples in the EPIC v2.0 analysis is likely due to further DNA degradation caused during the bisulfite conversion step, which significantly reduces the pool of DNA fragments available for hybridization by the EPIC v2.0 probes. This hypothesis is further supported by our observation that degraded samples with 95 bp DNA fragment size and relatively high input amounts (100 and 50 ng) did not yield reliable data. Reducing the harshness of bisulfite treatment (such as lower temperatures or shorter incubation times) [46] or using enzyme-based conversion methods [45, 47] could potentially enhance the analysis of highly fragmented DNA samples and improve the overall detection of DNAm measurement.

Additionally, although Illumina recommends a DNA input of 250 ng for EPIC v2.0 analysis, several authors assessed the EPIC v2.0 performance using low-input DNA samples [30, 45, 48, 49]. Kaur et al. [48] obtained valuable data from as little as 1 ng DNA input, though probe detection rates (still exceeding 50%) and reproducibility (Spearman’s rank correlation of 0.918) decreased with lower DNA input amounts. Our findings align with these observations, showing a decrease in probe detection rate, reproducibility, and precision with lower DNA input amounts and shorter DNA fragment sizes. Nevertheless, the deviation of these parameters with lower DNA input amount should not only be attributed to EPIC v2.0 performance but also by the intrinsic nature of DNAm level calculation. The DNAm level represents the proportion of methylated cytosines in a sample, and stochastic variations of DNA molecule sampling can lead to fluctuation in the analysis of DNAm levels in samples with low DNA input amounts. Naue et al. [50] reported that a DNA sample of 10 ng, representing 2,785 copies prior to bisulfite conversion, had a 95% confidence interval for Δβ relative to the true DNAm value of ± 0.019. We expect that part of the |Δβ| observed in low input DNA samples (10 ng) in this study reflects the inherent stochastic variations in DNA molecule sampling rather than only the performance of EPIC v2.0 microarray. This deviation from the real DNAm in a low input DNA sample cannot be reduced by using other DNAm measurement technologies, but only by increasing DNA input amounts. When performing analyses of differentially methylated position (DMP) and region (DMR) the commonly used threshold between the two sample groups (e.g., cases/controls or exposed/non-exposed individuals) is a |Δβ| of 0.1. In our study, we observed an increase in the percentage of CpG sites where |Δβ| ≥ 0.1, from 1.95% in samples with average DNA fragment size of 350 bp and a DNA input amount of 100 ng to 31,70% in samples with average sizes of 165 bp and 20 ng DNA input. This significant increase in CpG sites with a |Δβ| ≥ 0.1 indicates that DNA fragment size and input amount could impact DMP and DMR analysis. These variables should be taken into consideration when performing DMP and DMR analysis by matching the two sample groups or by including these variables as covariates. Furthermore, we noted that DNAm measurement at CpG sites with intermediate DNAm value (0.1–0.9) were more affected by DNA fragment size and lower input amount as measured by |∆β|. Higher |∆β| in intermediate DNAm values were previously observed in other studies when assessing EPIC microarray in low-input DNA samples [30] and tissues stored in suboptimal conditions [51]. Eventually, an enrichment of Type I probes and probes targeting CpG island regions was observed in degraded samples. A significant proportion of Type I probes target CpG-rich regions [52], making it challenging to determine whether this enrichment was inherent to the Type I probes or if CpG island regions themselves are more stable to induced fragmentation.

DNAm age inference represents one of the most promising and extensively researched applications within the field of epigenetics. In the medical context, DNAm age was applied to predict lifespan (time-to-death due to all-cause mortality) [53–55], but also to better understand the impact of lifestyle and environment on aging [56]. While in the forensic field, inferring the age of an unknown sample donor at a crime scene can provide crucial leads in police investigations to narrow down the suspect pool. Our analysis revealed that array-based DNAm age prediction is feasible when DNA samples have average DNA fragment sizes of 350 bp (DI = ~ 2.1) and a DNA input amount of 100 ng is used. However, when the average DNA fragment size is 165 bp (DI = ~ 8.88) or 95 bp (DI = ~ 8008.4), accurate age prediction was not possible. We believe that a combination of missing data and larger |∆β| is the reason of the lower age prediction accuracies observed in degraded samples with short DNA fragment size and low DNA inputs. Lee et al.[23] applied EPIC microarray technology in forensics to predict the age of an individual from fresh, fragmented, and low-input DNA samples. However, fragmented and low-input DNA were tested separately. In our study, we demonstrated that the combination of these two variables could alter DNAm measurement more significantly than each variable alone. We recommend a thorough evaluation of the combined effect of fragmented and low-input DNA samples on forensic estimations before applying EPIC v2.0 technology in forensic case work.

DNAm analysis of cfDNA in plasma or other body fluids could provide valuable information about tumor diagnosis, monitoring and prognosis [57]. cfDNA is retrieved in limited amounts from both healthy individuals and cancer patients. Additionally, cfDNA typically presents a peak at 167 bp for healthy individuals and 143 bp for cancer patients in the distribution of DNA fragment size [58]. Analysis of DNAm in cfDNA provides a dual challenge due to the low DNA input and short fragment size. Although cfDNA was not directly used in this study, our analysis indicates that EPIC v2.0 technology could yield altered DNAm measurement when analyzing DNA fragment sizes of 165 bp using low input amounts. Therefore, we propose to increase DNA input if the DNA is very fragmented. Our evaluation also has implications for the analysis of FFPE samples, which are primarily used for histological examinations but also for long-term tissue preservation. FFPE-derived DNA often exhibits low quality, which can impact DNAm analysis with EPIC arrays [59]. Dupont et al. [59] observed that FFPE samples had a mean DI = 2.51, close to the DNA samples with average DNA fragment size of 350 bp (DI = 2.09) used in this study, and that DI was correlated with EPIC array probe detection rate [60]. This correlation was observed also in our study, and we suggest that the DI parameter could be used as a preliminary quality screening tool to perform EPIC v2.0 analysis and as a predictor of probe detection rate.

Finally, there are a few limitations in our work. Our dataset was composed of DNA samples coming from a single individual and the investigated average DNA fragment size was limited to the range of 350 to 95 bp. Furthermore, the distribution of the four average DNA fragment sizes showed slight variations, with samples with average DNA fragment size of 350 bp having broader peak compared to those with average size of 95 bp samples. DNA samples used in this study were artificially fragmented. The starting DNA used in this study was of high-quality, clinical and forensic samples could present further layers of complexity such as DNA alterations/damage (e.g. chemical crosslinking in FFPE samples [61]) or different DNA fragment size distributions. These differences could make DNAm analysis more challenging and prevent a direct translation of our results to forensic and clinical scenarios. Future work should focus on investigating wider DNA fragment sizes, clinical and mock forensic degraded samples, and alternative experimental and computational approaches to improve the detection of DNAm in suboptimal samples. Lee et al. [45] proposed different bisulfite conversion kits, elution volumes, input volumes for the array, and probe detection calling pipeline for low-input DNA samples. These alternative workflows could be applied to fragmented and low input DNA samples in future studies.

DNA fragment size and input amount affect DNAm measurements as analyzed with the EPIC v2.0 microarray, with DNA fragment size having a greater impact than DNA input amount. Reliable DNAm data could be obtained down to an average DNA fragment size of 165 bp with 20 ng input amount. However, reproducibility, precision, probe detection rate, and age prediction are compromised by shorter DNA fragment size and input amounts. When dealing with degraded samples, we recommended evaluation of DNA quality and quantity before conducting DNAm analysis with the EPIC v2.0 microarray. Overall, we showed the potential and limitations of the EPIC v2.0 microarray with low quality and quantity DNA samples.

EPIC v2.0: HumanMethylationEPIC v2.0 BeadChip; DNAm: DNA methylation; QC: quality control; |Δβ|: absolute beta value difference; TCGA: The Cancer Genome Atlas; ARIES: Accessible Resource for Integrated Epigenomics Studies; cfDNA: circulating free DNA; FFPE: Formalin-fixed, paraffin-embedded; Dis: Degradation Indexes; SeSAMe: Sensible Step-wise Analysis of DNA Methylation BeadChip; PCA: Principal Component Analysis; SNP: Single Nucleotide Polymorphism; R: squared Pearson correlation coefficients; FDR: False Discovery Rate; IQR: interquartile range.

Ethics approval and consent to participate

The study conformed to the Declaration of Helsinki and was approved by the Committees of Health Research Ethics in the Capital Region of Denmark (H-22034131). Informed written consent was collected from the individual. Patient material and data were pseudonymized.

Consent for publication

Consent for publication was collected from the individual and included in the informed written consent.

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. The R script used for data analysis in this study is publicly available at https://github.com/BrandoPoggiali/EPIC_v2.0_degraded_samples_paper.git.

Competing interests

The authors declare no competing interests.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Acknowledgements

We would like to thank Anja Ladegaard Jørgensen for assistance with this work.

Author’ contributions

B.P.: designing the experiment, formal analysis, investigation, data curation, and writing original draft. M.E.D.: designing the experiment, investigation, reviewing and editing the manuscript. M.L.K: designing the experiment, reviewing and editing the manuscript. A.V.: reviewing and editing the manuscript. C.B: supervision, reviewing and editing the manuscript. V.P.: supervision, reviewing and editing the manuscript. J.T.H.: reviewing and editing the manuscript, funding acquisition. J.D.A.: designing the experiment, investigation, supervision, and funding acquisition. All authors approved the final version of the manuscript.

Dhar GA, Saha S, Mitra P, Nag Chaudhuri R. DNA methylation and regulation of gene expression: Guardian of our health. The Nucleus. 2021 Dec 16;64(3):259–70.
Robertson KD. DNA methylation and human disease. Nat Rev Genet. 2005 Aug;6(8):597–610.
Bibikova M, Le J, Barnes B, Saedinia-Melnyk S, Zhou L, Shen R, et al. Genome-Wide Dna Methylation Profiling Using Infinium ^® Assay. Epigenomics. 2009 Oct;1(1):177–200.
Weisenberger DJ. Characterizing DNA methylation alterations from The Cancer Genome Atlas. Journal of Clinical Investigation. 2014 Jan 2;124(1):17–23.
Relton CL, Gaunt T, McArdle W, Ho K, Duggirala A, Shihab H, et al. Data Resource Profile: Accessible Resource for Integrated Epigenomic Studies (ARIES). Int J Epidemiol. 2015 Aug;44(4):1181–90.
Plusquin M, Guida F, Polidoro S, Vermeulen R, Raaschou-Nielsen O, Campanella G, et al. DNA methylation and exposure to ambient air pollution in two prospective cohorts. Environ Int. 2017 Nov;108:127–36.
Lai CQ, Parnell LD, Lee YC, Zeng H, Smith CE, McKeown NM, et al. The impact of alcoholic drinks and dietary factors on epigenetic markers associated with triglyceride levels. Front Genet. 2023 Feb 15;14.
Dhingra R, Kwee LC, Diaz-Sanchez D, Devlin RB, Cascio W, Hauser ER, et al. Evaluating DNA methylation age on the Illumina MethylationEPIC Bead Chip. PLoS One. 2019 Apr 19;14(4):e0207834.
Li M, Li Y, Qin H, Tubbs JD, Li M, Qiao C, et al. Genome-wide DNA methylation analysis of peripheral blood cells derived from patients with first-episode schizophrenia in the Chinese Han population. Mol Psychiatry. 2021 Aug 5;26(8):4475–85.
Zhang X, Xiang Y, He D, Liang B, Wang C, Luo J, et al. Identification of Potential Biomarkers for CAD Using Integrated Expression and Methylation Data. Front Genet. 2020 Sep 9;11.
Draškovič T, Hauptman N. Discovery of novel DNA methylation biomarker panels for the diagnosis and differentiation between common adenocarcinomas and their liver metastases. Sci Rep. 2024 Feb 7;14(1):3095.
Noguera-Castells A, García-Prieto CA, Álvarez-Errico D, Esteller M. Validation of the new EPIC DNA methylation microarray (900K EPIC v2) for high-throughput profiling of the human DNA methylome. Epigenetics. 2023 Dec 31;18(1).
Kaur D, Lee SM, Goldberg D, Spix NJ, Hinoue T, Li HT, et al. Comprehensive evaluation of the Infinium human MethylationEPIC v2 BeadChip. Epigenetics Communications. 2023 Sep 27;3(1):6.
Udomruk S, Orrapin S, Pruksakorn D, Chaiyawat P. Size distribution of cell-free DNA in oncology. Crit Rev Oncol Hematol. 2021 Oct;166:103455.
Khurram I, Khan MU, Ibrahim S, Saleem A, Khan Z, Mubeen M, et al. Efficacy of cell-free DNA as a diagnostic biomarker in breast cancer patients. Sci Rep. 2023 Sep 15;13(1):15347.
Shen W, Dong H, Tang H, Zhang Y, Jia S, Luo Y. Diagnosis of prostate cancer using cell-free DNA methylation profiles from expressed prostatic secretions. Journal of Clinical Oncology. 2023 Feb 20;41(6_suppl):389–389.
Lauková L, Konečná B, Vlková B, Mlynáriková V, Celec P, Šteňová E. Anti-cytokine therapy and plasma DNA in patients with rheumatoid arthritis. Rheumatol Int. 2018 Aug 23;38(8):1449–54.
Raptis L, Menard HA. Quantitation and characterization of plasma DNA in normals and patients with systemic lupus erythematosus. J Clin Invest. 1980 Dec;66(6):1391–9.
Xu Y, Song Y, Chang J, Zhou X, Qi Q, Tian X, et al. High levels of circulating cell‐free <scp>DNA</scp> are a biomarker of active <scp>SLE</scp>. Eur J Clin Invest. 2018 Nov 23;48(11).
Spector BL, Harrell L, Sante D, Wyckoff GJ, Willig L. The methylome and cell-free DNA: current applications in medicine and pediatric disease. Pediatr Res. 2023 Jul 16;94(1):89–95.
Medina JE, Dracopoli NC, Bach PB, Lau A, Scharpf RB, Meijer GA, et al. Cell-free DNA approaches for cancer early detection and interception. J Immunother Cancer. 2023 Sep 11;11(9):e006013.
McDonough SJ, Bhagwate A, Sun Z, Wang C, Zschunke M, Gorman JA, et al. Use of FFPE-derived DNA in next generation sequencing: DNA extraction methods. PLoS One. 2019;14(4):e0211400.
Lee JM, Park SU, Lee SD, Lee HY. Application of array-based age prediction models to post-mortem tissue samples. Forensic Sci Int Genet. 2024 Jan;68:102940.
Lee HY, Hong SR, Lee JE, Hwang IK, Kim NY, Lee JM, et al. Epigenetic age signatures in bones. Forensic Sci Int Genet. 2020 May;46:102261.
Pruszkowska-Przybylska P, Dupont ME, Jacobsen SB, Smerup M, Tfelt-Hansen J, Morling N, et al. Evaluation of DNAmAge in paired fresh, frozen, and formalin-fixed paraffin-embedded heart tissues. PLoS One. 2024;19(5):e0299557.
Sawyer S, Gelabert P, Yakir B, Llanos-Lizcano A, Sperduti A, Bondioli L, et al. Improved detection of methylation in ancient DNA. Genome Biol. 2024 Oct 10;25(1):261.
Wagner S, Plomion C, Orlando L. Uncovering Signatures of DNA Methylation in Ancient Plant Remains From Patterns of Post-mortem DNA Damage. Front Ecol Evol. 2020 Jan 31;8.
Gokhman D, Malul A, Carmel L. Inferring Past Environments from Ancient Epigenomes. Mol Biol Evol. 2017 Oct 1;34(10):2429–38.
Schmidt M, Maixner F, Hotz G, Pap I, Szikossy I, Pálfi G, et al. DNA methylation profiling in mummified human remains from the eighteenth-century. Sci Rep. 2021 Jul 29;11(1):15493.
Christiansen SN, Andersen JD, Kampmann ML, Liu J, Andersen MM, Tfelt-Hansen J, et al. Reproducibility of the Infinium methylationEPIC BeadChip assay using low DNA amounts. Epigenetics. 2022 Dec 2;17(12):1636–45.
Wickham H. ggplot2: Elegant Graphics for Data Analysis [Internet]. Springer-Verlag New York; 2016. Available from: https://ggplot2.tidyverse.org
R Core Team. R: A Language and Environment for Statistical Computing [Internet]. Vienna, Austria; 2021. Available from: https://www.R-project.org/
Zhou W, Triche TJ, Laird PW, Shen H. SeSAMe: reducing artifactual detection of DNA methylation by Infinium BeadChips in genomic deletions. Nucleic Acids Res. 2018 Jul 31;
Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. J R Stat Soc Series B Stat Methodol. 1995 Jan 1;57(1):289–300.
Gu Z. IlluminaHumanMethylationEPICv2anno.20a1.hg38: Annotation for Illumina’s EPIC v2.0 methylation arrays [Internet]. 2024. Available from: https://www.illumina.com/products/by-type/microarray-kits/infinium-methylation-epic.html
Rainer J. EnsDb.Hsapiens.v86: Ensembl based annotation package. 2017.
Pelegí-Sisó D, de Prado P, Ronkainen J, Bustamante M, González JR. methylclock : a Bioconductor package to estimate DNA methylation age. Bioinformatics. 2021 Jul 19;37(12):1759–60.
Horvath S. DNA methylation age of human tissues and cell types. Genome Biol. 2013;14(10):R115.
Horvath S, Oshima J, Martin GM, Lu AT, Quach A, Cohen H, et al. Epigenetic clock for skin and blood cells applied to Hutchinson Gilford Progeria Syndrome and ex vivo studies. Aging. 2018 Jul 26;10(7):1758–75.
Zhang Q, Vallerga CL, Walker RM, Lin T, Henders AK, Montgomery GW, et al. Improved precision of epigenetic clock estimates across tissues and its implication for biological ageing. Genome Med. 2019 Dec 23;11(1):54.
Rodriguez-Casanova A, Costa-Fraga N, Castro-Carballeira C, González-Conde M, Abuin C, Bao-Caamano A, et al. A genome-wide cell-free DNA methylation analysis identifies an episignature associated with metastatic luminal B breast cancer. Front Cell Dev Biol. 2022 Oct 25;10.
Hahn J, Bressler J, Domingo-Relloso A, Chen MH, McCartney DL, Teumer A, et al. DNA methylation analysis is used to identify novel genetic loci associated with circulating fibrinogen levels in blood. Journal of Thrombosis and Haemostasis. 2023 May;21(5):1135–47.
Hop PJ, Zwamborn RAJ, Hannon EJ, Dekker AM, van Eijk KR, Walker EM, et al. Cross-reactive probes on Illumina DNA methylation arrays: a large study on ALS shows that a cautionary approach is warranted in interpreting epigenome-wide association studies. NAR Genom Bioinform. 2020 Dec 17;2(4).
Sahoo K, Sundararajan V. Methods in DNA methylation array dataset analysis: A review. Comput Struct Biotechnol J. 2024 Dec;23:2304–25.
Lee SM, Loo CE, Prasasya RD, Bartolomei MS, Kohli RM, Zhou W. Low-input and single-cell methods for Infinium DNA methylation BeadChips. Nucleic Acids Res. 2024 Apr 24;52(7):e38–e38.
Dai Q, Ye C, Irkliyenko I, Wang Y, Sun HL, Gao Y, et al. Ultrafast bisulfite sequencing detection of 5-methylcytosine in DNA and RNA. Nat Biotechnol. 2024 Jan 2;
Simons RB, Karkala F, Kukk MM, Adams HHH, Kayser M, Vidaki A. Comparative performance evaluation of bisulfite- and enzyme-based DNA conversion methods. Clin Epigenetics. 2025 Apr 3;17(1):56.
Kaur D, Lee SM, Goldberg D, Spix NJ, Hinoue T, Li HT, et al. Comprehensive evaluation of the Infinium human MethylationEPIC v2 BeadChip. Epigenetics Communications. 2023 Sep 27;3(1):6.
Peters TJ, Meyer B, Ryan L, Achinger-Kawecka J, Song J, Campbell EM, et al. Characterisation and reproducibility of the HumanMethylationEPIC v2.0 BeadChip for DNA methylation profiling. BMC Genomics. 2024 Mar 6;25(1):251.
Naue J, Hoefsloot HCJ, Kloosterman AD, Verschure PJ. Forensic DNA methylation profiling from minimal traces: How low can we go? Forensic Sci Int Genet. 2018 Mar;33:17–23.
Poggiali B, Dupont ME, Jacobsen SB, Smerup MH, Christiansen SNN, Tfelt-Hansen J, et al. DNA methylation stability in cardiac tissues kept at different temperatures and time intervals. Sci Rep. 2024 Oct 24;14(1):25170.
Zhang W, Young JI, Gomez L, Schmidt MA, Lukacsovich D, Varma A, et al. Critical evaluation of the reliability of DNA methylation probes on the Illumina MethylationEPIC v1.0 BeadChip microarrays. Epigenetics. 2024 Dec 31;19(1).
Chen BH, Marioni RE, Colicino E, Peters MJ, Ward-Caviness CK, Tsai PC, et al. DNA methylation-based measures of biological age: meta-analysis predicting time to death. Aging. 2016 Sep 28;8(9):1844–65.
Zhang Y, Wilson R, Heiss J, Breitling LP, Saum KU, Schöttker B, et al. DNA methylation signatures in peripheral blood strongly predict all-cause mortality. Nat Commun. 2017 Mar 17;8(1):14617.
Levine ME, Lu AT, Quach A, Chen BH, Assimes TL, Bandinelli S, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging. 2018 Apr 18;10(4):573–91.
Ryan J, Wrigglesworth J, Loong J, Fransquet PD, Woods RL. A Systematic Review and Meta-analysis of Environmental, Lifestyle, and Health Factors Associated With DNA Methylation Age. The Journals of Gerontology: Series A. 2020 Feb 14;75(3):481–94.
Cisneros-Villanueva M, Hidalgo-Pérez L, Rios-Romero M, Cedro-Tanda A, Ruiz-Villavicencio CA, Page K, et al. Cell-free DNA analysis in current cancer clinical trials: a review. Br J Cancer. 2022 Feb 1;126(3):391–400.
Qi T, Pan M, Shi H, Wang L, Bai Y, Ge Q. Cell-Free DNA Fragmentomics: The Novel Promising Biomarker. Int J Mol Sci. 2023 Jan 12;24(2).
Dupont ME, Jacobsen SB, Christiansen SNN, Tfelt-Hansen J, Smerup MH, Andersen JD, et al. Fresh and frozen cardiac tissue are comparable in DNA methylation array β-values, but formalin-fixed, paraffin-embedded tissue may overestimate DNA methylation levels. Sci Rep. 2023 Sep 29;13(1):16381.
Dupont ME, Christiansen SN, Jacobsen SB, Kampmann ML, Olsen KB, Tfelt-Hansen J, et al. DNA quality evaluation of formalin-fixed paraffin-embedded heart tissue for DNA methylation array analysis. Sci Rep. 2023 Feb 3;13(1):2004.
Mathieson W, Thomas GA. Why Formalin-fixed, Paraffin-embedded Biospecimens Must Be Used in Genomic Medicine: An Evidence-based Review and Conclusion. J Histochem Cytochem. 2020 Aug;68(8):543–52.

No competing interests reported.

Download PDF

Journal Publication

published 21 Aug, 2025

Read the published version in Biological Procedures Online →

Editorial decision: Revision requested
07 May, 2025
Reviews received at journal
06 May, 2025
Reviews received at journal
27 Apr, 2025
Reviewers agreed at journal
24 Apr, 2025
Reviewers agreed at journal
24 Apr, 2025
Reviewers invited by journal
23 Apr, 2025
Submission checks completed at journal
23 Apr, 2025
First submitted to journal
18 Apr, 2025

You are reading this latest preprint version

Evaluation of the HumanMethylationEPIC v2.0 Bead Chip using low quality and quantity DNA samples

Status:

Journal Publication

Version 1

Abstract

Figures

Background

Methods

Results

Discussion

Conclusions

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1