Strong concordance of DNA methylation measures between venous and capillary blood - Implications for human epigenome studies

Summary

DNA methylation (DNAm) is influenced by the tissue composition, which can be attributed to cellular heterogeneity. Venous and capillary blood are common blood sources in human studies, but may have differences in their cell type composition and resulting DNAm profiles. Using the Illumina EPIC v2 microarray, we compared matched (n = 20) venous and capillary blood samples from the VHAS and illustrated congruency in a typical DNAm workflow. Quality control measures indicated similar performance, with high CpG correlation (Spearman’s ρ > 0.9). Blood source was not associated with significant variation in the data, and an epigenome-wide association study (EWAS) indicated no difference as a result of blood source. Fourteen DNAm-based predictors, including epigenetic clocks and inflammatory predictors, did not differ by sample source (p > 0.05). Stratified EWASs of sex and age by blood source identified similar CpG sites. These results suggest that in large-scale DNAm studies, capillary and venous blood sample results are comparable.

Graphical abstract

Highlights

• •Blood source does not drive DNA methylation variation between capillary and venous samples
• •Estimated cell types and epigenetic age are consistent across blood sources
• •No differentially methylated CpG sites between capillary and venous blood
• •Sex and ELOVL2 DNA methylation patterns are consistent across two blood draw sources

Epigenetics; Genomics

Introduction

Human clinical and epidemiological studies, including studies with epigenetic measures, often rely on the collection of whole blood samples. The majority of these samples are obtained from either capillary or venous sources, with arterial blood sampling only required in special cases due to its greater risk of complication.¹ As blood is not separated by defined boundaries in the circulatory network, there are similarities between blood samples obtained from different components of the circulatory vessels; however, certain differences exist. Most pertinently, venous blood is deoxygenated, while capillary blood is a unique combination of oxygenated and deoxygenated blood from the capillary crossover between venules and arterioles.² Both capillary blood sampling and venipuncture are commonly employed sources of blood sampling. Capillary blood collection, generally performed via finger prick in adults or heel-prick in infants, is a simple procedure that can be self-conducted and permits the collection of smaller volumes of blood. Capillary blood collection is considered to provide greater physical comfort compared to venipuncture, which involves drawing larger volumes of blood. However, venipuncture is more invasive and can be challenging to implement in certain populations such as children, older adults, and obese individuals.³ In terms of storage, capillary blood can be stored either at room temperature as a dried blood spot (DBS) on absorbent paper or in small-volume microtainer or capillary tubes. DBS sampling is advantageous in many settings, as the collection cards can be stored and shipped at room temperature, with low biohazard risk.^4,5,6

Hematological investigations comparing venous and capillary blood indicate that there are differences in levels of hematocrit, hemoglobin, red blood cell, and leukocyte counts.^3,7,8 Clinical as well as population-based cohort studies often use venous blood samples for hematological and biomarker profiling, and in recent years, they have increasingly included DNA methylation (DNAm) quantification for epigenome-wide association studies (EWASs) (including the Framingham Heart Study, UK Biobank, and Irish Longitudinal Study on Ageing).^9,10,11,12 DNAm is a mitotically heritable, chemical modification (-CH₃) to nucleotides in DNA, typically in the context of cytosine-phosphate-guanine (CpG) residues. As a modification to DNA nucleotides, it forms a part of the regulatory machinery involved in many aspects of cell function, including cell-fate determination and environmental responses.^13,14

DNAm is the basis for several types of epigenetic predictive biomarkers,¹⁵ such as epigenetic clocks, inflammation biomarkers DNAm-IL-6¹⁶ and DNAm-CRP,¹⁷ and lifestyle factors such as smoking history¹⁸ and alcohol consumption.¹⁹ These DNAm-based biomarkers are used to interrogate health and biological aging across large cohorts as the nexus of biological response to altered environments in the face of aging and disease conditions. While inflammation biomarkers can be measured directly from blood, and smoking and alcohol consumption habits can be queried via survey, DNAm-predicted biomarkers can be useful in cases where original blood sources are limited, or where DNAm-predicted biomarkers may have stronger associations with an outcome.^20,21 Currently, there are minimal comparative investigations of venous and capillary blood in the context of DNAm research, and extant research has only focused on comparisons of DBS and venous blood preparations. For example, DNAm profiles of 62 matched DBS from Whatman paper cards and frozen venous blood samples from the Generation Scotland: Scottish Family Health Study have demonstrated highly correlated DNAm patterns²² (r = 0.991), and showed similar differentially methylated sites associated with age and sex in EWAS analyses.

One such cohort, which is positioned to investigate DNAm congruency across venous and capillary blood, is the Vietnam Health and Aging Study (VHAS), a longitudinal cohort that aims to characterize long-term health and resilience following exposure to violence during the Vietnam/Second Indochina War.²³ The VHAS possesses a combination of capillary and venous banked samples across longitudinal collection waves, as older subjects found venipuncture more comfortable than capillary sampling, prompting a switch in blood collection method between waves 1 and 2. Here, we present an exploratory investigation to test whether the DNAm profiles generated using venous blood and capillary blood draws from the same time point, both stored in K2-EDTA anticoagulant vacutainers, are congruent. Based on previous DBS studies, we hypothesize that the source of blood collection across longitudinal waves will not overly influence DNAm findings in this cohort, from either the DNAm-based biomarker or EWAS analyses, and will legitimize comparisons across waves of data collection.

Results

DNA methylation quality control characteristics did not differ by blood draw source

In this study, we aimed to determine the extent of possible congruency in DNAm patterns between venous and capillary blood samples, profiled on the EPICv2 array. To do this, we leveraged a subset of the VHAS cohort, comprising 20 matched venous and capillary blood samples from individuals in the age range of 66–86 years of age, including 13 males and seven females. To assess whether sample quality differed between venous or capillary blood draw, we first compared standard DNAm sample quality control metrics derived from EPIC v2 array data using paired t-tests. All the samples passed the quality checks based on predefined thresholds. We observed no significant differences between the DNAm sample quality parameters from venous and capillary blood (Table 1).

Table 1.

Sample quality control metrics do not differ between venous and capillary blood draws

Quality control metric	Venous draw (mean ± sd)	Capillary draw (mean ± sd)	p value
1. Detection p value probes	1040 ± 335	974 ± 217	0.364
2. Missing bead probes	984 ± 318	932 ± 272	0.549
3. Methylated median intensity	12.5 ± 0.182	12.4 ± 0.203	0.521
4. Unmethylated median intensity	11.9 ± 0.218	11.9 ± 0.164	0.702
5. Bisulfite conversion yield	39.2 ± 8.86	38.7 ± 10.7	0.874

High correlation was observed in DNA methylation beta values across all the CpG probes on the array

DNAm intensity from the EPICv2 array is captured as a beta value, ranging between 0 (fully unmethylated) and 1 (completely methylated). In order to determine the concordance in DNAm beta values between the venous-drawn and capillary-drawn samples, we estimated Spearman correlations of all the CpGs measured on the array averaged across all the samples. We identified high Spearman correlations at the CpG level across all samples between the blood draw types (Spearman’s ρ = 0.99, p < 2.2 × 10⁻¹⁶) (Figure 1). The lowest correlation observed was a considerably strong correlation (Spearman ρ = 0.96, p < 2.2 × 10⁻¹⁶). To ensure that the high concordance was not driven by CpGs that were either highly methylated (>0.9) or lowly methylated (<0.1) in both capillary and venous samples, we excluded these sites (n = 339,997 CpGs), and a high correlation was still observed (Spearman’s ρ = 0.9976, p < 2.2 × 10⁻¹⁶).

Figure 1.

Mean methylation of CpGs from venous and capillary blood is highly correlated

Probe-to-probe correlation across 20 individuals was highly correlated between venous and capillary blood.

Venous or capillary blood draw source did not drive the variation in DNA methylation

Broadly, it is well-established that variability in DNAm is driven by cell type heterogeneity, sex, age, and genetic variation, which should be accounted for.^24,25 Therefore, we sought to determine whether the method of blood collection, capillary versus venous, contributed to DNAm variation in this dataset. To capture global patterns of variation in our cohort, we performed a principal component analysis on noob and funnorm-normalized CpGs (936,990 CpGs). As expected, predicted blood cell type proportions and sex were significantly associated with the top 3 principal components, explaining approximately 36% of the total variation in the data. The origin of the blood sample, from either venous or capillary sources, was not associated with the top ten PCs (Figure S1A), nor did hierarchical clustering indicate a separation of samples by blood draw type, but rather showed that samples cluster by participant (Figure S1B). This demonstrated that the source of blood did not drive variability across samples at a high-level resolution. To complement this unsupervised PCA-based approach, we performed an EWAS to systematically compare DNAm between venous and capillary blood at the individual CpG level. With blood draw source as the main effect, no CpG sites passed the significance threshold of FDR adjusted p < 0.05, indicating no detectable DNAm difference between paired venous and capillary samples (Figure S1C).

Blood collection type did not affect the DNA methylation-based cell type proportion estimates

Cell-type heterogeneity is a main driver of DNAm in whole blood.^24,25 In the absence of cell count data, the majority of studies in DNAm research rely on estimated immune cell proportions using DNAm-based reference deconvolution algorithms. These predicted cell type proportions are then included as covariates in statistical models to account for cell type variation. We tested whether the blood draw source influenced predicted proportions of immune cells, and thus we estimated 12 immune cell type proportions using the whole blood reference-based deconvolution algorithm, and compared the proportions between the blood sources using the automatic probe selection method. Using this method, we noted no statistical differences in the cell type proportion estimates per individual between the venous blood and capillary blood samples when calculated together or separately (Figure 2). Furthermore, when examining the probe coefficients selected for cell type estimation using automatic probe selection, the selected probes and their coefficients were identical across venous and capillary blood samples when calculated separately. Bland-Altman plots (Figure S2) suggest high correlation, with minimal bias, as the mean difference between estimated cell type proportions in venous and capillary blood was < |0.005| across all cell types. While individual samples did not have identical predicted cell type proportions across venous and capillary blood, there was no patterned shift in the proportion of any cell type in either venous or capillary blood, and there were no significant differences in the proportion of predicted cell types per individual (Figure S2).

Figure 2.

DNA methylation-predicted cell type proportions do not significantly differ between venous and capillary blood

DNA methylation-based estimation of cell type proportion is a common epigenetic technique, and no differences were noted between capillary and venous blood. Each point represents an individual, with the box-and-whisker plot indicating the median, first- and third quartiles of the data, and 1.5x the first or third quartiles of the data.

DNA methylation-based tools comparison indicated no difference between capillary and venous blood samples

DNAm-based tools, including epigenetic clocks, inflammation scores, and lifestyle predictors, are commonly employed tools in epigenetic research. We sought to test whether the predicted estimates from these tools were concordant between the venous and capillary blood samples. We observed no significant differences in epigenetic clocks, their derived measures, or the inflammation and lifestyle DNAm-based predictors based on blood draw types (Table 2). Moderate Spearman correlations between epigenetic age and chronological age were observed (ρ > 0.5), with consistently stronger correlations in venous blood (Table S1). Epigenetic age acceleration (EAA), a derived measure from epigenetic age, is commonly used to predict the rate of aging and is calculated as the residuals from the linear regression of epigenetic age against chronological age. EAA was calculated two ways: (1) with pooled venous and capillary samples in the linear regression model and (2) with separate linear models for capillary and venous samples. No significant differences (p < 0.05) were identified in either case of EAA calculation, nor across all the tested DNAm-based tools (Table 2). As has been found previously,^26,27 males consistently had a higher EAA than females across all clocks tested (Figure S3).

Table 2.

DNA methylation predictors do not differ between venous and capillary blood samples

Predictor	Venous draw (mean ± sd)	Capillary draw (mean ± sd)	p value
Epigenetic age
Horvath DNAmAge	73.3 ± 9.5	72.2 ± 8.63	0.656
Hannum DNAmAge	40.2 ± 5.81	39.5 ± 5.79	0.650
GrimAge	69.3 ± 8.78	69.6 ± 8.16	0.923
PhenoAge	68.1 ± 8.97	66.9 ± 9.15	0.659
Skin and Blood DNAmAge	60.8 ± 6.62	59.7 ± 6.51	0.505
Epigenetic clocks - EAA calculated together
Horvath DNAmAge EAA together	0.614 ± 6.25	−0.614 ± 5.35	0.398
Hannum DNAmAge EAA together	0.384 ± 3.92	−0.384 ± 3.94	0.480
GrimAge EAA together	−0.094 ± 5.18	0.094 ± 4.79	0.899
PhenoAge EAA together	0.629 ± 6.90	−0.629 ± 6.42	0.548
SkinBlood DNAmAge EAA together	0.592 ± 3.7	−0.592 ± 4.02	0.254
DunedinPACE together	1.23 ± 0.125	1.24 ± 0.146	0.698
Epigenetic clocks - EAA calculated separately
Horvath DNAmAge EAA separately	0.000 ± 6.24	0.000 ± 5.35	1
Hannum DNAmAge EAA separately	0.000 ± 3.92	0.000±3.94	1
GrimAge EAA separately	0.000 ± 5.18	0.000 ± 4.79	1
PhenoAge EAA separately	0.000 ± 6.89	0.000 ± 6.40	1
SkinBlood DNAmAge EAA separately	0.000 ± 3.72	0.000 ± 4.02	1
DunedinPACE separately	1.23 ± 0.124	1.24 ± 0.146	0.742
Lifestyle and inflammation biomarkers
Smoking (McCartney)	2.74 ± 0.868	2.84 ± 0.881	0.692
Smoking score (Elliot)	−1.54 ±4.25	−1.23 ±4.27	0.816
DNAm IL6	0.086 ± 0.065	0.079 ± 0.069	0.735
DNAm CRP	64.4 ± 37.8	71.3 ± 40.5	0.489
BMI (McCartney)	−0.392 ± 0.115	−0.393 ± 0.128	0.989
Alcohol (McCartney)	−12.7 ± 0.342	−12.7 ± 0.255	0.701
HDL Cholesterol (McCartney)	2.40 ± 0.151	2.34 ± 0.163	0.105
LDL Cholesterol (McCartney)	−4.01 ± 0.353	−4.00 ± 0.266	0.914
Percent Body Fat (McCartney)	−5.98 ± 3.78	−6.37 ± 3.62	0.666

Comparison of candidate ELOVL2 CpG methylation in venous and capillary blood indicates no difference across samples

Increased DNAm of the ELOVL2 promoter is a well-established age-related phenomenon that occurs across multiple tissues.^28,29,30 To examine whether venous or capillary blood possessed age-related ELOVL2 methylation variability, we considered the DNAm patterns of fourteen CpGs in the promoter of the gene ELOVL2 which are known to increase with age across most tissue types (cg24724428, cg21572722, cg16867657, cg22143569, cg05446010, cg15557662, cg03577977, cg25151806, cg16323298, cg01799681, cg01142742, cg14153064, cg09017996, and cg02759911), and which have been characterized in other studies.^28,29,30 Here, we compared the DNAm of ELOVL2 CpGs in venous and capillary blood and found no significant difference (FDR adjusted p < 0.05) in DNAm level per individual between venous and capillary blood (Figures 3 and S4, and Table 3), suggesting a similar DNAm pattern of ELOVL2 between these two blood sources.

Figure 3.

No significant differences between fourteen ELOVL2 CpG DNA methylation values in capillary and venous blood

DNA methylation at CpGs in the ELOVL2 promoter shows cross-tissue concordance in other tissues, which are recapitulated between venous and capillary blood. Each point represents an individual, with the box-and-whisker plot indicating the median, first- and third quartiles of the data, and 1.5x the first or third quartiles of the data.

Table 3.

Comparison of mean methylation of ELOVL2 CpGs in capillary and venous blood indicates no differences

CpG in ELOVL2	Mean methylation value in capillary blood	Mean methylation value in venous blood	95% confidence interval	p value	FDR (adj.p)
cg24724428	0.332681	0.325864	−0.05282601; 0.06645968	0.819	0.937
cg21572722	0.532015	0.535855	−0.04434982; 0.03666999	0.849	0.937
cg16867657	0.776158	0.781774	−0.04152068; 0.03028939	0.754	0.937
cg22143569	0.022606	0.02304	−0.002459749; 0.001591473	0.668	0.937
cg05446010	0.018333	0.018761	−0.002189325; 0.001332773	0.626	0.937
cg15557662	0.022824	0.023751	−0.003299668; 0.001445942	0.434	0.937
cg03577977	0.932822	0.929937	−0.002191459; 0.007962052	0.256	0.937
cg25151806	0.882053	0.883319	−0.03111013; 0.02857815	0.932	0.937
cg16323298	0.282976	0.283876	−0.02381298; 0.02201313	0.937	0.937
cg01799681	0.248924	0.246282	−0.02769154; 0.03297532	0.861	0.937
cg01142742	0.890729	0.893184	−0.008884242; 0.003973805	0.445	0.937
cg14153064	0.341592	0.334625	−0.01922249; 0.03315543	0.594	0.937
cg09017996	0.928595	0.935183	−0.012147399; −0.001030559	0.021	0.298
cg02759911	0.790273	0.797089	−0.05166751; 0.03803448	0.761	0.937

Single-site EWAS on biological variables indicates similarity between venous and capillary blood samples

Age and sex are two important biological variables known to influence DNAm in multiple tissues, including whole blood.^24,31 After the comparison of age-associated candidate CpGs above, we conducted a single-site EWAS investigation of these two important biological phenotype characteristics (Figure S5). First, we separately identified variable autosomal CpGs in each blood source and found the 463,661 CpGs that were variable in both venous and capillary blood. Second, we conducted an EWAS in capillary and venous blood separately, on the commonly variable CpGs, for age and sex, including predicted proportions of naive B and naive CD8T cells in the model to account for cell type heterogeneity in this small sample of 20 individuals. To identify differentially methylated CpGs associated with age and sex, we employed both a statistical and effect size threshold (FDR adjusted p < 0.05, and absolute linear regression coefficient >0.05, respectively) and compared the rank coefficients for age and sex to determine the general comparability of venous and capillary blood sample DNAm. No single CpG site was significantly associated with age in either venous or capillary blood samples. We observed modest rank correlation (τ = 0.49, p value < 2.2e⁻¹⁶) of coefficient estimates for age between all 463,661 variable probes (Table 4), suggesting that the age-related changes to DNAm are detected at a similar strength in both venous and capillary blood.

Table 4.

Kendall correlation coefficients of the regression coefficient of age and sex from EWASs of capillary and venous blood, conducted separately

Variable	Comparison	Rank Coefficient (τ)	p value
Sex	correlation between 30 CpGs with FDR <0.05 and regression coefficient >0.05	−0.149	0.256
Sex	correlation of CpG coefficient estimate between all 463661 probes included in analysis	0.471	<2.2e−16
Age	correlation of CpG coefficient estimate between all 463661 probes included in analysis	0.496	<2.2e−16

When interrogating sex, 41 CpGs were associated with sex in capillary blood samples (FDR adjusted p < 0.05, regression coefficient > |0.05|), and 43 CpGs were associated with sex in venous blood samples (FDR <0.05, absolute regression coefficient > |0.05|) (Figure 4). Of the 43 sex-associated CpGs in venous blood, 30 CpGs (69.77%) were differentially methylated in both blood types, with concordant directions of effect according to the regression coefficients, with all absolute differences between regression coefficients < 0.02 (Table S2). Of these 30 CpGs, 17 showed higher methylation in males compared to females, and 13 showed lower methylation in males compared to females (Table S2). The rank correlation (τ) of the regression coefficients of these 30 commonly differentially methylated CpGs for sex was −0.149 (p value = 0.256), while the overall rank correlation (τ) between all 463,661 variable probes included in the venous and capillary EWAS was 0.47 (p value < 2.2e⁻¹⁶) (Table 4). Overall, similar results are obtained, with minimal differences in coefficients and a moderate rank coefficient across all probes, suggesting a high concordance of the DNAm patterns related to sex and age in both venous and capillary blood.

Figure 4.

Volcano plots indicate similarity of direction of sex effect across venous and capillary blood

The regression coefficients and p-values of sex across CpG sites in (A) venous and (B) capillary blood. CpG sites with FDR-adjusted p-values < 0.05 and regression coefficient > |0.05| are highighted in red and blue. Positive coefficients indicate higher DNA methylation in males compared to females; negative coefficients indicate lower DNA methylation in males compared to females.

Discussion

Large-scale cohort studies establish standardized protocols and methodologies in advance and generally maintain consistency throughout waves of data collection. However, unforeseen circumstances, such as the global pandemic that occurred during wave 1 of VHAS sample collection, may necessitate changes in sampling methodology. Changes in patient preference may also arise over the course of longitudinal studies, with some participants who initially consented to venipuncture later preferring a less invasive capillary blood collection, or vice versa, as occured in VHAS. Similarly, international consortia establishing cohorts in low- and middle-income countries (LMICs) may not anticipate logistical challenges with sampling methods until the process has commenced and must then adapt to changing circumstances. In addition to these practical considerations, not all cohort studies use identical blood sampling methodologies for DNAm profiling; therefore, systematic evidence comparing blood sampling methods will enhance the integration of findings across studies, enabling cross-cohort comparisons. Here, we sought to examine whether DNAm profiles in whole blood from venous and capillary draws were comparable based on output from the EPICv2 array. While studies have previously compared venipuncture and capillary blood draw sources in the context of hematological parameters,^7,32,33 transcriptome,^34,35 and the metabolome,³⁶ there are relatively fewer studies of systematic investigations of the DNA methylome based on blood draw sources. Those done to date have primarily compared DBS to venous blood DNAm, using the Illumina Infinium 450K array.^5,22,37 Taking advantage of a matched subset of venous and capillary blood drawn samples from a Southeast Asian adult population (VHAS), we carried out an epigenetic analysis comparing the two blood collection strategies in the context of measured DNAm levels across the array CpGs, predicted immune cell type proportions, DNAm-based tools, and EWAS investigations from blood collected in vacutainer and microtainer tubes. We observed a high correlation of DNAm levels at the individual CpGs measured between the venipuncture and capillary blood drawn samples. Further, we noted no significant differences between the two blood draw types in DNAm-based epigenetic age estimates, inflammatory scores, and lifestyle predictors. In the single CpG EWAS, we observed concordance in 69% of significant DNAm signatures associated with sex between venous and capillary blood samples. Taken together, we provide compelling evidence to support that DNAm measures obtained from capillary blood samples and venous blood sampling are congruent.

When comparing the DNA methylomes between the venous and capillary drawn samples, we observed a strong correlation of beta values as measured by the EPICv2 array. Similarly, we noted high correlation in the beta values across all measures of quality, suggesting the methylomes are providing similar DNAm readouts irrespective of the two blood draw origins. Given the congruency in DNAm values between the venous and capillary blood samples, we also compared the estimates obtained from the DNAm-based tools between the two blood draw types and observed an agreement in the predicted outputs across the DNAm-based tools (Table 2), across epigenetic age prediction, EAA, and inflammation and lifestyle predictors, re-emphasizing that the epigenetic information extracted from these tools is not affected by the blood collection strategy. Calculation of EAA is influenced by the cohort of individuals used in the linear regression, which motivated a comparison of EAA calculated with venous and capillary samples together and separately. No such differences were evident when EAA was calculated using all samples together, and stratification by blood origin also indicated no consistent differencs in EAA estimates.

While we did not observe an effect of blood collection strategy in our DNAm data, as evaluated by hierarchical clustering and PCA, cell composition, as expected, was the primary contributor to the observed DNAm variation. No significant differences in predicted immune cell type proportions were evident between venipuncture and capillary blood samples, highlighting that the estimated cellular proportions in adult blood were not influenced by the blood collection strategy. This corresponds with previous findings, which similarly illustrate a DNAm-based cell deconvolution tool derived from pyrosequencing measurements with correlations in estimated cellular proportions between venous blood and capillary blood.³⁸ While these findings were observed in DNAm-derived cell proportion estimates, studies conducted in actual blood counts also exhibited comparable cellular estimates between venous and capillary drawn samples,^39,40,41 though some studies reported a trend of increased granulocytes in capillary blood, which was speculated as a result of immediate local accumulation of granulocytes in the fingertip due to skin puncture.⁴² Beyond these analyses, our direct EWAS comparison of venous versus capillary blood showed no detectable DNAm differences between the two collection methods, suggesting remarkable concordance between blood draw sources and confirming that the collection method does not introduce measurable DNAm variation. We also demonstrated the concordance between venous and capillary blood drawn samples with an EWAS exemplar, focusing on sex, an important biological variable with an established DNAm signature in multiple tissues.^43,44 At our adopted thresholds, we observed an overlap (>69%) of differentially methylated CpGs associated with sex between the two blood collection strategies, with concordance in linear regression coefficients, as a proxy for effect size in this example EWAS, confirming the reliability of findings obtained either by venous blood or capillary blood drawn samples. This suggests that the interpretability of DNAm findings is not likely to be overly influenced by sample origin in either capillary or venous blood.

Limitations of the study

While the findings presented here are critical in the research design and interpretation of epigenetic findings focused on whole blood, our study also has limitations. Firstly, the sample size was small; hence, the investigations should be explored in a larger cohort and in an independent population, as our study was conducted in samples from Southeast Asia. A smaller sample size may have impacted statistical power in the EWAS, though equitably in samples from both blood draw types. Secondly, the comparative DNAm analyses were conducted in adult samples from a relatively small age range (approximately 20 years in older adults); however, age is an important characteristic that contributes to DNAm variation in whole blood²⁴ as changes in DNAm patterns occur at several CpGs on the genome throughout the lifespan. In fact, the rate of DNAm change occurring in pediatric samples is greater than that observed in adults.⁴⁵ Therefore, it is imperative to conduct these epigenetic investigations in samples from different age groups, particularly in children, though DNAm-based tools in pediatric populations are limited. Thirdly, we acknowledge that measured cell counts were not available in the current study, which would have enabled us to determine whether the estimated cell type proportions derived from venipuncture samples or capillary blood drawn samples are more accurately reflective of actual cell proportions. However, the concordance between cell counts and DNAm-derived immune cell type proportions has been previously demonstrated,¹⁴ and the objective of the current study is to test the correlations in the estimated cell type proportions between the samples corresponding to the two blood draw types. Fourthly, while our epigenetic investigations were limited to the measured CpGs on the microarray, and may not be reflective of global DNAm contexts, these findings would be critical and relevant to those conducting microarray research.

Together, we demonstrated that DNAm estimates obtained from samples via venipuncture and capillary blood draws are comparable for CpGs measured on the EPIC v2 array. This remarkable concordance supports that these two methods of blood collection strategies are reliable alternatives for each other in array-based DNAm investigations. This work is of practical relevance when unforeseen circumstances necessitate changes in collection strategies, specifically in large-scale population studies, longitudinal studies, when facing logistical challenges in resource-limited settings, and in the context of meta-analyses when multiple cohorts with samples collected either by venous or capillary blood draws are combined to address a common research question.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Hannah-Ruth Engelbrecht (hannah.engelbrecht@bcchr.ca.).

Materials availability

This study did not generate any new unique reagents or materials.

Data and code availability

• •Normalized data are available on GEO at GSE312123 and are publicly available
• •Original code can be found under Venous-Capillary-Comparison at https://github.com/kobor-lab/Venous-Capillary-Comparison
• •Any additional information to reanalyze data reported in this manuscript is available from the corresponding author upon request.

Acknowledgments

We are grateful to the study participants who have donated their biological sample to this study. This research uses the FlowSorted.BloodExtended.EPIC software packages developed at Dartmouth College, which are governed by the licensing terms provided by Dartmouth Technology Transfer. MSK is a fellow of the CIFAR Child and Brain Development Program. This work was supported by the National Institute on Aging of the National Institutes of Health under award number R01AG052537, and Canadian Institutes of Health Research (CIHR)/Project Grant (Sponsor Reference Number: PJT-175309).

Author contributions

Design of study: H.-R.E. and C.K.; data analysis and interpretation: H.-R.E., C.K., and B.C.Z.; writing the main draft: H.-R.E. and C.K.; data acquisition: K.D. and J.L.M.; critical draft review: T.W.M., T.K.T., K.K., Z.Z., and M.S.K.; study conceptualization: T.K.T., K.K., Z.Z., and M.S.K.; funding acquisition: T.K.T., K.K., Z.Z., and M.S.K.; final approval of the version to be published: T.K.T., K.K., Z.Z., and M.S.K.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Paired venous and capillary blood	Vietnam	N/A
Critical commercial assays
DNeasy Blood & Tissue kit	QIAGEN	Cat# 69506
EZ-DNA methylation kit	Zymo Research	Cat# D5002
MethylationEPIC v2.0 BeadChip Array	Illumina	Cat# 20086312
Deposited data
DNA methylation data	this paper	GEO: GSE312123
Software and algorithms
Rstudio	Statistical software, version 2.2.2	N/A
Biobase	R package, Huber et al., 2015	N/A
dendextend	R package, Galili, 2015	N/A
DunedinPACE	R package, Belsky et al., 2022	N/A
EpiSmoker	R package, Bollepalli et al., 2019	N/A
ewastools	R package, Murat et al., 2020	N/A
ExperimentHub	R package, Morgan & Shepherd, 2022	N/A
factoextra	R package, Kassambara & Mundt, 2020	N/A
FlowSorted.Blood.EPIC	R package, Salas et al., 2022	N/A
ggplot2	R package, Wickham, 2016	N/A
ggrepel	R package, Slowikowski, 2024	N/A
gridExtra	R package, Auguie, 2022	N/A
hmisc	R package, Harrell Jr., 2026	N/A
Horvath Online Calculator	Horvath, 2013,	https://dnamage.genetics.ucla.edu/new
IlluminaHumanMethylationEPICv2anno.20a1.hg38	Illumina	Illumina, https://support.illumina.com/array/array_kits/infinium-methylationepic-beadchip-kit/downloads.html
lumi	R package, Du et al., 2008	N/A
methylumi	R package, Sean Davis, 2017	N/A
minfi	R package, Aryee et al., 2014; Fortin et al., 2017	N/A
plyr	R package, Wickham, 2011	N/A
psych	R package, William Revelle, 2026	N/A
Rcurl	R package, Lang, 2023	N/A
Reshape	R package, Wickham, 2007	N/A
rstatix	R package, Kassambara, 2023	N/A
stats	R package, R Core Team, 2022	N/A
sva	R package, Leek et al., 2012	N/A
wateRmelon	R package, Pidsley et al., 2013	N/A
Other
Resource website for this publication	this paper	https://github.com/kobor-lab/Venous-Capillary-Comparison
Reference pre-processing script	this paper, Zhuang et al., 2025	https://github.com/kobor-lab/EPICv2_QC_preprocessing

Experimental model and study participant details

The Vietnam Health and Aging Study is a longitudinal study of ethnically Vietnamese individuals over 60 years of age (n = 2 447) from northern Vietnam, which aims to examine the effects of war and trauma exposure on long-term health and resilience.²³ In addition to survey data from respondents, several biomarkers, including but not limited to functional biomarkers (e.g., grip strength), and blood biomarkers (e.g., CRP), are being profiled for this cohort, while a subsample has been selected for molecular biomarkers (e.g., epigenetic clocks) characterization.⁴⁶ This study was approved by the University of British Columbia Human Research Ethics Board (H18-03136), the University of Utah Institutional Review Board (IRB_00099861), Mount Saint Vincent University’s Research Ethics Board (2018–047), Singapore Management University’s IRB (IRB-18-031-A034(318)), Hanoi Medical University’s Independent Review Board in Bio-medical Research (IRB No. 00003121) and Vietnam’s Ministry of Health. All participants provided written informed consent for inclusion in the study as described previously.²³

Initially, in wave 1 data collection in 2018, blood sampled for biomarker profiling in the VHAS cohort was capillary blood, collected in BD Microtainer MAP microtubes with K2-EDTA anticoagulant via finger puncture.²³ However, as described in the 2023 dissemination report,⁴⁷ the method of blood collection at wave 2 (2021–2022) was changed to venous blood sampling to improve participant comfort. Both venous and capillary blood samples were collected by trained phlebotomists. During wave 2 sample collection, matched venous and capillary samples were obtained for a subset of individuals for venous and capillary blood DNA methylation comparison (n = 20, 66–86 years, from 13 male and seven female participants, with sex self-reported). Whole blood was frozen at −20°C within 15 min of collection for storage, then at −80°C, prior to transport on dry ice to the University of British Columbia, Canada, for nucleic acid extraction and DNA methylation quantification.

Method details

Quantification and data preprocessing

DNA was extracted from both venous and capillary blood samples using the QIAGEN DNeasy Blood & Tissue kit (QIAGEN, Aarhus, Denmark). DNA extraction was followed by bisulfite conversion using the EZ-DNA methylation kit (Zymo Research, CA, United States), and finally DNA methylation from all samples was quantified using the Illumina MethylationEPIC v2.0 BeadChip Array (“EPICv2”, Illumina, San Diego, CA, United States), according to manufacturers’ protocols. The resulting DNA methylation data obtained after being run on the Illumina iScan 2000 was read into the R environment (R 4.2.2) and subsequently pre-processed according to existing methods.⁴⁸ Briefly, sample quality control checks were performed using functions embedded in the methylumi,⁴⁹ ewastools,⁵⁰ and minfi^51,52 R packages, to account for dye-type bias and probe type differences on the array and to assess bisulfite conversion and individual sample quality.⁴⁸ Sex was defined as concordant self-reported sex and chromosomal sex estimated from DNAm intensities on the X and Y chromosomes (XX for self-reported females, XY for self-reported males). DNAm data was noob-corrected and normalized using functional normalization (the preprocessFunnorm function) from the minfi package.^51,52 All the following analysis was performed using R 4.2.2 unless otherwise stated. On the EPICv2, there are multiple probes which assess DNA methylation at the same locus, in either duplicate or more replications. These probes were consolidated by finding the mean DNA methylation beta value across all replicates. Variability across microarrays is common, and should be assessed. Four technical replicates were included across the microarrays, two for venous blood and two for capillary blood. The correlations across CpG sites between the technical replicate and its originator sample, as well as the root-mean-square error (RMSE) were assessed in both raw data and pre-processed, normalized data. Pearson correlations between technical replicates was >0.99 both prior to and after normalization. RMSE values were <0.035 for raw data, and <0.03 for normalized, pre-processed data across four pairs of technical replicates.

Quantification and statistical analysis

Estimation of immune cell proportions

DNA methylation-estimated cell type proportions were calculated using the most high-resolution blood cell estimation reference to date, which predicts 12 blood cell type proportions.¹⁴ We used the automatic probe select option, which identifies the top 100 most informative probes per cell type for cell type deconvolution. Firstly, cell type proportions were calculated with the combined capillary and venous blood samples to identify whether these blood samples differ in blood cell type proportion. Secondly, predicted cell type proportions were estimated for capillary and venous samples separately. The coefficients and probes selected were compared between different blood sample types. Paired t-tests were used to compare the mean measure per predictor between capillary and venous blood sample types.

Clustering and principal components analysis

Unsupervised hierarchical clustering with the R stats package⁵³ was performed on all probes to ascertain whether samples clustered by individual, regardless of blood draw origin, to confirm that individuals clustered together.⁴⁸ In addition, principal components analysis (PCA) was performed on all probes to ascertain the principal underlying factors of variation between capillary and venous blood sample types.⁵⁴

DNA methylation-based tools estimation

Noob and funnorm-corrected data were used as input for the Horvath Online Calculator⁵⁵ to generate epigenetic age estimates. The following DNA methylation-based lifestyle, health, and inflammation predictors were calculated for comparison between venous and capillary blood: Horvath epigenetic clock (DNAmAge),⁵⁵ Hannum epigenetic (Hannum DNAmAge) clock,⁵⁶ PhenoAge clock (PhenoAge),⁵⁷ GrimAge clock (GrimAge),⁵⁸ Skin and Blood clock (Skin and Blood),⁵⁹ DunedinPACE predictor (Pace of Aging),¹⁰ DNAm-IL-6,¹⁶ DNAm-CRP,⁶⁰ DNAm-BMI,⁶¹ DNAm-alcohol consumption,¹⁹ DNAm smoking measures,¹⁸ and DNAm-High density lipoprotein (HDL) cholesterol and DNAm-Low density lipoprotein (LDL) cholesterol⁶¹ were calculated using R 4.2.2. Spearman correlations were used to assess the concordance of epigenetic and chronological age. Epigenetic age acceleration (EAA) for all epigenetic clock measures was calculated as the residual from the regression of epigenetic age against chronological age. EAA was calculated with the venous and capillary samples together, and separately, and was compared for all samples. Paired t-tests were used to compare the mean differences per predictor between capillary and venous blood sample types.

ELOVL2 methylation comparison

Increased methylation of CpGs in ELOVL2 is well-established as a marker of increasing age.^28,29,30 To test whether blood source influences the association between age-related methylation changes to DNA methylation sites, we compared the DNA methylation at fourteen ELOVL2 CpGs (cg24724428, cg21572722, cg16867657, cg22143569, cg05446010, cg15557662, cg03577977, cg25151806, cg16323298, cg01799681, cg01142742, cg14153064, cg09017996, and cg02759911) across all venous and capillary samples with t-tests, multiple test corrected using the Benjamini-Hochberg method.

Comparing paired blood source DNA methylation

To comprehensively compare venous and capillary blood DNA methylomes across all the measured CpGs, we conducted an EWAS where blood draw source was the variable of interest. Probes on X and Y chromosomes were removed (24,518 probes), and autosomal variable CpG sites were identified with both blood sources of DNA methylation were combined. CpG beta values with ≥5% variability within 5th and 95th percentiles of the range of beta values across all CpG probes were retained as variable sites (536,514 sites). The EWAS was conducted with a linear model which adjusted for age, sex, and predicted B naive and CD8 naive T cell proportions. The Benjamini-Hochberg method for multiple test correction was applied to the resulting p-values from the model to determine the False Discovery Rate (FDR) of each CpG. CpGs with an absolute regression coefficient >0.05, greater than the RMSE between technical replicates, and which met FDR <0.05 were retained as “significant hits”, analogous to an experimental analysis.

Comparing sex- and age-related DNA methylation

Sex and age are salient biological variables, identical for venous and capillary blood at the time of blood draw, which are known to influence DNA methylation. Independent epigenome-wide association studies (EWASs) for capillary and venous blood samples were conducted to assess the similarity of influence of these factors on the blood methylome. We first applied a variability filter and identified autosomal CpG probes with beta values with ≥5% variability within 5^th and 95^th percentiles of the range of beta values across all CpG probes (497,993 in venous blood, 505,234 in capillary blood) on the EPIC v2. Only variable CpG probes common to both venous and capillary blood were retained for analysis (463 661 CpGs). Secondly, we performed independent linear regression EWASs on the betas of the common, variable CpG probes, with sex as a main effect, while accounting for age and predicted B naive and CD8 naive T cell proportions, as these two cell type proportions were not strongly correlated with each other (r < 0.3) nor with the age and sex of the participants. The two cell type proportions had a low correlation with each other (r < 0.3) in each blood sample type to avoid multicollinearity in the linear regression model. Kendall’s tau correlation was used to assess the linear regression coefficient of each CpG from either venous or capillary blood for both age and sex. The same thresholds as above (FDR <0.05, estimate > |0.05|) were considered the threshold for significance and compared across venous and capillary blood samples.