Remote Blood Collection for Telomere Length Measurement: Assessing the Impact of Sample Characteristics and Handling on DNA Quality and Assay Outcomes

ABSTRACT

Background

Telomere length (TL) is a valuable marker of aging and stress that reflects both genetic and environmental influences. Quantitative PCR (qPCR) TL measurement is a powerful and cost‐effective assay, especially in population studies with limited quantities of source material. Nevertheless, collecting and transporting high‐quality blood samples can be logistically challenging, and research suggests that several preanalytical and analytical factors can influence the reliability and precision of the qPCR assay. Here we describe a procedure for collecting blood remotely in a large‐scale study. We then assess the influence of various features of the samples, as well as their collection, transportation, and storage on DNA quality and TL assay outcomes.

Method

Participants used at‐home collection kits to collect a few drops of whole blood in BD Microtainers during a baseline (n = 265) and 1‐year follow‐up (n = 178) assessment. DNA was extracted using a magnetic‐bead method, and DNA yield, purity, and integrity were assessed. TL was measured using qPCR. To assess inter‐assay variation, the coefficient of variation (CV) was calculated across repeated TL measurements (three runs) for each sample. When there was adequate material for duplicate extractions of DNA from the same blood samples, we calculated the intra‐class correlation (ICC) of the resultant TL values to assess assay precision.

Results

Our analyses revealed that as little as 50 μL of blood yielded sufficient DNA for highly precise TL measurement (ICC = 0.962, n = 365). Transportation time and an additional year of storage time at −80°C did not meaningfully affect DNA quality or assay outcomes. However, blood clotting was associated with longer telomere estimates, whereas greater temperature exposure was related to shorter telomere estimates.

Conclusions

We established that whole blood collected remotely in BD Microtainers can provide a valid sample source for qPCR TL measurement. We also outline important logistical considerations related to sample collection and handling and provide recommendations for researchers who want to use this method.

1. Introduction

Telomeres—the protective complexes at the ends of eukaryotic chromosomes—are made up of short tandem DNA repeats and associated proteins (Blackburn 2001). Telomeres naturally shorten with age through successive rounds of DNA replication, which leads to cell cycle arrest and senescence when telomeres become critically short (Chan and Blackburn 2002). This process can also be accelerated by inflammation (Jose et al. 2017) or reactive oxidative stress (Ahmed and Lingner 2018; Fouquerel et al. 2019). As such, telomere length (TL) variation over time is shaped both by genetic or inherited factors and by non‐genetic, environmental exposures and lived experience (Broer et al. 2013; Factor‐Litvak et al. 2017; Lin and Epel 2022; Shalev et al. 2013).

Given these properties, TL has emerged as a useful biomarker of cellular aging, major diseases, and associated risk factors (Aviv and Shay 2018; Blackburn et al. 2015; Cheng et al. 2021; De Meyer et al. 2018; Haycock et al. 2017; Ridout et al. 2017; Ruiz et al. 2022; Schneider et al. 2022; Wang et al. 2021; Willis et al. 2018). Chronic psychological stress, which is an important contributor to many physiological disease states, has also been linked to shorter TL in some, but not all, studies (reviewed in Epel and Prather 2018; Lin and Epel 2022; Mathur et al. 2016). Conversely, stress reduction and meditation training have been associated with better telomere maintenance (reviewed in Conklin et al. 2019).

The advent of quantitative PCR (qPCR) made it possible to obtain average TL measurements from sample sources with small quantities of material (Cawthon 2002, 2009). TL measurements derived from the qPCR assay (qPCR TL) are expressed as a ratio: the quantity of Telomere repeat (T) signal relative to the quantity of a Single copy gene (S) signal, or T/S. Because the qPCR TL assay is inexpensive and relatively easy to execute, it has been used to facilitate large population‐based studies that typically require hundreds of samples or more—a prominent example being the UK Biobank study, which used qPCR to measure 462,666 samples (Codd et al. 2022). This large‐sample requirement has been a motivator for developing the most reliable, standardized qPCR TL measurement techniques possible.

From sample collection to data analysis, there are many preanalytical and analytical factors that can contribute to variability in qPCR TL measurements (reviewed in Lin et al. 2019; Lindrose et al. 2021). With this in mind, the Telomere Research Network (TRN), funded by the National Institute of Aging and the National Institute of Environmental Health, has developed guidelines for TL measurements that address sample collection and storage, the TL assay, and data analysis (trn.tulane.edu). In a recent multi‐lab study, the TRN validated the use of the intra‐class correlation (ICC) coefficient as a metric of assay precision for TL estimates taken from repeated DNA extractions from the same sample (Lin et al. Geroscience, manuscript in revision). Because the ICC of repeated extracts captures the entire assay process, it is recommended by the TRN as a metric of assay performance.

Sample source selection is a critical factor in collecting pure DNA and reliable TL data. DNA is commonly extracted from a variety of sources, including saliva, dried blood spots, and whole blood. Saliva carries several benefits as a sample source because it is non‐invasive and can be gathered passively or with a swab. Saliva also poses no risk of blood‐borne pathogen exposure, and once in the lab, saliva samples are easy to store at room temperature. However, DNA extracted from saliva may contain nucleic acid contamination from non‐human sources, and some amount of degradation is common with saliva DNA (Smith et al. 2022; Wolf et al. 2024). Dried blood spots (DBS) represent another possible sample source. Using this method, small drops of blood are gathered from a finger or heel‐prick and deposited directly onto filter paper, which negates the need for a phlebotomist. However, only small amounts of blood can be gathered this way, and the DNA extracted from these samples can be of low yield or low quality. The advantages and disadvantages of DBS cards are discussed extensively in Rej et al. (2021). Importantly, a recent study comparing telomere measurements from a selection of blood‐based and oral tissues taken from the same cohort found that both saliva and DBS had lower DNA integrity and yield than other blood‐based sample sources (Wolf et al. 2024). Given these limitations, whole blood remains the most attractive tissue source for extracting reliable amounts of good‐quality DNA. However, whole blood collection typically requires a venous blood draw by a trained phlebotomist, and the blood collected needs to be cooled and stored quickly, making it logistically challenging for population studies with many participants.

In addition to these considerations, the travel restrictions imposed during the COVID pandemic highlighted the importance of remote sample collection methods, which allow participants to collect samples at home and ship them directly to research sites. These circumstances motivated us to develop a new protocol for remote blood collection using Becton Dickinson Microtainer tubes, which contain EDTA to stabilize blood and minimize clotting. In this report, we aim to determine the utility and potential limitations of this remote collection method by assessing the DNA quality, telomere outcomes, and assay precisions from blood samples collected in this manner.

Samples were collected as part of the Contemplative Coping during COVID‐19 (CCC) study—a longitudinal study that followed participants across a 1‐year period of the COVID‐19 pandemic. The primary goals of the CCC study were to understand the types of pandemic‐related stressors individuals experienced, whether they drew on contemplative practices, such as meditation, to cope with those stressors, and if engagement with contemplative practice promoted better biological and psychological outcomes—as measured by blood cell TL and self‐report measures of stress, mental health, and well‐being (Conklin et al. 2023). As part of the CCC study, hundreds of nationally dispersed participants donated blood for TL measurement at the beginning and end of this 1‐year longitudinal study, making it a good sample set to test our proposed methods for obtaining precise TL measurements from remotely collected samples. This design allowed us to assess the repeatability of the qPCR TL assay across duplicate extracts as well as its reliability across two time points. We also assessed how features of the blood samples, as well as their collection, transportation, and storage impacted the DNA quality and TL assessments. We demonstrate that blood samples collected remotely in microtainers can lead to high‐quality DNA extractions and high‐precision qPCR TL assay results.

2. Materials and Methods

2.1. Participants

The UC Davis team recruited a national cohort of meditation practitioners from the USA for a study investigating relationships between meditation practice, stress exposure, stress coping, and TL during the COVID‐19 pandemic (Conklin et al. 2023; VandenBos et al. 2025). To be included, participants had to be at least 18 years old, currently residing in the United States, and have some previous meditation experience. A total of 836 people applied to participate in the study, 608 were invited to participate by the research team, and 389 started the baseline assessment. Of those who started the baseline assessment, 335 completed an initial set of questionnaires and were shipped a blood collection kit. A total of 265 of those participants were able to collect and return a usable blood sample. Participants were paid $25 for each sample collection, and all procedures were approved by the UC Davis Institutional Review Board. Demographics for the participants who provided telomere data are reported in Table 1. See the Supporting Information for a CONSORT diagram depicting participant and sample exclusion criteria, and for further details regarding the demographic variables collected as well as reported demographics for the full study sample (Table S1).

TABLE 1.

Demographics for participants with telomere measurements.

Variable	n	M (SD)
Age	265	50.8 (15.5)
BMI at baseline	264	25.1 (5.4)
BMI at 1 year	165	24.9 (4.8)

	n	%
Ethnicity
White	181	68.3
Indigenous, Pacific Islander, or Multi‐ethnic	31	11.7
Asian	29	10.9
Black, African American or Afro‐Caribbean	15	5.7
Hispanic or Latino	7	2.6
Prefer not to say	2	0.8
Sex
Female	203	76.9
Male	59	22.4
Non‐binary/third gender	2	0.8
Prefer not to say	1	0.0
Gender
Woman	193	72.8
Man	58	21.9
Gender expansive	14	5.3
Sexual orientation
Heterosexual	188	70.9
LGBQA	77	29.1
Education
High school diploma or equivalent	3	1.1
Some college	23	8.7
College diploma	61	23.0
Some graduate/professional school	29	10.9
Graduate/professional degree	149	56.2
Income
$0–50 k	49	18.5
$50–100 k	90	34.0
$100–150 k	51	19.3
$150–200 k	26	9.8
$200 k+	49	18.5

2.2. Sample Collection

Participants were mailed at‐home collection kits, which included detailed instructions, lancets, and microtainers for sample collection, and alcohol wipes, gauze, and adhesive bandages to prepare and cover the puncture site. These materials were mailed in an insulated shipping container, along with cold packs and a pre‐paid shipping label to return the samples by mail (see Supporting Information for a list of materials included in the collection kit and the instructions provided to participants). Participants were also emailed an instructional video explaining how to collect the sample and were offered an optional one‐on‐one video call if they wanted additional guidance. The microtainer collection tubes (Becton Dickinson, cat # 365992) contained approximately 1.0 mg of K2EDTA per tube to prevent coagulation and DNA degradation and were demarcated at 400 and 600 μL. Participants were instructed to collect between 400 and 600 μL of blood if possible and to send back whatever amount of material they were able to collect if less than this specified amount. The volume of blood collected ranged from < 25 μL to over ~1000 μL (see Figure S1). Participants were instructed to place their sample in the insulated shipping container immediately after collecting it, along with frozen cold packs, and to take the package directly to a FedEx facility for same‐day shipping. Once received, samples were stored in a −80°C freezer at UC Davis. Baseline samples were collected between June of 2020 and January of 2021, and 1‐year follow‐up samples were collected between June of 2021 and February of 2022. Once all baseline and 1‐year samples were collected, the UC Davis team transported the full set of samples to UC San Francisco on dry ice for DNA extraction and TL measurement.

2.3. DNA Extraction and Assessment

Genomic DNA was extracted using the Beckman Coulter GenFind v3 magnetic‐bead‐based kit (cat # C34880/C34881), which involves lysing cells and using magnetic beads to bind and purify DNA. To confirm that we could extract enough DNA from our smaller samples, we ran a small pilot study using 11 samples with as little as 50 μL of blood. The standard GenFind v3 extraction protocol requires a starting volume of 200 μL, so all tubes were topped off to 200 μL with DPBS (Thermo Fisher Scientific, cat # 14190144) before proceeding with cell lysis. This pilot test confirmed that samples with as little as 50 μL of blood yielded sufficient DNA for the TL assay. All samples were eluted in 50 μL elution buffer; therefore, the DNA concentration used in subsequent statistical analyses represents the DNA yield. In our full set of samples, the lowest concentration with passing DNA quality criteria was 7.1 ng/μL, which provided a sufficient amount of DNA for DNA quality assessment (spectrometry and gel analysis) and for three TL assay runs (separate plates) with triplicate wells.

Our goal for the full set of samples was to assess the repeatability, or precision, of the TL assay by duplicating the entire procedure twice for samples with a sufficient volume of blood, and then calculating the ICC for the telomere estimates derived from these duplicate DNA extractions (Eisenberg 2016; Lin et al. Geroscience, manuscript in revision). Thus, all samples with more than 100 μL of blood (n = 399; 231 baseline samples and 168 1‐year samples) were split into two tubes for two separate extractions. Blood volume varied widely across the full set of extraction tubes (see Figure 3A and Figure S3A), but blood volumes for paired tubes from a single sample were roughly equivalent.

FIGURE 3.

Characterization of DNA extractions. (A) Histogram showing variation in starting blood volume for DNA extractions. See Figure S1 for the volume of blood samples collected. Panels (B–F) depict 114 extracted DNA pairs, each with one clotted (gray) and one un‐clotted (white) sample that met the DNA purity criteria of OD₂₆₀/OD₂₃₀ ≥ 1.0 and OD₂₆₀/OD₂₈₀ ≥ 1.7 and ≤ 2.0. (B) DNA concentrations were typically higher in extracts taken from samples with a larger volume of starting material and in tubes containing blood clots compared to those with no clots. (C) DNA concentration (ng/μL) was higher in clotted samples (p < 0.0001). (D) OD₂₆₀/OD₂₃₀ was lower in clotted samples (p < 0.0001). (E) OD₂₆₀/OD₂₈₀ was higher in clotted samples (p < 0.001). (F) Clotted extracts were also associated with longer T/S estimates (p < 0.0001). (G) Representative image of the agarose gels run to assess DNA integrity, which shows no signs of degradation. ***p < 0.001; ****p < 0.0001.

Despite the presence of EDTA in the Microtainer collection tubes, some samples contained blood clots, which prompted us to investigate the impact of clotting on DNA quality and telomere assay outcomes. Before splitting the sample into duplicate tubes for DNA extraction, each blood sample was judged to be “clotted” (35.4%) or “un‐clotted” (64.6%). Clotted samples could be difficult to pipet, and sometimes only one tube within a pair contained a clot (usually the second, “transfer” tube). Other times, both tubes contained a partial clot.

The quantity of DNA purified in each extraction was measured through spectrophotometry, using the UV–Vis Nanodrop function (NanoDrop 2000/2000c, Thermo Fisher Scientific, cat # ND2000CLAPTOP). The purity of each DNA sample was inferred using the UV–Vis Nanodrop function. Extracts were judged to be relatively free of contaminants if the OD₂₆₀/OD₂₃₀ ratio was greater than or equal to 1.0 and the OD₂₆₀/OD₂₈₀ ratio was greater than or equal to 1.7 but less than or equal to 2.0. The integrity of each extract was further evaluated by visual inspection: samples normalized for DNA concentration (100 ng of DNA/lane) were run on 0.8% agarose gels (80 mL solidified agarose with 2 μg/mL ethidium bromide) at 80 V for 50 min with HyperLadder 1 kb markers (Meridian Bioscience, cat # BIO‐33053). Two operators (DLS and JL) visually inspected each gel and independently judged the DNA to be “intact” or “degraded.” All samples were determined to be fully intact, and no sample was judged as showing signs of degradation.

Extracted DNA was stored at −80°C until TL measurements were made. The DNA extractions were completed within a 3‐week period at the beginning of 2023. TL measurements were made during an overlapping 3‐week period, such that all samples were assayed within 3 weeks of extraction.

2.4. qPCR TL Assay

The standard qPCR TL assay used was first developed by Cawthon in 2002 and is graphically illustrated in Figure 1A. Our assay workflow is outlined in Figure 1B. This method involves measuring the amplification of Telomere repeats (T) relative to the amplification of a Single copy gene (S) in two separate qPCR reactions and then taking the resultant T/S ratio (illustrated in Figure 1A).

FIGURE 1.

Telomere length qPCR assay depiction. (A) qPCR TL measurement involves PCR amplification and quantification of Telomeric DNA (T) and DNA from a Single copy gene (S) with appropriate primers. (B) The TL qPCR assay workflow, as described in the text. In brief, DNA is denatured by heating at 96°C for 10 min on a 96 well plate before being dispensed to two 384 wells (one for T and one for S). We used 4 μL of normalized DNA and added 36 μL buffer (20 mM Tris–HCl, pH 8.4; 50 mM KCl; and 6 ng/μL Escherichia coli DNA) before denaturation.

As depicted in Figure 1B, DNA for each extract was first normalized to 20 ng/μL across 10 individual 96‐well plates and stored at −80°C. In cases where the extract yielded less than 20 ng/μL of DNA, the original DNA concentration was used, provided that the measured concentration was higher than the second‐lowest concentration on the standard curve (0.97 ng). No samples were excluded at this stage. Duplicate DNA samples were always pipetted to the same 96‐well source plate. Paired baseline and 1‐year DNA samples were also positioned together on the same plate. We denatured and dispensed the DNA and appropriate (T or S) reaction mixture into triplicate wells on 384‐well plates for every sample. Identical pipetting patterns were carried out for the “T” and “S” reaction plates.

The primers for the telomere PCR were tel1b (5′‐CGGTTT(GTTTGG)5GTT‐3′), used at a final concentration of 100 nM, and tel2b (5′‐GGCTTG(CCTTAC)5CCT‐3′), used at a final concentration of 900 nM. The primers for the single‐copy gene (human beta‐globin) PCR were hbg1 (5′‐GCTTCTGACACAACTGTGTTCACTAGC‐3′), used at a final concentration of 300 nM, and hbg2 (5′‐CACCAACTTCATCCACGTTCACC‐3′), used at a final concentration of 700 nM. The final reaction mix contained: 20 mM Tris‐hydrochloride (pH 8.4); 50 mM potassium chloride; 200 μM each of dATP, dCTP, dGTP, and dTTP; 1% dimethyl sulfoxide; 0.4× SYBR green I; 22 ng Escherichia coli DNA; 0.4 units of platinum Taq DNA polymerase (Invitrogen Inc., Carlsbad, CA); and approximately 6.6 ng of genomic DNA per 11 μL reaction.

A Roche Lightcycler 480 was used to carry out the T and S monoplex reactions on the 384‐well plates. The telomere (T) thermal profile consisted of denaturing at 96°C for 1 min followed by 30 cycles of denaturing at 96°C for 1 s and annealing or extension at 54°C for 60 s with fluorescence data collection. The single copy gene (S) thermal profile consisted of denaturing at 96°C for 1 min followed by eight cycles of denaturing at 95°C for 15 s, annealing at 58°C for 1 s, and extension at 72°C for 20 s followed by 35 cycles of denaturing at 96°C for 1 s, annealing at 58°C for 1 s, extension at 72°C for 20 s, and holding at 83°C for 5 s with data collection. The PCR efficiency was 91.3% ± 1.6% for the telomere reaction and 93.7% ± 1.8% for the single copy gene reaction.

The T/S ratio for each sample was measured in triplicate runs (i.e., on three separate plates for both the T and S reactions), with triplicate wells for each reaction. The triplicate wells containing samples were interspersed with DNA standards and control samples throughout each plate (Figure 1B). Each PCR run contained a three‐fold serial dilution of a commercial human genomic DNA (Sigma‐Aldrich, cat #11691112001) containing 26, 8.75, 2.9, 0.97, 0.324, and 0.108 ng of DNA as the reference standard. The average R ² across assays was 0.9999 for both T and S independently. The quantity of targeted templates in each sample was determined relative to the reference DNA by the maximum second derivative method in the Roche LC480 program. Efficiency estimates were based on the standard curves provided by the Roche LC480 program. Dixon's Q test with a cutoff of > 0.66 was used to exclude up to one T and one S outlier from each set of triplicate wells: 13.3% of wells from the T reaction and 15.1% of wells from the S reaction were removed. The T/S ratio for each sample was calculated from the average of the T and S triplicate wells after outlier removal. We used 8 DNA samples extracted from human cancer cell lines as controls. Each control sample was run twice in quadruplex wells (8 samples × 4 replicates × 2 runs = 64 wells of controls per 384 well plate), resulting in 16 T/S values from each assay plate. The average of each of the two sets of eight samples across thirty 384‐well assay plates was used to derive a normalizing factor, and then the average of the 16 normalizing factors was used to normalize each run. The average CV of the control samples was 4.6% ± 0.9% before normalization. After normalization, by definition and design, the average CV was 0%. All assays for the study were performed using the same lots of reagents. Lab personnel who performed the assays were provided with de‐identified samples and were blind to all demographic data.

2.5. Data Analysis

We used the average coefficient of variation (CV) of the T/S values across the three assay runs (separate plates) for every sample as a measure of inter‐assay variation. Because most of our samples contained sufficient material to repeat the entire TL assay for duplicate DNA extractions, we also calculated the ICC as a measure of assay precision, or repeatability. Following TRN guidelines, T/S values were z‐scored by extraction before ICCs were calculated. CVs were calculated in Microsoft Excel and ICCs were calculated in R version 4.4.0 (2024‐04‐24) and R studio Version 2024.04.2 + 764 using the “rpt” function from the rptR package (Stoffel et al. 2019). ICC confidence intervals were estimated by parametric bootstrapping (n = 1000).

We assessed the impact of various aspects of the assay methodology on DNA concentration and purity, telomere estimates, and the external validity of the telomere estimates. We analyzed features of the samples (e.g., clotting) and the collection, transportation, and storage of the samples using a series of ICC calculations, Wilcoxon tests, correlations, and linear mixed effects models. Variables were assessed for |skew| < 1 and |kurtosis| < 3. Pearson correlations were computed when both variables were normally distributed (i.e., blood volume, OD₂₆₀/OD₂₈₀, T/S values, and temperature, which was bimodally distributed), and Spearman correlations were computed if either variable was skewed, kurtotic, or ordinal (i.e., BMI, DNA concentration, OD₂₆₀/OD₂₃₀, transportation duration, and cold pack condition). Correlations were computed in R using the “cor.test” function. 95% confidence intervals were calculated for Pearson's correlations but not for Spearman's correlations. Wilcoxon tests were conducted in GraphPad Prism 10 (Version 10.3.1). Multilevel models were used to assess combinations of person‐, extract‐, and sample‐level covariates while estimating the data dependencies (random effects) among samples nested within participants and among extracts nested within samples taken from different timepoints. Linear mixed effects models were fit with the “lmer” function from the lme4 package in R. For each set of predictors, we determined the best fitting model by comparing fit statistics for models with alternative combinations of predictors using the “dredge” function from the MuMIn package (Bartoń 2024). Models were ranked according to AICc. Each set of predictors was checked for multicollinearity using the Variance Inflation Factor or “vif” function from the car package.

Our predictions regarding the impact of blood volume, transportation duration, temperature, and cold pack condition on DNA quality and T/S ratios were pre‐registered on the Open Science Framework: https://osf.io/5gf62. The remaining analyses were exploratory, and changes to the pre‐registered analysis plan are discussed in the Supporting Information. The code for the full set of analyses conducted in R can be found on the pre‐registration page.

3. Results and Discussion

Researchers have long sought a DNA sample source that can be gathered easily and non‐invasively from geographically distributed participants, while also preserving sample quality. Inspired by these criteria, the CCC study was designed to collect hundreds of high‐quality, moderately sized, fresh blood samples that were kept cold and shipped immediately after collection. Here we report participants' success rates for collecting their own samples in this manner, the quality of the blood samples achieved, and the external validity and precision of the resulting telomere measurements. We also assessed how several aspects of the samples and their collection, transportation, and storage influenced these results.

3.1. Sample Collection Success Rates

Of the 335 participants who were shipped a collection kit during the baseline assessment, 275 (82.0%) returned a blood sample, of which 265 (96.4%) were usable (i.e., contained enough blood and were not delayed in transit by a day or more). However, this required the shipment of an additional 83 replacement kits to participants who had difficulty collecting enough blood with the finger puncture method (~60 people) or whose samples were delayed in transit (~23 people). Other participants opted out of the blood draw based on blood phobia, the logistics required to return the blood sample, or hesitancy about research involving their DNA (~10 people stated a reason for opting out). These numbers are approximate given that some participants experienced multiple issues, and our early records did not differentiate each issue.

A total of 323 people were contacted to participate in the 1‐year follow‐up assessment. Of those, 245 (75.9%) agreed to collect a blood sample and were shipped a collection kit; 28 opted out of the blood collection (often due to difficulty collecting a sample at the baseline assessment), and 51 did not reply. Of the 245 who agreed to collect a sample, 197 (80.4%) participants returned a sample and 178 (90.3%) of those were usable. Again, several participants could not collect enough material (20) or experienced shipping delays (25), so 37 replacement kits were sent during this assessment (see CONSORT Diagram in the Supporting Information).

In summary, of the 389 people who formally enrolled in the study, 68.1% provided a viable blood sample at baseline and 45.8% provided viable follow‐up samples 1 year later. At each time point, approximately 75% of people who had progressed to that stage of the study and agreed to collect a sample were successful, and 93% of the samples returned were usable. Others agreed to collect a sample and then changed their minds because they did not want to collect it themselves or because the logistical demands of collecting and returning the sample were too complex. Between 10% and 20% of people who tried to collect a sample had trouble collecting enough blood. To mitigate this issue, we adjusted the kits to include enough materials for people to make up to three collection attempts, if they were willing, which reduced the number of replacement kits sent during the 1‐year follow‐up. Approximately 25 samples were rendered unusable at each time point due to shipping delays.

Researchers planning to use this sample collection method can use these figures as guidance to estimate the number of participants they should aim to recruit, the number of materials to budget for, and some of the problems they might encounter. However, it is worth keeping in mind that these samples were collected during the height of the COVID‐19 pandemic, when many people were sheltering in place. These circumstances may have given some people more or less time to participate in these research activities, while also increasing shipping demands and contributing to the shipping delays observed.

3.2. Blood Collected Remotely in BD Microtainers Yields a Sufficient Quantity of High‐Quality DNA

We determined that extractions from as little as 50 μL of blood yielded enough DNA (M = 44.8 ng/μL, range: 6.5–163.9 ng/μL) to perform DNA quality assessments and for three runs of the T and S qPCR reactions, with plenty of material left for repeat runs if necessary. Of the 443 viable samples collected, 399 samples had sufficient material for duplicate DNA extractions, while 44 samples could only be extracted once, which resulted in a total of 842 usable extracts.

Of the 842 usable extracts, 37 failed to meet the DNA quality criteria: six failed to meet the OD₂₆₀/OD₂₃₀ criteria of ≥ 1.0 (three from baseline and three from 1 year), and 33 failed to meet the OD₂₆₀/OD₂₈₀ criteria of ≥ 1.7 and ≤ 2.0 (24 from baseline and nine from 1 year)—resulting in a total of 805 extracts that met both DNA purity standards and 365 extract pairs that met these standards (see CONSORT Diagram in the Supporting Information for further details). Unless specified, the analyses presented in this manuscript include only those extracts that met these DNA purity standards; parallel analyses for the full set of extractions are presented in the Supporting Information for comparison.

Descriptive statistics for the DNA metrics and extract level T/S values are presented in Table 2 and Table S2. Descriptive statistics for the variables related to sample handling and sample‐level T/S values are presented in Table 3 and Table S3.

TABLE 2.

Extract‐level descriptive statistics for extracts that met DNA purity criteria.

Variable	n	Mean	SD	Median	Min	Max	Skew	Kurtosis	SE
Blood volume used for extraction (μL)	805	139.3	61.1	150.0	15.0	200.0	−0.3	−1.6	2.2
DNA concentration (ng/μL)	805	45.29	28.42	38.50	6.90	163.90	1.20	1.43	1.00
OD 260/280	805	1.9	0.1	1.9	1.7	2.0	−0.6	0.9	0.0
OD 260/230	805	2.6	1.5	2.4	1.1	28.0	10.9	149.7	0.1
Extract average T/S (across runs)	805	0.83	0.18	0.80	0.42	1.46	0.53	0.24	0.01

TABLE 3.

Sample‐level descriptive statistics for extracts that met DNA purity criteria.

Variable	n	Mean	SD	Median	Min	Max	Skew	Kurtosis	SE
Estimated blood volume (μL)	438	290	181	250	15	1000	0.6	−0.1	9
Baseline	261	274	176	250	15	1000	0.6	−0.1	11
1‐year	177	313	186	300	15	900	0.7	−0.1	14
Transportation duration (hours)	414	25.4	4.5	25.3	2.2	54.7	1.0	10.1	0.2
Baseline	239	26.3	4.7	25.9	2.2	54.7	1.0	12.4	0.3
1‐year	175	24.2	3.7	23.9	13.8	42.8	0.7	2.5	0.3
Max temperature in collection location (°F)	409	75.8	17.8	79.5	25.0	115.0	−0.6	−0.3	0.9
Baseline	244	75.7	18.1	79.8	25.0	113.0	−0.6	−0.3	1.2
1‐year	165	76.1	17.4	79.3	28.9	115.0	−0.5	−0.4	1.4
Max temperature in Davis on delivery day (°F)	440	84.4	15.3	89.6	46.4	107.1	−0.7	−0.8	0.7
Baseline	262	83.8	13.2	89.6	53.6	104.0	−1.0	−0.4	0.8
1‐year	178	85.2	17.9	93.9	46.4	107.1	−0.6	−1.3	1.3
Cold pack condition	437	3.1	0.8	3.0	1.0	4.0	−0.5	−0.1	0.0
Baseline	260	2.0	0.8	2.0	1.0	4.0	−0.6	−0.2	0.1
1‐year	177	1.9	0.7	2.0	1.0	3.0	−0.2	−0.8	0.1
Sample average T/S (across extracts)	440	0.82	0.18	0.80	0.43	1.46	0.53	0.19	0.01
Baseline	262	0.83	0.18	0.80	0.43	1.45	0.54	0.13	0.01
1‐year	178	0.82	0.18	0.80	0.45	1.46	0.53	0.22	0.01

Note: Davis, California, USA is the location of the lab where collected samples were delivered and stored. The condition of each cold pack was coded as (1) frozen, (2) partially frozen, (3) melted and cold, or (4) melted and warm.

3.3. External Validity of the TL Estimates

We next assessed the external validity of our TL estimates. First, we examined the relationship between sample T/S values at baseline (averaged across all runs and extracts for a given individual) with those at 1 year, and found these values to be highly correlated, r ₍₁₇₃₎ = 0.95, p < 0.001, 95% CI (0.93, 0.96). We also examined relationships between sample T/S values at baseline and 1 year in relation to age, BMI, and sex. As expected from previous literature (Müezzinler et al. 2013), we found moderate negative correlations between TL and age at baseline, r ₍₂₆₀₎ = −0.58, p < 0.001, 95% CI (−0.65, −0.49), and 1 year, r ₍₁₇₆₎ = −0.53, p < 0.001, 95% CI (−0.63, −0.42). We also found the expected negative relationship between TL and BMI (Müezzinler et al. 2014) at baseline, r _s = −0.21, p = 0.002, and 1 year, r _s = −0.23, p = 0.003. Consistent with prior research (Lansdorp 2022), t‐tests also revealed that males (M = 0.74, SD = 0.16) had significantly shorter telomeres than females (M = 0.85, SD = 0.18) at baseline, t ₍₂₅₇₎ = 4.35, p < 0.0001; this was also true for samples collected at 1 year (males: M = 0.71, SD = 1.3; females: M = 0.85, SD = 0.18; t ₍₁₇₃₎ = 4.46, p < 0.0001). These results confirm that the TL estimates derived from this method display the expected patterns commonly observed in the literature.

3.4. DNA Concentration and Blood Clotting Predict DNA Purity Measures and T/S Estimates

As might be expected, starting blood volume for each extract was positively correlated with the final concentration of extracted DNA, r _s = 0.47, p < 0.001. OD₂₆₀/OD₂₃₀ ratios were negatively correlated with both blood volume, r _s = −0.28, p < 0.001, and DNA concentration, r _s = −0.42, p < 0.001 (see Figure 2A and Figure S2A). By contrast, OD₂₆₀/OD₂₈₀ values were positively correlated with both blood volume, r ₍₈₀₃₎ = 0.24, p < 0.001, 95% CI (0.17, 0.31), and DNA concentration, r _s = 0.41, p < 0.001 (Figure 2B and Figure S2B). After inspecting the plot in Figure 2B, we also conducted a polynomial regression, which revealed significant linear, β = 0.31, 95% CI (0.24, 0.37) and quadratic, β = −0.10, 95% CI (−0.14, −0.06), effects of OD₂₆₀/OD₂₈₀ values on DNA concentration. There was no significant correlation between extract level T/S values and starting blood volume, r ₍₈₀₃₎ = 0.03, p = 0.330, 95% CI (−0.03, 0.10), but there was a small negative correlation between T/S values and DNA concentration, r _s  = 0.08, p = 0.030 (Figure 2C and Figure S2C). However, when we recomputed this latter correlation after excluding extracts with a variable starting concentration of less than 20 ng/μL of DNA, there was no longer a significant correlation between T/S values and DNA concentration, r _s = 0.02, p = 0.674. See Section 3.8 for further discussion of the impacts of variability in starting DNA concentrations on the estimation of T/S values.

FIGURE 2.

DNA quality and extract T/S values by DNA concentration for extracts that met DNA purity criteria. Plots show Spearman's correlations between DNA concentration and (A) OD 260/230 values, (B) OD 260/280 values, and (C) extract T/S values.

In the full set of extracts, there were 162 extracts with clots (19.2%), including 127 extract pairs in which one tube contained a clot while the other did not, and 114 pairs in which both DNA extracts met the nanodrop purity criteria. Wilcoxon tests revealed that DNA in the clotted samples (57.3 ± 24.4 ng/μL) was significantly more concentrated than in unclotted tubes (34.6 ± 19.9 ng/μL, p < 0.0001; see Figure 3B,C and Figure S3B,C). The average OD₂₆₀/OD₂₃₀ ratio was reduced with clotting (2.18 in clotted compared to 2.47 in unclotted; p < 0.0001, see Figure 3D and Figure S3D), whereas the average OD₂₆₀/OD₂₈₀ ratio was slightly higher in the presence of clotting (1.89 for clotted compared to 1.87 for unclotted, p < 0.001, Figure 3E and Figure S3E). Similarly, a paired‐sample t‐test revealed that clotted extracts (M = 0.834) had significantly higher T/S values than unclotted extracts (M = 0.798), t ₍₁₁₃₎ = 7.71, p < 0.0001 (Figure 3F and Figure S3F).

We also assessed whether the presence of clots affected CVs, ICCs, or relationships between sample T/S values and measures of external validity. Using the 730 duplicate extracts that met our DNA criteria (258 with clots and 472 without), we determined that the CVs across three runs were equivalent for the clotted and unclotted samples (Table 4). However, the ICC for the 236 unclotted extract pairs was numerically larger than for the 129 pairs where one or both tubes contained a clot (Table 4, Table S4). When we compared the magnitude of association between sample T/S collected at baseline and 1 year, we observed a weaker correlation between timepoints for samples that contained clots, r ₍₇₉₎ = 0.93, p < 0.001, 95% CI (0.89, 0.95), than for samples containing no clots, r ₍₉₂₎ = 0.96, p < 0.001, 95% CI (0.95, 0.98), Fisher's t = −1.86, p = 0.031. By contrast, when we compared correlations between sample T/S values and age as a measure of external validity, we found no significant difference between clotted, r ₍₇₈₎ = −0.56, p < 0.001, 95% CI (−0.70, −0.39), and unclotted samples, r ₍₁₈₀₎ = −0.58, p < 0.001, 95% CI (−0.67, −0.48); Fisher's t = −0.217, p = 0.414.

TABLE 4.

The impact of sample clotting on qPCR TL assay reliability and precision for extracts that met DNA purity criteria.

Sample type	# samples	CV	SD of CV	# pairs	ICC	SE	95% CI
All pairs	730	0.031	0.019	365	0.964	0.004	0.956, 0.971
Clots	258	0.031	0.019	129	0.955	0.008	0.937, 0.968
No clots	472	0.031	0.018	236	0.972	0.004	0.964, 0.978

Note: CVs and ICCs of samples with and without clots. The CV of every sample was derived from three runs, and the average CV and SD across each set of sample extracts are reported. To characterize the impact of sample clotting on qPCR TL assay precision, ICCs were calculated for the 365 total duplicate extractions, the 129 pairs in which one or both tubes had a clot, and 236 pairs in which neither tube had a clot.

Overall, these results indicate that while clotting seems to have a minimal effect on the reliability and precision of the qPCR assay, it can affect the accuracy of the final T/S values, even though the results still hold with known patterns of external validity.

3.5. An Extra Year of Sample Storage at −80°C Does Not Affect DNA Metrics or TL Assay Precision

The baseline blood samples in our study were stored for a full year longer than the follow‐up samples from the same participants. This allowed us the opportunity to compare DNA quality and assay precision from blood samples stored at −80°C for 2.5 and 1.5 years, respectively. We found that DNA concentration, OD₂₆₀/OD₂₃₀, and OD₂₆₀/OD₂₈₀ were all statistically indistinguishable for the baseline and 1‐year groups (Figure 4A–C and Figure S4A–C). Similarly, the CV of three runs and the ICC of duplicate extractions indicated that the length of storage time at −80°C had no appreciable impact on the reliability or precision of the telomere assay (Figure 4D and Figure S4D).

FIGURE 4.

The impact of sample storage on DNA quality and assay reliability for extracts that met DNA purity criteria. (A) DNA concentration, (B) OD₂₆₀/OD₂₃₀, and (C) OD₂₆₀/OD₂₈₀ for samples collected and stored at −80°C from the beginning of study (baseline), compared to samples collected and stored at −80°C 1 year later. Data presented include 470 DNA extracts from baseline (416 duplicate DNA extracts from 208 blood samples and 54 single DNA extracts from 54 blood samples) and 335 DNA extracts from the 1‐year follow‐up (314 duplicate DNA extracts from 157 blood samples and 21 single DNA extracts from 21 blood samples). All DNA extracts met the DNA purity criteria of OD₂₆₀/OD₂₃₀ ≥ 1.0 and OD₂₆₀/OD₂₈₀ ≥ 1.7 and ≤ 2.0. (D) The impact of storage time on qPCR TL assay precision, including the average CV across three runs for 416 baseline extracts compared to 314 1‐year extracts, and the ICC of 365 duplicate extractions for 208 baseline pairs compared to 157 duplicate extractions from 1 year later.

3.6. Transportation Time Does Not Affect DNA Purity or T/S Values, but Temperature‐Related Variables Are Negatively Correlated With T/S Values

Next, we examined whether aspects of sample transportation and storage affected assay outcomes. Because our samples were collected remotely and each sample was subject to a different transportation duration, we assessed whether the time from sample collection to storage at −80°C affected any of our DNA quality metrics or T/S estimates. We found no relationships between transportation time and OD₂₆₀/OD₂₃₀ values, r _s = 0.04, p = 0.315, OD₂₆₀/OD₂₈₀ values, r _s = −0.05, p = 0.190, or T/S values, r _s = 0.07, p = 0.13. Collectively, these results suggest that transportation time had negligible effects on assay outcomes, and these findings held when transportation duration outliers were removed. It is worth noting, however, that the variability in transportation times for these samples was relatively constrained (generally 24–34 h; see Table 3), so it may be that longer transportation times could still result in greater degradation.

At each study time point, our samples were collected over a 6‐month window from participants distributed across the United States, meaning that they were collected during different seasons and subject to different temperatures during shipping. As a proxy for these environmental factors, we assessed the influence of the maximum daily temperature in a participant's location on the day they collected their sample, as well as the maximum daily temperature in Davis, CA on the day the sample was delivered to the lab (see Table 3). Although these two temperature variables were highly correlated, r ₍₄₁₉₎ = 0.72, p < 0.001, 95% CI (0.67, 0.76), we suspect that the participants' local temperature is more likely to reflect the climate participants were exposed to, while temperatures in Davis are more likely to reflect the maximum temperature the samples could have been exposed to, since Davis temperatures are often considerably higher than other regions of the US. We also recorded whether the cold packs included with the blood sample were still frozen or had melted during shipping (1 = frozen, 2 = partially frozen, 3 = melted but cold, 4 = melted and warm; Table 3). This categorical variable, while crude, is another proxy for the temperature exposure of the samples during transit.

There was no significant correlation between the maximum temperature in the sample collection location and OD₂₆₀/OD₂₃₀ values, r _s = 0.05, p = 0.174, or OD₂₆₀/OD₂₈₀ values, r ₍₇₀₈₎ = −0.04, p = 0.352, but there was a small negative correlation between collection site temperature and sample T/S values, r ₍₄₀₇₎ = −0.12, p = 0.012, 95% CI (−0.22, −0.03), (Figure 5A and Figure S5A). When these analyses were repeated with the maximum temperature in Davis, there was no significant correlation between the maximum temperature in Davis and OD₂₆₀/OD₂₃₀ values, r _s = 0.05, p = 0.177, but there was a small negative correlation between temperature and OD₂₆₀/OD₂₈₀ values, r ₍₈₀₁₎ = −0.09, p = 0.012, 95% CI (−0.16, −0.02), and a negative correlation between Davis temperature and sample T/S values, r ₍₄₃₈₎ = −0.19, p < 0.001, 95% CI (−0.28, −0.10), (Figure 5B and Figure S5B).

FIGURE 5.

Extract T/S values by regional temperatures for extracts that met DNA purity criteria. Plots show Pearson's correlations between sample T/S and (A) the maximum local temperature on the day of sample collection in the participants' zip code and (B) the maximum temperature in Davis, California, USA on the day the samples were delivered to the lab.

We then assessed relationships between cold pack condition and temperature, transportation duration, DNA purity criteria, and T/S values. As might be expected, we observed positive associations between the extent of cold pack melting and temperature in the sample collection location, r _s = 0.51, p < 0.001 (Figure 6A), the temperature in Davis on the day of delivery, r _s = 0.36, p < 0.001 (Figure 6B), and length of time the samples were in transit, r _s = 0.30, p < 0.001 (Figure 6C). There was no significant correlation between cold pack condition and OD₂₆₀/OD₂₃₀, r _s = 0.04, p = 0.315, or OD₂₆₀/OD₂₈₀ values, r _s = 0.01, p = 0.828. There was, however, a small negative correlation between cold pack condition and sample T/S values, r _s = −0.11, p = 0.027 (Figure 6D), suggesting that samples that were kept colder had slightly longer T/S estimates.

FIGURE 6.

Cold pack condition in relation to temperature, transportation duration, and T/S estimates for extracts that met DNA purity criteria. Boxplots depicting associations between cold pack condition and (A) the maximum temperature in the participants' zip code on the day their sample was collected, (B) the maximum temperature in Davis, California, USA (the location of the storing lab) on the day the sample was delivered, (C) the time the sample was in transit between collection and freezing, and (D) sample T/S values.

Because our samples showed no signs of DNA degradation detectable by standard gel electrophoresis, the relationships observed between temperature and T/S values may reflect seasonal variation in bulk TL—potentially due to changes in immune cell composition (Beaulieu et al. 2017). However, the fact that higher temperatures and warmer cold packs were both associated with longer T/S values suggests that there may be some temperature‐driven sample degradation. While we cannot distinguish these possibilities in the present study, our results do suggest that seasonal variation and temperature should be measured and considered when samples are collected at different times from different locations, and that attention should be paid to the cold chain during sample handling and transportation.

3.7. Comparing the Repeatability of the TL Assay Across Multiple Extractions and Time Points

We used linear mixed effect models to compare the repeatability of the TL assay for (1) two independent assay replications taken from the same sample at a given time point, and (2) two samples taken from the same individuals at separate time points 1 year apart. First, we eliminated the potential effects of time point and calculated the ICC for the average T/S estimates (across three runs of the assay) for all participants with two high quality extracts at baseline only (416 observations from 208 samples taken from 208 people). Consistent with the ICCs reported above, we found a high degree of reliability between the T/S values taken from the same sample/individual: ICC = 0.965, SE = 0.005, 95% CI (0.956, 0.973), indicating that 96.5% of the variance was accounted for by difference between individuals, with the remaining 3.5% of variance potentially due to other sources including measurement error or assay variability.

Next, we included the 1‐year follow‐up samples in the model and constrained the dataset to include only participants who had two high quality extracts at both time points (for a total of 508 observations). This model included a fixed effect for time point and random effects for participant identity (n = 127) and sample identity (n = 254). Likelihood ratio tests indicated that the addition of random effects for participant identity (p < 0.001) and sample identity (p < 0.001) both improved model fit. The repeatability (ICC) for participant identity was estimated to be 0.925, SE = 0.012, 95% CI (0.907, 0.946), indicating that 92.5% of the variance was attributable to between‐subject differences, and that measurements taken from the same individual were highly consistent. Additionally, after participant identity was accounted for, the repeatability (ICC) for sample identity was estimated to be 0.039, SE = 0.009, 95% CI (0.024, 0.059), indicating that the unique variance attributed to sample identity (which in this case, corresponds with the timepoint) was only 3.9%, suggesting that the amount of between‐extract variance was small after participant identity was accounted for. There was no effect of time point on mean TL, ß = −0.0007, SE = 0.0055, t ₍₁₂₆₎ = −0.13, p = 0.896. These findings suggest that the qPCR assay can provide highly reliable estimates across time. We also see no evidence of change in TL estimates across the 1‐year period in this naturalistic study design. While this is consistent with reports that suggest that it may take years for observable changes to occur in TL, we also note that there was no active intervention or disease condition present in our study.

3.8. Comparing Person‐Level, Extract‐Level, and Blood‐Sample‐Level Covariates: Age, Sex, and DNA Concentration/Blood Clotting are the Strongest Predictors of T/S Values

In the prior analyses, we characterized the separate influence of person‐level (age, sex, BMI), extract‐level (DNA concentration, clotting, OD₂₆₀/OD₂₈₀, and OD₂₆₀/₂₃₀), and sample‐level covariates (transportation time, temperature exposure, cold pack condition, and storage time) on our assay outcomes. Next, we used a series of mixed effects models to examine the relative influence of these sets of variables on T/S values when they were modeled in tandem. For each set of predictors, we specified a set of models for baseline and for the 1‐year time point. First, we specified a preliminary model that included only covariates at the extract‐level or sample‐level. Then, we specified a global model that included covariates from the other levels. Finally, we specified a model of “best fit” by iteratively comparing each combination of predictors from the global model to identify the most parsimonious model as ranked by AICc model fit. Continuous variables were z‐scored for all models.

Table 5 presents the first set of models, which examined the extract‐level variables alone and in conjunction with person‐level variables. In the preliminary baseline model, blood clotting and OD 260/280 values each significantly predicted T/S values (Model 1), whereas DNA concentration, blood clotting, and OD 260/280 values were each significant predictors of T/S values in the 1‐year model (Model 4). We then added the person‐level covariates to extract‐level covariates in the global models (Models 2 and 5) and used the dredge function to determine the best fitting model at each time point. The best fitting baseline model included age, sex, blood clotting, and OD₂₆₀/OD₂₈₀ values (Model 3), while the best fitting 1‐year model included age, sex, and DNA concentration (Model 6).

TABLE 5.

Mixed effects models predicting telomere length with extract‐level and person‐level covariates.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
	Baseline extract‐level covariates only	Baseline extract‐level + person‐level covariates	Baseline extract‐level + person‐level best fit	1‐year extract‐level covariates only	1‐year extract‐level + person‐level covariates	1‐year extract‐level + person‐level best fit
Observations (n)	461	461	461	307	307	307
Participants (n)	258	258	258	163	163	163
Fixed effects
Intercept	0.824 (0.011)***	0.842 (0.010)***	0.839 (0.010)***	0.806 (0.014)***	0.833 (0.013)***	0.837 (0.013)***
Timepoint (Ref. = baseline)	—	—	—	—	—	—
Age	—	−0.099 (0.009)***	−0.100 (0.009)***	—	−0.090 (0.012)***	−0.094 (0.012)***
Sex (Ref. = male)	—	−0.082 (0.021)***	−0.084 (0.022)***	—	−0.087 (0.030)**	−0.092 (0.028)**
BMI	—	−0.015 (0.009)	—	—	−0.019 (0.012)	—
DNA concentration	0.009 (0.005)	0.008 (0.004)	—	0.017 (0.007)*	0.017 (0.007)*	0.032 (0.005)***
Blood clotting (Ref. = no clot)	0.020 (0.007)**	0.021 (0.007)	0.029 (0.006)***	0.021 (0.009)*	0.021 (0.008)**	—
OD 260/280	0.014 (0.004)***	0.014 (0.004)	0.016 (0.004)***	0.011 (0.006)*	0.011 (0.006)*	—
OD 260/230	−0.024 (0.013)	−0.027 (0.012)	—	−0.031 (0.028)	−0.030 (0.028)	—
Variance components
Participant intercept	0.032 (0.178)	0.020 (0.141)	0.020 (0.142)	0.031 (0.177)	0.021 (0.144)	0.023 (0.151)
Residual variance	0.001 (0.030)	0.001 (0.030)	0.001 (0.030)	0.001 (0.033)	0.001 (0.033)	0.001 (0.034)
Fit statistics
Conditional R 2	0.973	0.973	0.972	0.967	0.97	0.967
Marginal R 2	0.012	0.371	0.360	0.019	0.348	0.337
AICc	−810.43	−903.71	−921.41	−528.26	−569.69	−591.22
BIC	−781.74	−862.87	−892.72	−502.54	−533.16	−591.21

Note: For all models, fixed effects and variance components are restricted maximum likelihood estimates, with standard errors (for fixed effects) and standard deviations (for variance components) in parentheses. Ref. indicates the reference category for categorical predictors. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

Table 6 presents a second set of models, which assessed sample‐handling variables alone and in the presence of the person‐level and extract‐level variables. In the preliminary baseline model, maximum Davis temperature was the only significant predictor of T/S values (Model 7), whereas in the 1‐year model, transportation duration and cold pack condition were significant predictors of T/S values (Model 10). When we added the person‐level and extract‐level predictors to the global models (Models 8 and 11) and applied the dredge function, we found that the best fitting baseline model included only age and DNA concentration (Model 9), whereas the best fitting 1‐year model included age, sex, and DNA concentration (Model 12). Standardized betas for the fixed effects examined in the best fitting Models 3, 6, 9, and 12 are presented in Table 7 for comparison across predictors.

TABLE 6.

Mixed effects models predicting telomere length with sample handling variables plus extract‐level and person‐level covariates.

	Model 7	Model 8	Model 9	Model 10	Model 11	Model 12
	Baseline sample‐level covariates only	Baseline person + extract + sample‐level	Baseline person + extract + sample‐level best fit	1‐year sample‐level covariates only	1‐year person + extract + sample‐level	1‐year person + extract + sample‐level best fit
Observations (n)	391	391	391	269	269	269
Participants (n)	231	231	231	151	151	151
Fixed effects
Intercept	0.803 (0.024)***	0.815 (0.022)***	0.826 (0.010)***	0.932 (0.033)***	0.909 (0.029)***	0.842 (0.014)***
Age	—	−0.094 (0.010)***	−0.100 (0.010)***	—	−0.083 (0.014)***	−0.095 (0.013)***
Sex (Ref. = male)	—	−0.061 (0.024)*	—	—	−0.087 (0.030)**	−0.094 (0.030)**
BMI	—	−0.015 (0.009)	—	—	−0.019 (0.013)	—
DNA concentration	—	0.009 (0.005)	0.024 (0.004)***	—	0.020 (0.007)**	0.035 (0.006)***
Blood clotting (Ref. = no clot)	—	0.023 (0.007)**	—	—	0.025 (0.009)**	—
OD 260/280	—	0.012 (0.005)**	—	—	0.009 (0.006)	—
OD 260/230	—	−0.026 (0.013)*	—	—	−0.023 (0.029)	—
Max collection temperature	0.002 (0.016)	0.004 (0.014)	—	0.033 (0.025)	0.028 (0.021)	—
Max Davis temperature	−0.066 (0.021)**	−0.027 (0.018)	—	−0.006 (0.020)	0.001 (0.017)	—
Transportation duration	−0.022 (0.012)	−0.017 (0.010)	—	0.054 (0.021)*	0.025 (0.019)	—
Cold pack condition (Ref. = frozen)
Partially frozen	0.031 (0.032)	0.033 (0.027)	—	−0.132 (0.038)***	−0.091 (0.033)**	—
Melted cold	0.056 (0.042)	0.039 (0.035)	—	−0.200 (0.056)***	−0.118 (0.049)*	—
Melted warm	0.090 (0.065)	0.071 (0.054)	—	—	—	—
Variance components
Participant intercept	0.029 (0.171)	0.020 (0.143)	0.021 (0.144)	0.030 (0.172)	0.022 (0.147)	0.023 (0.151)
Residual variance	0.001 (0.036)	0.001 (0.031)	0.001 (0.033)	0.002 (0.040)	0.001 (0.033)	0.001 (0.034)
Fit statistics
Conditional R 2	0.960	0.971	0.967	0.955	0.970	0.968
Marginal R 2	0.070	0.359	0.323	0.111	0.377	0.345
AICc	−596.74	−664.52	−729.61	−399.15	−428.82	−483.60
BIC	−561.49	−602.47	−709.92	−370.95	−376.80	−462.36

Note: For all models, fixed effects and variance components are restricted maximum likelihood estimates, with standard errors (for fixed effects) and standard deviations (for variance components) in parentheses. Ref. indicates the reference category for categorical predictors. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

TABLE 7.

Standardized estimates for best‐fit mixed effects models.

Fixed effects	Model 3	Model 6	Model 9	Model 12
	Baseline person + extract‐level best fit	1‐year person + extract‐level best fit	Baseline person + extract + sample‐level best fit	1‐year person + extract + sample‐level best fit
Intercept	0.07 (−0.04, 0.18)	0.12 (−0.02, 0.27)	−9.10e‐04 (−0.11, 0.11)	0.13 (−0.02, 0.28)
Age	−0.54 (−0.63, −0.44)	−0.51 (−0.64, −0.38)	−0.55 (−0.66, −0.45)	−0.50 (−0.64, −0.36)
Sex (Ref. = male)	−0.46 (−0.70, −0.23)	−0.52 (−0.83, −0.21)	—	−0.51 (−0.83, −0.19)
BMI	—	—	—	—
DNA concentration	—	0.19 (0.13, 0.26)	0.13 (0.09, 0.17)	0.20 (0.14, 0.26)
Blood clotting (Ref. = no clot)	0.16 (0.10, 0.22)	—	—	—
OD 260/280	0.06 (0.03, 0.09)	—	—	—

Note: Standardized betas and 95% confidence (in parentheses) intervals are provided for the fixed effects included in the best‐fit models presented in Tables 5 and 6.

This series of analyses confirmed that age and sex are reliable predictors of TL, while the predictive power of the extract‐level and sample‐level factors was less consistent across time points, depending on the variables modeled. When all factors were considered, our final models (9 and 12) suggested that in addition to age and sex, extract DNA concentration was the most significant predictor, while the remaining covariates offered little predictive power over and above these three variables. However, while our protocol used a standardized quantity of 20 ng/μL of DNA per PCR reaction for most reactions, we also retained samples for which there was less than 20 ng/μL of DNA of starting material available. We therefore conducted another set of analyses to assess the consequences of this variability in starting DNA concentration on these models.

When we eliminated extracts with less than 20 ng/μL of DNA of starting material and re‐estimated Models 1 through 12, we found that age, sex, and blood clotting were the most consistent predictors in the best fitting models (see Tables S5–S7). Because DNA concentration is associated with clotting, 260/280 values, and 260/230 values, we suspect that variance in the starting concentration of DNA can explain some of the inconsistencies observed in the first set of models. Although there is no established empirical cutoff for the amount of starting material needed for the qPCR reaction, these findings do suggest that the starting DNA concentration should be consistent when possible. They also suggest that blood clotting is the most consequential sample‐level factor analyzed in this dataset.

3.9. Assessing the Impact of Assay Replication on Precision, Accuracy, and Cost

In our final set of analyses, we investigated whether there were any systematic effects of assay plate and whether running a third assay plate (i.e., replication on a separate plate) improved the precision or external validity of our telomere estimates. The standard practice for the qPCR TL assay in our group is to run each DNA sample (in triplicate wells) on two separate pairs of T and S assay plates initially, and to run a third pair of assay plates for any samples having T/S values from the first two plates that vary by more than 7%. The two closest values are then averaged to calculate the T/S value for a given DNA extract. In this study, we opted to run each DNA extract on three separate assay plates from the start and repeated this whole procedure twice for blood samples with enough material for two DNA extractions.

First, we assessed whether there were differences between assay plates on measures of reliability or precision. The CV of every sample (0.030, +/− 0.018) was measured for three runs and averaged across 10 plates (Figure 7A). The ICC was calculated for an average of 40 duplicate extractions per full plate (43 per plate including samples that did not pass DNA purity criteria), and the ICCs ranged between 0.939 and 0.988 across all 10 plates (Figure 7B). Complete CV and ICC data are tabulated in Figure 7C (see also Figure S6 for CVs and ICCs of all duplicate pairs including samples that did not pass DNA purity criteria). These patterns suggest that there was no apparent plate effect on the CV of three runs or the ICC from two separate extractions.

FIGURE 7.

Assay reliability and precision by plate for extracts that met DNA purity criteria. (A) The CV and (B) ICC across 10 plates. Error bars represent standard deviations. (C) Tabular data for CV and ICC. The CV represents three runs across 10 plates. The ICC is computed for T/S values from duplicate extraction pairs averaged across each of the 10 plates, with the number of pairs on each plate and 95% CI indicated. Data presented in this figure come from 365 DNA pairs (i.e., all pairs with OD₂₆₀/OD₂₃₀ ≥ 1.0 and OD₂₆₀/OD₂₈₀ ≥ 1.7 and ≤ 2.0).

Next, we assessed whether completing three full runs of the TL assay for each sample (as opposed to two) would meaningfully improve the between‐extract precision using the full set of 365 pure extract pairs. First, we computed the ICC for T/S values from duplicate extractions that were each calculated as the average of all three runs (ICC = 0.964) and compared this to the ICC for extract T/S ratios calculated from the average of the two closest runs (ICC = 0.952; See Table 8 and Table S8). Next, we compared these values to our standard procedure: taking the average of the first two runs for samples that varied by less than 7%, and for samples where these first two runs varied by more than 7% (about 15% of DNA samples in this study), we considered the third run and averaged the values for the two closest of these three runs (ICC = 0.956). Lastly, we compared these values to an adjustment in our standard procedure: taking the average of the first two runs for samples that varied by less than 7% and then taking the average of all three runs in cases where the first two runs varied by more than 7% (ICC = 0.961). In all four of these analysis approaches, the confidence intervals indicate that the between‐extract ICCs were statistically indistinguishable, suggesting that the addition of a third assay run does not substantively improve the assay precision (Table 8 and Table S8), although it increases costs by up to 25%.

TABLE 8.

CV and ICC analysis of 365 duplicate extractions considering varying numbers of assay runs.

# assay runs considered	# pairs	CV	SD of CV	ICC	SE of ICC	95% CI
3 runs	365	0.031	0.019	0.964	0.004	0.956, 0.971
2 best	365	0.010	0.009	0.952	0.005	0.942, 0.961
2 runs+, average best 2	365	0.020	0.014	0.956	0.005	0.946, 0.964
2 runs+, average all 3	365	0.023	0.018	0.961	0.004	0.952, 0.968

Note: All calculations conducted on the 365 extract pairs with OD₂₆₀/OD₂₃₀ ≥ 1.0 and OD₂₆₀/OD₂₈₀ ≥ 1.7 and ≤ 2.0. “3 runs” represents the ICC for duplicate extract T/S values calculated as the average of all three runs. “2 best” represents the ICC for duplicate extract T/S values calculated as the average of the two closest out of the three runs. “2 runs +, average best 2,” represents the ICC for duplicate extract T/S values calculated as the average of the first two runs, unless those two T/S run values varied by > 7%, in which case the third run was considered and the two closest of the three runs were averaged. “2 runs +, average all 3” represents the ICC for duplicate extract T/S values calculated as the average of the first two runs, unless those two T/S run values varied by > 7%, in which case the average of all three runs for those samples was used.

We also assessed whether the associations between our telomere estimates and external validity metrics differed if T/S values were calculated from two or three assay runs per extract, as well as from one or two extractions per sample (representing the full replication of the entire assay procedure including the extraction step). Overall, we found that including two or three assay plate replicates made little to no difference on the stability of the relationships between the T/S estimates and measures of external validity (Table 9). Using T/S values from only one extract, rather than two, tended to introduce greater variability in those associations; though none of the correlations statistically differed from one another across the four T/S estimate calculations used. Collectively, these findings suggest that for a sample of this size with assays run by highly experienced technicians, the improvements in accuracy were negligible when averaging three replications of the TL assay step rather than two, and only slightly more pronounced when comparing the results from one extraction to the average of two extractions, which involves a full replication of the entire assay procedure from start to finish.

TABLE 9.

External validity correlations with T/S values calculated from various numbers of DNA extracts and assay replicates.

	df	r	p	95% CI
Sample T/S at baseline and 1‐year
1) Sample T/S averaged across 3 runs across extracts	173	0.95	< 0.001	[0.93, 0.96]
2) Sample T/S averaged across 2 runs across extracts	173	0.94	< 0.001	[0.93, 0.96]
3) Sample T/S from extract 1 (averaged across all three runs)	125	0.93	< 0.001	[0.90, 0.95]
4) Sample T/S from extract 2 (averaged across all three runs)	125	0.92	< 0.001	[0.89, 0.94]
Age and sample T/S at Baseline
1) Sample T/S averaged across 3 runs across extracts	260	−0.58	< 0.001	[−0.65, −0.49]
2) Sample T/S averaged across 2 runs across extracts	260	−0.58	< 0.001	[−0.65, −0.49]
3) Sample T/S from extract 1 (averaged across all three runs)	206	−0.60	< 0.001	[−0.68, −0.50]
4) Sample T/S from extract 2 (averaged across all three runs)	206	−0.58	< 0.001	[−0.66, −0.48]
Age and sample T/S at 1 year
1) Sample T/S averaged across 3 runs across extracts	176	−0.53	< 0.001	[−0.63, −0.42]
2) Sample T/S averaged across 2 runs across extracts	176	−0.53	< 0.001	[−0.63, −0.42]
3) Sample T/S from extract 1 (averaged across all three runs)	155	−0.58	< 0.001	[−0.68, −0.47]
4) Sample T/S from extract 2 (averaged across all three runs)	155	−0.56	< 0.001	[−0.66, −0.44]
BMI and sample T/S at Baseline
1) Sample T/S averaged across 3 runs across extracts		−0.21	0.001
2) Sample T/S averaged across 2 runs across extracts		−0.21	0.001
3) Sample T/S from extract 1 (averaged across all three runs)		−0.20	0.005
4) Sample T/S from extract 2 (averaged across all three runs)		−0.15	0.032
BMI and sample T/S at 1 Year
1) Sample T/S averaged across 3 runs across extracts		−0.23	0.003
2) Sample T/S averaged across 2 runs across extracts		−0.23	0.003
3) Sample T/S from extract 1 (averaged across all three runs)		−0.19	0.025
4) Sample T/S from extract 2 (averaged across all three runs)		−0.24	0.003

3.10. Comparisons With Other Source Materials

Lastly, we consider the costs and benefits of this microtainer sample‐collection approach compared to DBS and saliva sample collection. Our lab has experience with the microtainer method, saliva collected with the DNA Genotek Oragene kits, and DBS collected with the Whatman⁹⁰³ protein saver cards. The costs for DNA extraction and the TL assay are comparable for these sample sources, so the primary differences lie in the costs of the collection materials and shipping. For the microtainer method used in this study, we estimated the average overall cost to be ~$60/sample for all materials related to blood collection, sample preparation, and shipping (i.e., alcohol wipes, band aids, gauze, labels, etc.). While we do not have comparable estimates for collecting individual samples from each participant for each of the other methods, we have outlined the basic materials and shipping considerations in Table 10—noting that costs associated with the microtainer method can be somewhat cheaper for the basic collection materials but are likely more expensive to ship and store given the temperature requirements.

TABLE 10.

Cost comparisons for various sample sources and collection methods.

Sample source/collection method	Example collection materials	Collection materials cost/unit	Shipping materials	Shipping costs	Storage requirements
Whole blood in microtainer	BD vacutainer collection tube $430.00/case of 200 BD contact‐activated lancet blade: 1.5 mm × 2.0 mm Blue High Flow $74.40/Box of 200	$2.52 × 3 = $7.56	Insulated shipping container + 3 cold packs $6.80/shipment	~$18 average overnight FedEx cost/shipment	−80°C
Dried blood spots	Cytiva Whatman 903 Protein Saver Card $434.50/Pack of 100	$4.35	Shipping with cold packs advised	Assuming fewer cold packs and a smaller container shipping may be cheaper	Room temperature
Saliva	Oragene OGR‐500, DNA Genotek Saliva kits $19.50/kit	$19.50	Can be shipped at room temp, but cold packs recommended		Room temperature

In terms of assay precision, our internal records demonstrate comparable precision between the ICCs observed with microtainers in this study and the ICCs observed from two saliva studies (Study 1: ICC = 0.95, 95% CI [0.911–0.972] with 45 pairs of duplicate extractions; Study 2: ICC = 0.979, 95% CI [0.94–0.992] for 17 pairs of duplicate extractions). DBS, on the other hand, tends to be less precise (Study 1: ICC = 0.804, 95% CI [CI 0.624–0.9] with 28 pairs of duplicate extractions; Study 2: ICC = 0.734, 95% CI [0.572–0.845] for 46 pairs of duplicate extractions).

Given these considerations, we believe the microtainer collection method is the superior alternative to DBS when researchers want to collect whole blood remotely for qPCR telomere assays, though blood clotting and maintaining the cold chain during sample transportation are important considerations. Nevertheless, venipuncture is still necessary for alternative assay methods that require more starting material or when cell sorting is needed, and saliva methods may be preferable when working with children or vulnerable populations.

3.11. Limitations

The primary limitation of this assay method is that it results in bulk TL estimates that average across telomeres within a given cell, and across cells within a given sample. Although the majority of research examining TL in relation to human health has utilized bulk measures, these metrics are limited for two reasons. First, nucleated human cells contain 92 telomeres (i.e., one telomere capping each end of 46 linear chromosomes) that vary in length, and some evidence suggests that it is the shortest telomere that triggers DNA damage responses and cellular senescence (Lai et al. 2018). As such, the shortest telomere may be a more biologically relevant measure than average TL. Second, different cell types are known to have varying TLs (Lin et al. 2010), so the distribution of cell types in a given sample may mask true changes in TL when bulk estimates are used. This limitation pertains similarly to saliva and DBS, so more granular measures involving cell sorting are necessary for detailed mechanistic studies. Nevertheless, qPCR remains a useful and cost‐effective method for measuring TL in large cross‐sectional studies when limited sample material can be collected.

Another potential limitation is that the extract replicates for the baseline and 1‐year samples were positioned next to one another across assay plates. This was done intentionally to minimize the impact of well‐position effects but may have also inflated the sample repeatability.

Finally, it is also important to consider the demand characteristics and overall intensity and timing of the study when interpreting the sample collection rates, as these factors likely affected participants' motivation and compliance with the blood collection procedure. We recruited a cohort of participants with prior meditation experience to participate in a study examining the effectiveness of meditation in reducing pandemic‐related stress. As such, these participants may have been motivated to provide a sample because the study focus aligned with their pre‐existing goals or beliefs. The study also involved responding to extensive batteries of questionnaires at four time points in addition to providing two blood samples across a 1‐year period. This level of participant burden may have demotivated some participants from completing the sample collections; however, it may also have increased participants' overall investment in the study and further incentivized them to complete the blood draws. This study was also conducted during the height of the COVID‐19 pandemic, which similarly may have resulted in more availability and willingness for some participants, while making it more difficult for others to participate. As such, the commitment and compliance of participants observed in this study may not generalize to other circumstances.

4. Conclusions

This is the first study to use whole blood collected remotely in EDTA microtainers for qPCR TL measurements. In combination with magnetic bead extraction, we demonstrated that this collection method results in high‐quality DNA and highly precise telomere measurements from as little as 50 μL. We also confirmed that there were no meaningful effects of sample volume, transportation duration, or −80°C storage time on DNA quality or assay precision. We did, however, find that blood clotting seemed to increase TL estimates, suggesting that there is room for improving this sample collection technique to further minimize blood clotting. We also found negative relationships between TL estimates and the maximum local temperature during sample collection and delivery, suggesting that there may be seasonal effects influencing TL or heat exposure affecting sample integrity during transportation. Based on these findings, we recommend that future studies using this method either account for or constrain seasonal variation in collection times, and measure sample temperature in transit if possible.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: Supporting information.

Conklin, Q. A. , Smith D. L., Dai G., King B. G., Saron C. D., and Lin J.. 2025. “Remote Blood Collection for Telomere Length Measurement: Assessing the Impact of Sample Characteristics and Handling on DNA Quality and Assay Outcomes.” American Journal of Human Biology 37, no. 9: e70128. 10.1002/ajhb.70128.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.