Telomere length and epigenetic clocks as markers of cellular aging: a comparative study

Abstract

Telomere length (TL) and DNA methylation–based epigenetic clocks are markers of biological age, but the relationship between the two is not fully understood. Here, we used multivariable regression models to evaluate the relationships between leukocyte TL (LTL; measured by qPCR [n = 635] or flow FISH [n = 144]) and five epigenetic clocks (Hannum, DNAmAge pan-tissue, PhenoAge, SkinBlood, or GrimAge clocks), or their epigenetic age acceleration measures in healthy adults (age 19–61 years). LTL showed statistically significant negative correlations with all clocks (qPCR: r =  − 0.26 to − 0.32; flow FISH: r =  − 0.34 to − 0.49; p < 0.001 for all). Yet, models adjusted for age, sex, and race revealed significant associations between three of five clocks (PhenoAge, GrimAge, and Hannum clocks) and LTL by flow FISH (p < 0.01 for all) or qPCR (p < 0.001 for all). Significant associations between age acceleration measures for the same three clocks and qPCR or flow FISH TL were also found (p < 0.01 for all). Additionally, LTL (by qPCR or flow FISH) showed significant associations with extrinsic epigenetic age acceleration (EEAA: p < 0.0001 for both), but not intrinsic epigenetic age acceleration (IEAA; p > 0.05 for both). In conclusion, the relationships between LTL and epigenetic clocks were limited to clocks reflecting phenotypic age. The observed association between LTL and EEAA reflects the ability of both measures to detect immunosenescence. The observed modest correlations between LTL and epigenetic clocks highlight a possible benefit from incorporating both measures in understanding disease etiology and prognosis.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11357-022-00586-4.

Introduction

The precise conceptualization of aging as it relates to health is still being refined, with the ultimate goal of defining “young” or “old” by biomarkers reflecting cellular age. DNA methylation (DNAm)–based clocks (also known as epigenetic clocks) and telomere length (TL) have been identified as useful biomarkers in this context.

Epigenetic clocks were developed using machine-learning algorithms and use DNAm patterns at CpG sites across the genome to predict chronologic age or its correlates (e.g., lifespan, age-related phenotypes, or plasma protein levels). Epigenetic clocks include the Hannum DNAmAge clock [1], DNAmAge or “Pan Tissue” clock [2], SkinBlood clock [3], PhenoAge clock [4], and GrimAge clock [5]. The Hannum clock is derived from whole blood, while the “Pan Tissue” or “SkinBlood” clocks were designed to predict chronologic age across different tissue types. More recent epigenetic clocks, namely, PhenoAge or GrimAge, incorporate age-associated clinical and laboratory measures in their development and focus on predicting phenotypic age and lifespan, respectively.

Telomere length is also considered a marker of biological age. Telomeres are long tandem nucleotide repeats (TTAGGG)_n and a protein structure located at chromosome ends that are essential for chromosomal stability and shorten with each cell division [6]. Cellular replicative senescence or apoptosis is triggered when telomeres reach a critically short length [7–10]. Several methods are used to measure TL; most commonly used are the gold standard Southern blot Telomere Restriction Fragment (TRF), the clinically used flow cytometry with fluorescence in situ hybridization (flow FISH), and the high-throughput qPCR assay. Details on those methods and the pros and cons for each were reviewed elsewhere [11, 12]. Both epigenetic age and TL have been associated with a variety of age-related diseases and health outcomes including cardiovascular disease, metabolic syndrome, cancer, and all-cause mortality (reviewed in [7, 13, 14]).

Previous studies suggest minimal or no relationships between epigenetic age and TL but were limited in that most only evaluated qPCR TL and the DNAmAge clock [15–19]. Here, we explored the relationship between all five published epigenetic clocks and TL measured by qPCR assay and flow FISH.

Methods

Study participants and TL measurement methods used here have been described elsewhere [20] (Table 1). Briefly, the study participants were healthy adult donors from the Transplant Outcomes in Aplastic Anemia (TOAA) study, a research collaboration between the National Cancer Institute (NCI) and the Center for International Blood and Marrow Transplant Research (CIBMTR; https://www.cibmtr.org). DNA for qPCR TL or methylationEpic array was extracted from peripheral blood mononuclear cells (PBMCs; n = 219) or frozen whole blood (n = 416) using the QIAamp Maxi Kit procedure (Qiagen Inc., Valencia, CA, USA). Flow FISH telomere length was measured in a subset with cryopreserved PBMC samples (n = 144), and values from total lymphocyte TL are used in this study.

Table 1.

Characteristics of study participants

Variable	Flow FISH	qPCR
	N (%)	N (%)
Total	144	635
Sex
Male	90 (63%)	425 (67%)
Female	54 (38%)	210 (33%)
Race
White	110 (77%)	478 (75%)
African American	6 (4%)	40 (6%)
Other	26 (18%)	93 (15%)
Missing	2 (0.5%)	24 (4%)
	Median (range)	Median (range)
Age	36.9 (20–53.2)	33.8 (19.1–61.1)
,
Flow FISH (kb)1	7.0 (3.7–11.2)	7.0 (3.7–11.2)
qPCR (z-score)2	− 0.03 (− 2.3 to 4.3)	− 0.17 (− 2.3 to 4.9)
Epigenetic age clocks2,3
GrimAge clock	33.9 (19.8–50.2)	35.9 (19.8–63.6)
PhenoAge clock	16.2 (− 10.7 to 42.8)	21.1 (− 10.7 to 53.9)
Hannum clock	23.1 (1.4–46)	24.6 (− 1.2 to 49.6)
SkinBlood clock	32.9 (11.9–54.8)	30.9 (11.6–60.7)
DNAmAge clock	31.6 (13.2–50.1)	31.6 (6.2–58.6)
Age acceleration residual (AAR) measures
AAR GrimAge	− 0.5 (− 10.0 to 15.5)	− 4.2 (− 10.0 to 7.4)
AAR PhenoAge	− 0.3 (− 31.5 to 30.5)	− 7.9 (− 31.5 to 10.7)
AAR Hannum	0.02 (− 23.1 to 21.7)	− 3.7 (− 18.1 to 10.35)
AAR SkinBlood	0.3 (− 10.3 to 20.7)	0.09 (− 8.6 to 9.1)
AAR DNAmAge	− 1.03 (− 9.1 to 10.7)	0.5 (− 21.6 to 19.6)
IEAA4	0.2 (− 7.9 to 12.9)	0.2 (− 19.8 to 20.5)
EEAA4	− 8.3 (− 25.7 to 12.4)	1.3 (− 25.7 to 28.2)

Flow FISH fluorescence in situ hybridization (FISH) with flow cytometry, qPCR quantitative polymerase chain reaction, DNAm DNA methylation, AAR age acceleration residual, IEAA intrinsic epigenetic age acceleration, EEAA extrinsic epigenetic age acceleration

¹Cryopreserved PBMC samples were used, and reported values are from total lymphocytes

²DNA was extracted from peripheral blood mononuclear cells (PBMCs; n = 219) or frozen whole blood (n = 416)

³Epigenetic clocks were developed using penalized regression to select CpG sites that produce the most accurate estimate of chronological age only (for DNAmAge, Hannum, and SkinBlood clocks), or to predict mortality risk from a combination of phenotypic indicators, including chronological age and blood chemistries (PhenoAge), or selected blood proteins and smoking history, controlling for chronological age (GrimAge)

⁴Calculated from DNAmAge clock

DNA methylation measurement and epigenetic age calculation

We used the Illumina Infinium MethylationEPIC Bead™ array for whole-genome DNAm profiling, as described elsewhere [20]. We calculated epigenetic age using five clocks: Hannum (71 CpGs), DNAmAge (353 CpGs), SkinBlood (513 CpGs), PhenoAge (391 CpGs), or GrimAge (1030 CpGs), as previously described [1–5, 21, 22]. Briefly, epigenetic clocks were originally developed using penalized regression to select CpG sites that produce the most accurate estimate of chronological age only (for DNAmAge, Hannum, and SkinBlood clocks), or to predict mortality risk from a combination of phenotypic indicators, including chronological age and blood chemistries (PhenoAge), or selected blood proteins and smoking history, controlling for chronological age (GrimAge). Epigenetic age is then calculated as the weighted average, based on the regression coefficients, of DNAm levels for the clock-specific CpG sites and is presented in units of years [4, 23]. The majority of CpG sites are unique to each clock, with a small number of shared CpG sites between clocks [21, 24, 25]. We also calculated epigenetic age acceleration residual (AAR) measures [26] for all clocks, which represent the deviation of observed methylation age from expected for the individual’s chronological age. This includes AAR for Hannum, SkinBlood, PhenoAge, and GrimAge clocks. From the DNAmAge clock, we calculated the following: (a) [AAR DNAmAge]: residuals of the regression of DNAmAge on chronological age; (b) intrinsic epigenetic age acceleration [IEAA]: residuals of the multivariable regression of DNAmAge on chronological age and predicted methylation-based proportions of white blood cell (WBC) types. The statistical adjustment for white blood cell (WBC) composition allows IEAA to capture biological aging across tissue types, independent of immune cell aging; and (c) extrinsic epigenetic age acceleration [EEAA]: a measure of immune system age acceleration defined as the weighted average of DNAmAge and predicted proportions of the immune cells in the DNA content [27]. In all cases, a positive value of age acceleration suggests an older biological age than chronological age.

Statistical analysis

We used Pearson’s correlation coefficient and multivariable linear regression models to evaluate the strength of association between TL (flow FISH or qPCR) and epigenetic age or age acceleration measures. All tests were two-sided with statistical significance defined as p < 0.05. Multivariable models were adjusted for participant characteristics including chronological age, sex, and race. For comparison with epigenetic age acceleration measures, we used age-adjusted TL (calculated as the residuals from linear regressions of TL on age) and adjusted the models for sex and race. All measures of TL, epigenetic age clocks, and age acceleration residuals were standardized using z-scores for all analyses to allow for comparability of estimates across studies.

All data analysis was performed using SAS® statistical software, version 9.4, and visualizations were created using RStudio, version 1.3.959.

Results

Participant characteristics

The study included 635 healthy adults (median age = 34 years, range = 19–61). Of them, 425 (67%) were male, and 478 (75%) were white.

Relationship between chronologic age, and epigenetic age or telomere length

As expected, all epigenetic clocks were highly and positively correlated with chronological age (SkinBlood r = 0.96, DNAmAge r = 0.88, Hannum r = 0.85, PhenoAge r = 0.72, GrimAge r = 0.85, p < 0.001 for all). The negative correlation between chronological age and TL was modest, but statistically significant (r =  − 0.26, p < 0.001 for qPCR TL, and r =  − 0.34, p < 0.001 for flow FISH TL). Scatterplots of chronological age with all measures of TL and age clocks by DNA source are presented in Supplementary Fig. 1.

Relationship between epigenetic age and telomere length

Calculated epigenetic age by the different clocks was highly correlated with one another (r = 0.74 to 0.91, p < 0.0001 for all). The highest correlation was noted between DNAmAge and SkinBlood (r = 0.91) and the lowest between PhenoAge and SkinBlood (r = 0.74) (Fig. 1).

Fig. 1.

Correlations between leukocyte telomere length (TL) and epigenetic clocks or epigenetic age acceleration residual (AAR) measures. a Flow FISH total lymphocyte TL and epigenetic age clocks, b qPCR TL and epigenetic age clocks, c age-adjusted flow FISH lymphocyte TL and epigenetic age acceleration residual measures, and d age-adjusted qPCR TL and epigenetic age acceleration residual measures. Flow FISH, fluorescence in situ hybridization (FISH) with flow cytometry; qPCR, quantitative polymerase chain reaction; TL, telomere length; AAR, age acceleration residual; IEAA, intrinsic epigenetic age acceleration; EEAA, extrinsic epigenetic age acceleration. Note: DNA for qPCR telomere length or methylationEpic array was extracted from peripheral blood mononuclear cells (PBMCs; n = 219) or frozen whole blood (n = 416). Flow FISH total lymphocyte telomere length was measured in a subset with cryopreserved PBMC samples (n = 144)

TL showed statistically significant, moderate negative correlations with epigenetic age from all clocks for qPCR TL (r =  − 0.26 for SkinBlood, − 0.27 for DNAmAge, − 0.30 for Hannum, and − 0.32 for PhenoAge and GrimAge, p < 0.0001 for all) and flow FISH TL (r =  − 0.34 for SkinBlood, − 0.36 for DNAmAge, − 0.41 for GrimAge, − 0.43 for Hannum, and − 0.49 for PhenoAge, p < 0.0001 for all) (Fig. 1 a and b; Supplementary Fig. 2). Multivariable regression analyses adjusted for age, sex, and race revealed significant associations between epigenetic age from three of five clocks (PhenoAge, Hannum, and GrimAge) and both flow FISH (p < 0.01 for all) and qPCR (p < 0.001 for all) (Table 2). Relationships between TL (by qPCR or flow FISH) and DNAmAge or SkinBlood age were not statistically significant (p > 0.05 for all).

Table 2.

Multivariable regression for the association between epigenetic clocks or age acceleration measures and leukocyte telomere length

	Flow FISH total lymphocyte TL (N = 144)		qPCR TL (N = 635)
	β1	p-value	β1	p-value
Epigenetic age clocks2
GrimAge	− 0.61	0.003	− 0.35	< 0.0001
PhenoAge	− 0.57	< 0.0001	− 0.28	< 0.0001
Hannum	− 0.49	0.0006	− 0.26	0.0004
SkinBlood	− 0.26	0.36	− 0.11	0.42
DNAmAge	− 0.35	0.06	− 0.14	0.09
Epigenetic age acceleration measures3
AAR GrimAge	− 0.32	0.003	− 0.18	< 0.0001
AAR PhenoAge	− 0.39	< 0.0001	− 0.19	< 0.0001
AAR Hannum	− 0.24	0.0009	− 0.14	0.0004
AAR SkinBlood	− 0.07	0.37	− 0.03	0.42
AAR DNAmAge	− 0.17	0.06	− 0.07	0.09
IEAA DNAmAge	− 0.11	0.18	− 0.03	0.52
EEAA DNAmAge	− 0.32	< 0.0001	− 0.18	< 0.0001

DNA for qPCR telomere length or methylationEpic array was extracted from peripheral blood mononuclear cells (PBMCs; n = 219) or frozen whole blood (n = 416). Flow FISH telomere length was measured in a subset with cryopreserved PBMC samples (n = 144)

Flow FISH fluorescence in situ hybridization (FISH) with flow cytometry, qPCR quantitative polymerase chain reaction, TL telomere length, AAR age acceleration residual, IEAA intrinsic epigenetic age acceleration, EEAA extrinsic epigenetic age acceleration

¹All measures are in z-scores; β is interpreted as standard deviation (SD) change in epigenetic age clock or age acceleration measure under-study with each SD of TL; models were adjusted for chronological age, sex, and race

²Epigenetic clocks were developed using penalized regression to select CpG sites that produce the most accurate estimate of chronological age only (for DNAmAge, Hannum, and SkinBlood clocks), or to predict mortality risk from a combination of phenotypic indicators, including chronological age and blood chemistries (PhenoAge), or selected blood proteins and smoking history, controlling for chronological age (GrimAge)

³All age acceleration models used age-adjusted TL and were adjusted for sex and race

Relationship between epigenetic age acceleration and TL

Significant negative correlations between age acceleration residual (AAR) measures and age-adjusted TL by both flow FISH and qPCR were found for EEAA (r =  − 0.32, p < 0.0001 and r =  − 0.20, p < 0.001, for flow FISH and qPCR TL, respectively), and AAR GrimAge (r =  − 0.24 and r =  − 0.19, respectively, all p < 0.01), AAR PhenoAge (r =  − 0.37 and r =  − 0.20, respectively, all p < 0.01), and AAR Hannum (r =  − 0.27 and r =  − 0.15, respectively, all p < 0.01) (Fig. 1 c and d; Supplementary Fig. 2). These relationships remained statistically significant after adjusting for sex and race for both qPCR and flow FISH TL (p < 0.01 for all). Relationships between TL (by either qPCR or flow FISH) and IEAA or AAR for DNAmAge or SkinBlood clock were not statistically significant (Table 2).

Stratified analyses for qPCR TL correlations by DNA source (PBMC vs. whole blood) revealed no sizeable difference (r not exceeding ± 0.1) on any of the tested correlations, except between (a) EEAA and qPCR TL where a stronger correlation is observed in PBMC-derived samples (r =  − 0.24, p = 0.0003) compared with whole blood (r =  − 0.09, p = 0.06) and (b) AAR by Hannum and PhenoAge clocks, where a statistically significant negative association was observed in PBMC-derived samples (Hannum: r =  − 0.20, p = 0.004, PhenoAge r =  − 0.28, p < 0.0001) but not in whole blood (Hannum r =  − 0.05, p = 0.33, PhenoAge r =  − 0.07, p = 0.16) (Supplementary Table 1).

Discussion

In this study, we found statistically significant inverse associations between leukocyte TL and epigenetic age or age acceleration measures calculated using the PhenoAge, GrimAge, and Hannum epigenetic clocks, but not SkinBlood or DNAmAge clocks, after adjusting for chronologic age, sex, and race. Similarly, significant inverse associations were noted between age-adjusted qPCR or flow FISH TL and the immune age acceleration measure, EEAA, but not with the immune-independent measures (IEAA). These associations were consistent with both qPCR and flow FISH TL assays. The modest associations between TL and some, but not all, epigenetic age clocks suggest the dynamic aging processes captured by TL and epigenetic clocks, each measured at the same time point, could reflect varying biologic processes related to cellular aging.

Previous cross-sectional studies reported little to no significant relationship between TL, measured by qPCR in most studies, and DNAmAge clocks (summarized in Table 3) [15–17, 19]. In line with previous findings, our results also showed no significant association between TL and epigenetic age when comparing DNAmAge with flow FISH or qPCR (β =  − 0.35, p = 0.06; β =  − 0.14, p = 0.09, respectively). Unsurprisingly, results for the SkinBlood clock showed similarly null associations. Both the DNAmAge clock and the SkinBlood clock were developed to predict chronological age and share 60 CpG sites; the two clocks also show the highest correlation with one another in our analysis (r = 0.9). On the other hand, our study found significant negative relationships between TL by both flow FISH and qPCR with the PhenoAge clock, and GrimAge clock (p < 0.01 for all after adjusting for age, race, and sex). This suggests that epigenetic clocks that predicted phenotypic age, and not only chronological age, demonstrate relationships with TL.

Table 3.

Summary of published associations between telomere length and commonly used epigenetic age clocks or age acceleration measures

Author (year)	Study population	N	Age	TL assay	Epigenetic age	Associations
Marioni et al. (2016) [19]	LBC1921 LBC1936	920	70 +	qPCR	DNAmAge	Weak to no association
Chen et al. (2017) [18]	WHI FHS BHS	804 909 826	60–70 61–74 40–47	TRF	IEAA1	No association
					IEAA2	Weak to no association
					EEAA3	No association
					EEAA4	Negative association (WHI, FHS) No association (BHS)
Belsky et al. (2018) [15]	Dunedin	1037	38	qPCR	DNAmAge	No association
					Hannum	No association

¹IEAA measured using CPGs from the Hannum epigenetic clock

²IEAA measured using CPGs from Horvarth DNAmAge epigenetic clock

³EEAA measured using CPGs from Hannum epigenetic clock, adjusted for the proportion naive CD8 + T cells, memory CD8 + cells, and plasmablasts

⁴EEAA measured using CPGs from Horvath DNAmAge epigenetic clock

Our results also showed a relationship between leukocyte TL measured by either flow FISH, or qPCR and the immune-based age acceleration measure (EEAA), but not with measures that control for or do not include immune cell counts (IEAA or AAR DNAmAge, respectively). Previous cross-sectional studies using qPCR TL showed weak to no relationship with IEAA (β =  − 0.002, p = 0.007) in the largest study (N = 1895) [16, 17]. On the other hand, a statistically significant (p < 0.001) association between TL and EEAA was reported in a study of TL measured by TRF that included three large cohorts (Women’s Health Initiative, N = 804, r =  − 0.16; Framingham Heart Study, N = 909, r =  − 0.09; and Bogalusa Heart Study, N = 826, r =  − 0.07) [18]. The observed relationships in our current study between TL (by flow FISH or qPCR) and both the Hannum clock and the age acceleration measure EEAA, both derived from blood, could be due, in part, to age-related immune cell composition profile [28]. In flow FISH TL studies, TL is shorter in NK and memory enriched T cells, and longer in naive enriched T cells and B cells [29, 30]. Similarly, a high correlation between white blood cell counts and the Hannum clock or EEAA was previously shown [18]. Our findings combined with those of previous studies suggest the ability of both EEAA and TL to detect immune senescence.

As expected, the strength of association with chronological age was much higher for epigenetic age clocks compared to measured TL. This possibly suggests that the two markers are likely to reflect different aging processes. Epigenetic age clocks were mathematically developed with a focus on age prediction, while TL and its natural shortening primarily reflect cellular proliferative capacity and show significant variations between individuals of the same age group [29, 31]. In analysis comparing TL predicted by DNA methylation (DNAmTL) with actual TL measurement, we previously showed only moderate correlation between the two TL measures, despite DNAmTL’s ability to detect TL differences with sex, race, and chronic infection [20]. Taken together, the current study highlights the possible benefit of considering both TL and epigenetic clocks in studies of age-related health outcomes, given evidence that both measures may capture different aspects of aging. It also underscores the need for more molecular research in understanding the contribution of age-related markers in the human aging process.

The strengths of the current study included the relatively large sample size, use of different assays for TL measurements, and almost all available epigenetic age clocks. The use of the accurate flow FISH TL in a subset validated the results obtained from the high-throughput qPCR TL analysis, which is known by its sensitivity to pre-analytic factors [32]. The correlation between our qPCR assay and flow FISH lymphocyte TL used in this study was previously published (R² = 0.56, p < 0.0001) [33]. Limitations of this study include the limited age range (19–64 years) and cross-sectional data. Our flow FISH TL analysis was limited by the relatively small sample size because of the limited availability of viable cells.

In conclusion, our findings showed that the relationships between TL and epigenetic clocks were mainly observed with clocks that reflect phenotypic age. TL association with EEAA, but not other measures of age acceleration, reflects the ability of both markers to identify immunosenescence. The modest association between the two biological age markers (TL and epigenetic clocks) highlights the importance of incorporating both measures in understanding disease etiology and prognosis.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

Study design: Emily E. Pearce, Rotana Alsaggaf, and Shahinaz M. Gadalla. Sample acquisition: Stephen Spellman. Laboratory and bioinformatics: Geraldine Aubert, Casey L. Dagnall, Shilpa Katta, Steve Horvath, and Belynda Hicks. Statistical analysis: Emily E. Pearce and Rotana Alsaggaf. Data interpretation and manuscript drafting: Emily Pearce, Rotana Alsaggaf, Sharon Savage, and Shahinaz M. Gadalla. Manuscript critical review: all authors.

Funding

This study was supported by the intramural research program of the Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health. The Cancer Genomics Research Laboratory is funded with Federal funds from the National Cancer Institute, National Institutes of Health, under NCI Contract No. 75N910D00024.

The CIBMTR is supported primarily by Public Health Service U24CA076518 from the National Cancer Institute (NCI), the National Heart, Lung and Blood Institute (NHLBI), and the National Institute of Allergy and Infectious Diseases (NIAID); HHSH250201700006C from the Health Resources and Services Administration (HRSA); and N00014-20–1-2705 and N00014-20–1-2832 from the Office of Naval Research. Support is also provided by Be the Match Foundation, the Medical College of Wisconsin, the National Marrow Donor Program.

Declarations

Conflict of interest

The authors declare no competing interests.