Buildup from birth onward of short telomeres in human hematopoietic cells

Abstract Telomere length (TL) limits somatic cell replication. However, the shortest among the telomeres in each nucleus, not mean TL, is thought to induce replicative senescence. Researchers have relied on Southern blotting (SB), and techniques calibrated by SB, for precise measurements of TL in epidemiological studies. However, SB provides little information on the shortest telomeres among the 92 telomeres in the nucleus of human somatic cells. Therefore, little is known about the accumulation of short telomeres with age, or whether it limits the human lifespan. To fill this knowledge void, we used the Telomere‐Shortest‐Length‐Assay (TeSLA), a method that tallies and measures single telomeres of all chromosomes. We charted the age‐dependent buildup of short telomeres (<3 kb) in human hematopoietic cells from 334 individuals (birth‐89 years) from the general population, and 18 patients with dyskeratosis congenita‐telomere biology disorders (DC/TBDs), whose hematopoietic cells have presumably reached or are close to their replicative limit. For comparison, we also measured TL with SB. We found that in hematopoietic cells, the buildup of short telomeres occurs in parallel with the shortening with age of mean TL. However, the proportion of short telomeres was lower in octogenarians from the general population than in patients with DC/TBDs. At any age, mean TL was longer and the proportion of short telomeres lower in females than in males. We conclude that though converging to the TL‐mediated replicative limit, hematopoietic cell telomeres are unlikely to reach this limit during the lifespan of most contemporary humans.


| INTRODUC TI ON
Telomere length (TL) is highly variable across adult humans (Aviv & Shay, 2018). Several lines of evidence indicate, however, that this variation, as expressed in leukocyte TL (LTL), is principally established at birth. First, LTL variation (SD ~ 700 base-pairs (bp)) among newborns is like that among their parents (Factor-Litvak et al., 2016) and adults in the general population (Aubert et al., 2012;Steenstrup et al., 2017). Second, heritability, observed from birth onward, explains a major component of the LTL variation among individuals (Factor-Litvak et al., 2016;Hjelmborg et al., 2015;Slagboom et al., 1994). Third, LTL tracks after the first decade of life, such that individuals maintain their relative LTL ranking, for example, short, or long LTL, compared to their peers (Benetos et al., , 2019. Fourth, exposures and behaviors exert a nominal effect on LTL in adults (Bountziouka et al., 2022;Pepper et al., 2018). Having short or long LTL thus precedes by decades the clinical onset of aging-related diseases. Jointly, this LTL precedence and Mendelian randomization analyses of SNPs associated with LTL infer a causal role of TL in many aging-related diseases and longevity in humans (Codd et al., 2021). It follows that LTL at birth and progressive telomere shortening afterward might contribute to aging-related diseases (Aviv & Shay, 2018;Shay, 2016) and longevity (Arbeev et al., 2020;Codd et al., 2021) in humans. Data of change with age in LTL parameters are thus vital for understanding the role of telomeres in human health and longevity.
Epidemiologists and geneticists have relied on several TL measurement techniques in population studies. These include quantitative polymerase chain reaction (qPCR) (Cawthon, 2002(Cawthon, , 2009), Southern blotting (SB) of the terminal restriction fragments (TRFs) (Kimura, Stone, et al., 2010), and flow-fluorescent in situ hybridization (FISH) (Baerlocher et al., 2006;Rufer et al., 1998). Such techniques measure the mean length of the 92 telomeres at the ends of the p and q arms of the 23 pairs of human chromosomes. However, the shortest telomeres, rather than mean length of all these telomeres, trigger cessation of somatic cell replication (Hemann et al., 2001;Zou et al., 2004).
LTL dynamics (LTL at birth and its age-dependent shortening) have been charted based on measurements of mean TL (Aubert et al., 2012;Steenstrup et al., 2017). However, the buildup of short/ ultrashort telomeres in hematopoietic cells (HCs) over the human life course might be a better indicator of TL-dependent mechanisms that contribute to aging-related diseases. For technical reasons (Materials and Methods), however, SB, considered the "gold standard" of TL measurements, provides limited information about telomeres shorter than (<) 3 kb. A recently developed method, the Telomere-Shortest-Length-Assay (TeSLA), tallies and measures single telomeres (Lai et al., 2017), and provides a comprehensive picture of their distribution that includes telomeres with length of a few 100 bps. These features may explain the ability of TeSLA to generate further understanding of the potential role of telomeres in human biology and disease (Benetos et al., 2021;Lai et al., 2021).

| RE SULTS
We first describe features of SB and TeSLA, and then present data that chart for the first time a progressive buildup of telomeres <3 kb in HCs across nine decades of the human lifespan. patients with DC/TBDs. At any age, mean TL was longer and the proportion of short telomeres lower in females than in males. We conclude that though converging to the TL-mediated replicative limit, hematopoietic cell telomeres are unlikely to reach this limit during the lifespan of most contemporary humans.

| Features of SB and TeSLA
SB (Kimura, Stone, et al., 2010) and TeSLA (Lai et al., 2017) measure the lengths of the TRFs. TeSLA, however, detects and measures, in addition, TRFs that are <3 kb (Materials and Methods, illustrated in Figure S1).
We measured TL parameters in leukocytes from a cohort of 334 individuals (birth-89 years) originating from the general population (Table 1). Mean TL, measured using TeSLA (TeSmTL), strongly correlated with, but was consistently shorter than mean TL, measured with SB (SBmTL) (Table 2; Figure 1, left panel). This difference stems in part from the absence of telomeres <3 kb in the SB output. Excluding telomeres <3 kb from the TeSLA output thus raised the TeSmTL from 4.51 ± 0.77 kilo-base (kb) (SD) to 5.82 ± 0.63 kb, a value that was still ~2 kb shorter than SBmTL (7.83 ± 1.40 kb) for the entire cohort. The TRFs include the sub-telomeric region (Kimura, Stone, et al., 2010), a segment proximal to the canonical telomeres (TTAGGG tandem repeats), known as the X region (Steinert et al., 2004). The length of the X region largely depends on the restriction enzymes used to digest the DNA (Kimura, Stone, et al., 2010), and TeSLA employs different restriction enzymes than SB (Lai et al., 2017). For a subsample (n = 8), we generated, therefore, TRFs for SB by the same set of restriction enzymes used for TeSLA. This resulted in 1.69 ± 0.22 kb (mean ± SD) shorter SBmTL (7.22 ± 0.59 kb vs. 5.53 ± 0.40 kb; Figure 1, right panel

| Cross-sectional data of leukocyte TL parameters over nine decades of life
SBmTL in the cohort showed a steep and rapidly decelerating pace of telomere shortening during childhood, but relatively steady pace during adulthood. (Figure 2, top panel). This trajectory of SBmTL was best described by a model that included the natural logarithm of age (+1) together with age (Table 3). In this model, the linear age term slightly tilted the equation to fit the regression to the long SBmTL of newborns. As per earlier findings, LTL was longer in females compared to males (Aubert et al., 2012;Factor-Litvak et al., 2016;Gardner et al., 2014;Hjelmborg et al., 2015;Hunt et al., 2020;Steenstrup et al., 2017) (Table 3, and below). We next compared SBmTL in our sample with SBmTL in seven patients with dyskeratosis congenitarelated telomere biology disorders (DC/TBDs) (Alter et al., 2012;Niewisch et al., 2022). In previous work we defined the TL-mediated replicative limit in human HCs as the "telomeric brink" based on SBmTL in these patients (Steenstrup et al., 2017). In general, SBmTL in the oldest participants from the general population showed little overlap with the SBmTL in DC/TBDs ( Figure 2, top panel).
TeSmTL declined similarly with age as SBmTL ( Figure 2, bottom panel). It was also best described by a model that included the natural logarithm of age (+1) together with age and sex (Table 3) For comparisons of the age-dependent trajectories of TeSmTL with SBmTL, both sets of measurements were transformed to a

F I G U R E 1
The relation between mean TL, measured by SB versus mean TL, measured by TeSLA. TeSmTL (kb) in relation to SBmTL (kb). N = 334. Bold Line is the regression line (Table 2). Also shown is line of unity (Y = X). Right panel. Mean (±SE) TL (kb) was estimated in four ways. Light grey lines connect different methods applied to the same samples (n = 8). Methods: TeSLA: TeSmTL; TeSLA 3 kb, TeSmTL but calculated excluding telomeres <3 kb; SB 4R: SBmTL but DNA digested using the 4 restriction enzymes used in TeSLA; SB 2R: SB with DNA digested using the 2 restriction enzymes routinely applied. For comparability, both TRF estimates were calculated over the same range (3-20 kb). TeSLA generates data of single telomeres (162.9 ± 48.0 telomeres/sample in this study), enabling examination of within-sample single TL variation. Figure 4 displays the single-telomere TL distribution for four age groups. It shows a shift toward shorter single telomeres and a compression of the distribution with age, as confirmed by a decline with age of the SD of single-telomere TL, calculated within samples (t = 10.22, n = 334, p < 0.0001).
Focusing on the age-dependent increase in the proportion of telomeres <3 kb (TeS3kb), we found that the TeS3kb increase with age ( Figure 5) was best described with the same combination of terms as TeSmTL (Table 3). As per TeSmTL, little overlap was observed between TeS3kb in participants from the general population and DC/TBD patients (illustration of TeSLA for a patient with a DC/TBD disorder vs. a healthy adult of the same age is shown in Figure S2). To evaluate the robustness of this inference, given the uncertainty in estimating the telomeric brink, we also estimated the age at which the telomeric brink plus one SD (or −1 SD for TeS3kb) is reached, and found this to be at ages of 129, 132, and 142 for SBmTL, TeSmTL, and TeS3kb, respectively. We stress, however, that these estimates are based on a relatively small sample and extrapolations over more than 80 years, and they simply serve to illustrate the unlikelihood of reaching the telomeric brink, given the present life expectancy.

| The sex effect on mean and short LTL dynamics over the lifespan
Males displayed a shorter LTL than females ( to SBmTL (p = 0.063, Table 3). While numerically, the sex gap differed only slightly between results generated by the two methods, the explained variance by the (identical) complete model was lower for TeSLA based data (Table 3). We illustrate the more pronounced sex effect in the TeSLA-based data by calculating the standardized difference ( Figure 6), which is the sex effect scaled to the residual variation without sex (models in Table 3).

| DISCUSS ION
Our knowledge of LTL change with age is fragmented and surprisingly limited (Aubert et al., 2012;Steenstrup et al., 2017). Most population-based research relies on qPCR measurements, which generate data in relative units, that is, amplified telomere product (T) divided by a single-gene product (S), without information on absolute TL. T/S data, whose metric differs across laboratories, thus generate little quantitative insight on LTL shortening with age. Moreover, SB and flow-FISH, which is calibrated by SB, do not usually measure short telomeres among the telomeres in each cell (Kimura, Stone, et al., 2010), and thus overestimate LTL.
We also note that previous studies suggested that the mean value of the X region might be more than 3 kb across individuals (Cawthon, 2002;Hultdin et al., 1998), but our findings do not support this view, because HCs from patients with DC/TBDs display TRFs shorter than 0.5 kb (illustrated in Figure S2); these include the canonical region and the X region. Although we do not know which chromosomes or cells are the source of these ultrashort telomeres, the X region for these telomeres is shorter than 0.5 kb. This means that the trajectories of telomeres <3 kb we show in this work reflect changes with time that principally apply to the canonical component of the telomeres.

F I G U R E 3
Comparison of association with age between mean TL by TeSLA, with mean TL by SB. TeSmTL, mean TL by TeSLA; SBmTL, mean TL by SB. Upper panel. TL measurements with each method was transformed to standard normal distribution (mean = 0, SD = 1). Lower Panel. Telomere shortening in relation to age estimated using TeSmTL and SBmTL (first derivatives of regression lines in Figure 2 (for model see Table 3). SBmTL, blue line, TeSmTL, green line. In both panels, lines represent the pattern averaged over the sexes.  (Aubert et al., 2012). In addition, patients with DK/TBDs often present with neutropenia in addition to aplastic anemia (Niewisch et al., 2022).

F I G U R E 4 Distribution of TL (kb) within samples measured using
While the accumulation of HC telomeres <3 kb during the human life course might not severely affect erythropoiesis under steady state conditions, in some adults it can still limit the ability of lymphoid cells, particularly T cells, to undergo clonal expansion in response to infection. The magnitude of this expansion, a model suggests, might be stymied by age-dependent HC telomere shortening (Anderson et al., 2022). However, such a model is based on LTL, that is, mean TL, data. Except for small studies (Benetos et al., 2021;Lai et al., 2021),

no information is available about whether the shortest telomeres in
HCs and other somatic cells provide a better insight of TL-dependent biological processes and health outcomes in humans. Moreover, leukocytes consist of cells of myeloid and lymphoid lineages with different rates of age-dependent shortening with age (Aubert et al., 2012), whereas our trajectories are based on TL parameters of all these lineages. We note, however, that within an individual, TL variation across these lineages and other somatic cells is much smaller that the inter-individual TL variation (Aubert et al., 2012;Daniali et al., 2013;Kimura, Gazitt, et al., 2010). Therefore, we anticipate that the broad dynamics of TL shortening and buildup of the shortest telomere we show in this work would apply to specific lineages. That said, as indicated more than a decade ago, the focus on a specific leukocyte lineage largely depends on the investigators' TL-related hypothesis (Kimura, Gazitt, et al., 2010).
LTL is shorter in males than females from birth onward (Aubert et al., 2012;Factor-Litvak et al., 2016;Gardner et al., 2014;Hjelmborg et al., 2015;Hunt et al., 2020;Steenstrup et al., 2017), as confirmed in the present study. Statistical support for this sex gap was considerably stronger for TeSmTL and TeS3kb than for SBmTL. This difference cannot be attributed to greater precision of TeSLA compared to SB, because repeatability of the latter method is slightly higher (see We acknowledge limitations of the study, including the following: Although TeSLA is a reliable method to tally and measure single telomeres, the method does not identify the chromosomal origins of these telomeres. Our study is a composite of diverse participants, originating from different groups in different countries. We have not explored potential health outcomes in participants from the general population since sample sizes for any given age group was too small.
In conclusion, this is the first study showing the buildup of short telomeres in HCs from birth onward over nine decades of human life.
TeSmTL and the ability of TeSLA to capture short/ultrashort telomeres provide the most refined picture of LTL dynamics in humans.
Based on data in the general population and in patients with DC/ TBDs, the study suggests that the buildup of telomeres <3 kb over the lifespan might be insufficient to severely impede erythropoiesis under steady state condition in most healthy individuals as they get older. Therefore, the contribution of HC TL dynamics to agingrelated human diseases and longevity (Codd et al., 2021) might be exerted through other mechanisms, for example, clonal hematopoiesis of indeterminate potential (Aviv & Levy, 2019;Nakao et al., 2022).
In addition, the study has generated further insight into the X region, suggesting that it is shorter than that previously estimated.

F I G U R E 6
Sex difference in telomere length Estimates (± SE) are for three different telomere metrics, SBmTL, mean TL by SB; TeSmTL mean TL by TeSLA, and TeS3kb proportion of telomeres shorter than 3 kb. All three expressed relative to the pooled standard deviation of that metric calculated over the residuals of the models in Table 2, without sex.

| Subjects
LTL parameters were measured in DNA samples from 334 participants from five studies (Table 1). SBmTL data from four studies were published (Benetos et al., 2018(Benetos et al., , 2019Factor-Litvak et al., 2016;Nettle et al., 2021), and one study is ongoing. For the present study, we performed TeSLA on 18 patients with DC/TBDs and had enough DNA to perform SB in 7 of these patients. All TL measurements were performed at the Rutgers' laboratory. Participants in all studies provided written informed consents approved by ethic committees and institutional review boards for the research use of their samples.

| LTL measurements
DNA samples for both SB and TeSLA passed an integrity test by resolving 20 ng of DNA on a 1% (w/v) agarose gel.
SB measurements were performed (in duplicate) as previously described (Kimura, Stone, et al., 2010). Briefly, DNA was digested using the Hinf I and Rsa I restriction enzymes (Roche Applied Sciences, Mannheim, Germany). Digested DNA and DNA ladders were resolved on 0.5% agarose gels for 16 h (2 V/cm). We also performed SB measurements de novo in eight DNA samples digested with the restriction enzymes used for TeSLA (below). Precision of SBmTL measurements as indicated by the ICC is 0.98 (Nettle et al., 2021).
TeSLA measurements: These measurements were performed as previously described . In brief, extracted DNA is ligated at the overhangs of telomeres to single-stranded adaptors that contain seven nucleotides of telomeric C-rich repeats at the 3′ end, which is complementary to the G-rich overhang followed by a unique sequence for PCR. The DNA was then digested with restriction enzymes (BfaI,
TeS3kb and TeS4kb were arcsine-square-root-transformed to meet the homoscedasticity assumption. As LTL shortens at a faster rate in early life, we fitted different combinations of terms, including polynomials (linear, squared, and cubed) and age natural log-transformed (using age +1 because of newborns' age = 0), and compared the fit to the data using the Akaike information criterion and visual inspection of graphs, including raw data and the model fit. We also fitted the correlations of TL parameters with age using restricted cubed splines, leveraging the R package Hmisc. While the latter approach enables fitting very complex patterns, it has a limited ability to interpret results beyond visual inspection. We used the fitted spline to determine visually equations that reveal a close match between the spline and the equation fitted using a conventional approach. Cohort identity was included as random effect in the analyses since DNA donors came from five cohorts of different ages and countries of residence. Reported R 2 values refer to variation explained by the fixed effects only. Males have shorter LTL than females and we thus included sex in all analyses. Interactions between age and sex were not significant (not shown).

CO N FLI C T O F I NTE R E S T S TATE M E NT
All other authors declare they have no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data are available in the main text or the supplementary materials.