Identification of seven loci affecting mean telomere length and their association with disease

Telomeres are the protein bound DNA repeat structures at the ends of chromosomes, which have an important role in maintaining genomic stability¹. Furthermore, they play a critical role in regulating cellular replicative capacity². During somatic cell replication, telomere length (TL) progressively shortens due to the inability of DNA polymerase to fully replicate the 3’ end of the DNA strand. Once a critically short TL is reached, the cell is triggered to enter replicative senescence and subsequently cell death^{1, 2}. Conversely, in germ cells and other stem cells requiring renewal, TL is maintained by the enzyme telomerase, a ribonucleoprotein which contains the RNA template TERC and a reverse transcriptase TERT³. Both longer and shorter TL are associated with increased risk of certain cancers^{4, 5}, and reactivation of telomerase, which by-passes cellular senescence, is a common requirement for oncogenic progression⁶. Therefore, TL is an important determinant of telomere function.

Mean TL shows considerable inter-individual variability and has high heritability with estimates varying between 44–80%^7–9. Most of these studies have measured mean TL in blood leukocytes. However, there is evidence that within an individual, mean TL of leukocytes (LTL) and other tissues is highly correlated^{10, 11}. In cross-sectional population studies mean LTL is longer in women than in men and is inversely associated with age (declining by between 20–40 base-pairs (bp) per year)^{9, 12–14}. Furthermore, shorter age-and sex-adjusted mean LTL has been found to be associated with risk of several age-related diseases, including CAD^12–15 and has been advanced as a marker of biological ageing¹⁶. However, the extent to which the association of shorter LTL with age-related disorders is causal in nature remains unclear. Identifying genetic variants that affect TL and testing their association with disease could clarify any causal role.

So far, common variants at two loci on Chr3q26 (TERC)^17–19 and Chr10q24.33 (OBFC1)¹⁸, that explain <1% of the variance in TL, have shown a replicated association with mean LTL in genome-wide association (GWA) studies. To identify further genetic determinants of LTL we conducted a large scale GWA meta-analysis of 37,684 individuals from 15 cohorts, followed by replication of selected variants in a further 10,739 individuals from 6 additional cohorts.

Details of the studies included in the GWA meta-analysis and in the replication phase are given in Supplementary Note and key characteristics summarised in Supplementary Tables 1a and 1b, respectively. All subjects were of European descent, the majority of the cohorts were population-based and three of the replication cohorts were additional subjects from studies used in the meta-analysis. The genotyping platforms and the imputation method (to HapMap 2 build 36) used by each GWA cohort are summarised in Supplementary Table 2. Mean LTL was measured in each cohort using a quantitative PCR method and is expressed as a T/S ratio (Online Methods, Supplementary Note).

LTL, adjusted for age, sex, and any study-specific covariates was then analysed for association with genotype using linear regression within each study and results adjusted for genomic inflation control factors (Supplementary Table 2). An inverse-variance weighted meta-analysis for 2,362,330 SNPs (Online Methods) was performed with further correction for the overall genomic inflation control factor (λ = 1.007). The quantile-quantile plot for the meta-analysis is shown in Supplementary Figure 1.

SNPs in 7 loci showed association with mean LTL at genome-wide significance (P<5x10⁻⁸, Fig 1, Table 1, Fig 2 and Supplementary Fig 2). The association of the lead SNP on chromosome 2p16.2 (rs11125529) was very close to the threshold for genome-wide significance , while the lead SNP in a further locus on 16q23.3 (rs2967374) fell just short of this threshold (Table 1). We therefore sought further validation of these two loci. We confirmed the association of rs11125529 but not of rs2967374 (Table 1). The combined P value from the GWA meta-analyses and replication cohorts for rs11125529 was 7.50x10⁻¹⁰. There was no evidence of sex-dependent effects or additional independent signals at any of these loci (Online Methods, Supplementary Tables 3 and 4).

Details of key genes in each locus associated with LTL and their location in relation to the lead SNP are given in Supplementary Table 5. The most significantly associated locus we found was the previously reported TERC locus on 3q26 (Figs 1 and 2, Table 1)¹⁷. In addition to this, four further loci – 5p15.33 (TERT), 4q32.2 (NAF1, nuclear assembly factor 1), 10q24.33 (OBFC1, oligonucleotide/oligosaccharide-binding fold containing 1)¹⁸ and 20q13.3 (RTEL1, regulator of telomere elongation helicase 1) - harbour genes that encode proteins with known function in telomere biology^{3, 20–23}. NAF1 is a protein which is required for H/ACA box snoRNA assembly, the RNA family to which TERC belongs²⁰. Therefore, the three most significantly-associated loci (3q26, 5p15.33 and 4q32.2) harbour genes involved in the formation and activity of telomerase. We therefore examined whether the lead SNPs at these loci as well as the other identified loci associate with leukocyte telomerase activity in available data from 208 individuals. We did not find significant association of any of the variants with telomerase activity (Supplementary Table 6). However, the study only had 80% power (α of 0.05) to detect a SNP effect that explained 3.7% of the variance in telomerase activity and therefore smaller effects are likely to have been missed in this exploratory analysis.

We also replicated the previously reported OBFC1 locus¹⁸. OBFC1 is a component of the telomere binding CST complex also containing CTC1, and TEN1²¹. In yeast this complex binds to the single stranded G overhang at the telomere and functions to promote telomere replication. RTEL1 is a DNA helicase that has been shown to play important roles in setting telomere length, telomere maintenance and DNA repair in mice^{22, 23}. However, it should be noted that the lead SNP lies 94 kb from RTEL1. The remaining two loci (19p21 and 2p16.2) do not harbour obvious candidate genes related to telomere biology. The locus on 19p12 contains a cluster of genes encoding zinc finger proteins (ZNFs) while that on 2p16.2 spans both the ACYP2 gene which encodes a muscle specific acylphosphate and TSPYL6, a gene within intron 3 of ACYP2 that shares homology with nucleosome assembly factors. There is evidence that ACYP2 is linked to stress induced apoptosis in rat muscle²⁴.

In order to gain better functional insight into the associated loci we undertook various bioinformatics analyses (Online Methods). Details of the findings are given in the Supplementary Note and in Supplementary Table 7. SNPs in high LD with the lead SNP were found to lie within potential regulatory elements of TERC, NAF1 and OBFC1. However, similar SNPs were also present for other genes in some of the loci. These findings emphasise that although strong candidate genes are located in some of the loci, at this stage we cannot overlook the potential involvement of other genes within each region.

Each of the identified loci explains a relatively small proportion of the total variance in LTL (Table 1). In order to put this in context, we calculated the effect of the lead SNP at each locus in terms of equivalent age-related shortening of LTL based on an estimate of age-related attrition in T/S ratio calculated across all cohorts (Supplementary Fig 3). We saw per-allele effects using this measure equivalent to between 1.9–3.9 years of age-related attrition in T/S ratio (Table 1). The qPCR method used here to measure LTL cannot be used to directly calculate the effect on LTL in base pairs (bp). However, many prior studies that have used Southern blotting to measure LTL have shown that mean LTL attrition rate is about ~30 bp per year^{8, 12–14, 25}. This suggests that the per allele effect of the different SNPs on LTL in base pairs ranges from ~ 57 to 117 bp (Table 1).

As both shorter and longer mean LTL have been linked to increased risk of various diseases, we searched genetic association databases for disease associations with the LTL associated SNPs (Supplementary Table 8). The rs10936599 (TERC) allele associated with longer LTL associates with increased risk of colorectal cancer¹⁹ and with two autoimmune diseases, multiple sclerosis (longer LTL allele) and celiac disease (shorter LTL allele). The lead SNP for the 5p15.33 (TERT) locus is associated with different cancer types (both shorter and longer LTL alleles) and with increased risk of idiopathic pulmonary fibrosis (shorter LTL allele), a disease that has previously been shown to be associated with shorter LTL²⁶.

One of the most widely reported associations for LTL to date has been that between shorter mean LTL and CAD^{12–14, 25}. Because LTL is also affected by other risk factors for CAD such as oxidative stress^27–29, it has been unclear whether the association of shorter LTL with CAD is primary or secondary. To investigate whether the association could be causal, we examined the association of both individual lead SNPs and a genetic risk score (GRS) based on a combination of all seven SNPs, (adjusted for their effect size) with CAD in the CARDIoGRAM GWAS meta-analysis comprising 22,233 CAD cases and 64,762 controls of European descent³⁰, using the approach recently described by the ICBP Consortium³¹. Although the results for individuals variants were not significant, 6 out of 7 variants showed consistency in direction and the combined GRS analysis showed a significant association (p = 0.014) of the allele associated with shorter LTL with increased risk of CAD (Fig 3). Shorter mean LTL equivalent to one SD in LTL was associated with a 21% (95% CI 5% –35%) higher risk of CAD.

Here, we report five novel and confirm two previously reported loci that associate with mean LTL in humans. A specific motivation for our study is the observation that variation in LTL is associated with several age-related diseases and the desire to establish whether this link is causal. This is particularly challenging to disentangle because other environmental and life-style factors also impact on TL^{29, 32–34}. The most persuasive evidence for a causal role comes from in vitro and in vivo manipulation of telomerase activity which impacts on TL and has been shown to enhance or reverse senescence and ageing-associated phenotypes^35–39. Here, we show that some of the genetic variants associated with LTL are also associated with risk of specific cancers as well as other diseases some of which have been shown to be previously associated with shorter LTL, suggesting a causal link. An interesting finding was that alleles associated with both shorter and longer telomeres showed associations with specific cancers suggesting that variation in LTL in either direction may contribute to the development of specific cancers.

As an example of a complex disease that has been shown to be associated with shorter LTL we examined CAD. Through an analysis of a large GWA database of CAD³⁰, we found that while individually the lead SNPs at each of the TL-associated loci were not significantly associated with risk of CAD (probably at least in part reflecting their weak individual effects on LTL and low power), in a combined analysis, alleles associated with shorter LTL were associated with a significantly higher risk of CAD. Because the variants at each of the loci could have other biological effects that could impact on their association with CAD through LTL (and possibly explain why the NAF1 locus may be trending in the opposite direction), some caution is required in the interpretation of this association. Nonetheless, the finding is consistent with that in the prospective WOSCOPS study where, after adjustment for other CAD risk factors, baseline LTL was associated with a 44% higher risk of CAD over the ensuing mean 5.5 years of follow-up in individuals in the tertile with the shortest LTL compared with the longest LTL¹³. Our finding here therefore supports a causal association of shorter LTL with CAD and further mechanistic investigation of this relationship is warranted.

In summary, we provide novel insights into the genetic determination of a structure that is critically involved in genomic stability and cellular function. Our findings suggest that variants in several genes encoding proteins with known function in telomere biology as well as other genes influence LTL. The findings provide a framework for a genetic approach to investigating the causal role of telomere length in ageing-related diseases.