Introduction Telomeres are protective end structures of the chromosomes. Telomere length is dictated partly by hereditary – and partly by environmental , and epigenetic factors . The hereditary impact on TL has been estimated to range between 36–84% –. An equally strong telomere length inheritance was reported for monozygotic (MZ) as for dizygotic (DZ) twin pairs, indicating that the correlation in TL was mainly due to shared environmental factors . In contrast, relatively minor environmental effects on TL during life were suggested in MZ twins where identical homologue telomeres differed less in TL compared to the two alleles within one individual ,. Regarding the influence of life style and environment on telomere maintenance, the published data are conflicting and no consensus has been reached concerning the impact of e.g. smoking, blood pressure or serum lipids on TL (literature overview in ). Patients with smoking associated malignancies, such as human bladder, head and neck, lung, and renal cell cancers, have been shown to display shorter blood TL at diagnosis compared to controls ,. Short blood TL has therefore been suggested as a predisposition factor for these cancer types. For breast cancer, no difference in blood TL between patients and controls was found in one study , whereas we recently reported longer telomeres in peripheral blood cells of breast cancer patients and, furthermore, that long blood TL indicated a poor survival . Numerous studies have shown an inverse correlation between blood cell TL and age –. Hence, it might be assumed that this characteristic is also true at the individual level. However, data are essentially lacking on individual telomere attrition rates and its relation to the occurrence of malignancy. Martin-Ruiz et al. did not find an association between telomere length at baseline and malignancy related mortality in a longitudinal study on individuals >85 years old . In the present study, we have investigated individual blood cell telomere shortening in a large cohort of voluntarily donated samples. Our novel results show that the attrition rate was strongly correlated to telomere length at baseline, but unrelated to later tumor development. Results In the study cohort of 959 individuals, investigated at two occasions with 9–11 year intervals, an overall TL shortening occurred with age as expected (r = −0.164, P 0.3. This subcohort showed the same statistical outcome (r = −0.623, p 0.3. The result of the restricted correlation analysis was very similar to the result of the larger analysis, showing a strong correlation between telomere length at baseline and attrition rate. A large variation in telomere attrition at the individual level has been observed in previous longitudinal studies on telomere length –. In a very recent study by Aviv et al. , TL was measured in leukocytes collected on two occasions from 450 whites and 185 African Americans, participating in the Bogalusa Heart Study. The median time period between the first and second blood sampling was shorter compared to our study (∼6 years vs. ∼10 years), and the participants were fewer and younger (age range: 20.0–40.0 years at baseline). Nevertheless, they found that the age-dependent TL attrition rate was proportional to TL at baseline, which is in accordance to our present study. The majority of participants in their study displayed TL shortening (85.9% of African Americans and 88.0% of whites), whereas the rest displayed a stable or increased TL. Similar to our observation, they also found that the rate of TL shortening varied considerably among individuals. One explanation to the variations in attrition rate could be differences in epigenetic regulation, with secondary effects on telomere maintenance. Another reason might be that telomerase act preferentially on short telomeres, which has been shown in mice models and cell culture systems –. This would be in line with our observation of a very strong inverse correlation between individual TL at baseline and telomere attrition over time. Interestingly, we found a similar result in our separate family cohort. Hence, comparable data were obtained when analyzing telomere attrition rate at both the individual and at the family level. In the present study, individuals with the shortest TL actually elongated their telomeres over a decade, indicating that the TL maintenance machinery is focused on protecting the shortest telomeres. Nevertheless, other factors are likely to influence the TL attrition rate as well. In our study, the correlation value between blood RTL at baseline (sample 1) and attrition rate was r = −0.752 when analyzing the entire cohort. The corresponding r-squared value is hence 0.566, or ∼57%. This means that the telomere length at time point 1 could explain 57% of the variation in attrition rate. Thus, 43% of the variation might well be explained by other factors, such as life style, oxidative stress, inflammation etc. In the study by Aviv et al. , oxidative stress was proposed as a potential candidate for causing proportional telomere shortening. We agree that oxidative stress is likely to be important for telomere attrition, but the theory does not explain why a subset of the cohort demonstrated TL elongation. We suggest that a cellular TL regulating mechanism, rather than environmental/life style factors, is the major factor determining the rate of telomere attrition over time. The working hypothesis that blood cell TL can indicate a later development of a malignant tumor was not supported in the present study. This hypothesis emanates from data showing altered TL in cases with a variety of malignancies. In urinary bladder, head and neck, lung, and renal cell cancers, shortened blood telomeres have been described at diagnosis, whereas data on breast cancer indicate unchanged or longer telomeres compared to controls –. Since no difference in TL existed between cases and controls, neither ≥9 (sample 1) nor 0–11 (sample 2) years before the appearance of a malignancy, we conclude that blood TL is not a prediagnostic biomarker for malignancy per se. However, our cases suffered from a variety of tumors and we cannot exclude that blood TL might be a biomarker for specific tumor types. A support for this is a recent study indicating that short blood telomeres were associated with a decreased risk for melanoma but also an increased risk for basal cell carcinoma, whereas there was no trend for squamous cell carcinoma . In the largest tumor group in our material, prostate cancer, the blood TL ≥9 years before diagnosis seemed to indicate a poor prognosis. All prostate cancer cases with long blood telomeres (>median) were alive five years after diagnosis compared to 7000 individuals who had donated blood samples at a ∼10 year interval (9–11 years) and of these 343 persons had obtained a cancer diagnosis after the second blood sample (time from sample 2 to diagnosis: 0–11 years, mean 2.7) (Figure 6). From the same cohort, 686 age and sex matched controls were also selected. The age span was 30–61 years for sample 1 and 40–70 years for sample 2. Cancer cases were identified through record linkages with the regional Cancer Register. Due to insufficient amounts of buffy coat cells for DNA extraction or unsuccessful RT-PCR, 314 cases and 645 controls (totally 1918 samples) were included in the statistical analyses (cases: 176 men and 138 women; controls: 361 men and 284 women). 10.1371/journal.pgen.1000375.g006 Figure 6 Schematic drawing of blood draws for sample 1 (baseline) and sample 2 (follow up). To permit analysis of a possible family linked pattern regarding TL attrition, a multifamily cohort was also utilized, initially aimed at studying genetic and environmental factors influencing heredity of personality traits, upbringing, general health and longevity (a study designed and conducted in the late 90's by the author RA). In total, whole blood was available from 962 individuals in 68 families (445 men and 517 women) with an age span of 0–102 years. Thirteen of these families could be selected for the purpose of this study (se statistics below). Ethics Statement The study was approved by the Umeå University Ethical Committee. Telomere Length Determination DNA was extracted from buffy coats and whole blood using conventional methods. Relative telomere length was measured using quantitative real-time PCR as described previously ,. In short, telomeres and a single copy gene (β2-globin) were amplified in all samples including an internal reference control cell line (CCRF-CEM) to which all samples were compared. The ΔΔCt method was used for calculation of RTL values and a standard curve was created in each PCR run to monitor the PCR efficiency. The mean inter-assay coefficient of variation for this method ranges between 4–8% in our laboratory. Statistics Normality was shown regarding RTL distributions. Pearson partial correlation was performed to calculate age-adjusted correlations between continuous variables. ANCOVA was used for age and/or sex adjusted comparisons between groups. Cumulative survival for cancer patients with long vs. short telomeres was investigated using Kaplan-Meier with the log-rank test. Survival was defined as the number of months between diagnosis date to death or to last follow-up (Feb 2008). To investigate whether the rate of telomere loss with age was linked to TL at a young age, 13 separate families in the multifamily cohort were studied. In each family, samples from 10 or more (maximum 28) related individuals, i.e. no in-laws, were available in at least three generations. The age of the individuals in the youngest generation varied between 14 and 32 years and in the oldest generation between 70 and 101 years. The number of men and women was similar within all families except for one which contained more women. The RTL values were plotted against age and linear regression was used to generate intercept (“starting RTL”) and slope (telomere loss) values for each family. The calculated intercepts corresponded to the estimated RTL value at the age of 14. The slope was then plotted as a function of the intercept and the correlation was examined using Pearson's Correlation Coefficient. MLwiN , a software for multilevel analysis, was used to test for parallellity between the 13 regression lines. All other statistics were analyzed in SPSS 15.0. A P-value ≤0.05 was considered to be significant.