Links between biological aging and cancer risk in older German adults

Given the persistent epigenetic changes associated with cancer initiation, progression, and related treatments^7,8,25, we conducted multivariable linear regression analyses (Model 1 and Model 2, as described above) to examine the association between a cancer cases before baseline and BA measured at baseline (Supplementary Table 2). Participants with a cancer diagnosis prior to baseline had higher baseline BA than cancer-free participants for all five BA metrics with coefficient estimates (which represent adjusted differences in BA) ranging from 0.80 to 1.15. However, only PCHannum and PCGrimAge reached the significance threshold after Bonferroni-correction in the fully adjusted models (p-value < 0.01, Supplementary Table 2).

For the long-term risk analysis, the time scale started at the second blood draw for both cases and controls. Compared to short-term risk, BA exhibited a stronger predictive value for long-term cancer risk, as presented in Table 3and Supplementary Table 3. All baseline PC-clocks demonstrated significant associations with long-term cancer risk even after Bonferroni correction for multiple testing (HR per SD ranged from 1.34 to 1.67, p-value < 0.005, Table 3).

We replicated the analyses of the associations of PC-clocks slope with cancer incidence after 8-year follow-up in subgroups classified by sex, age group and family history of cancer (Supplementary Tableto). Increasing cancer risk was seen with increasing slope for all PC-clocks in both men and women, and the Bonferroni-corrected significance threshold was reached for three PC-clocks among men (PCHorvath, PCSkinBloodClock and PCHannum, Supplementary Table). When the models were separately fitted for age groups, stronger associations were seen for older participants (age > 60 years), with slopes in PCHorvath, PCHannum and PCPhenoAge reaching the Bonferroni-corrected significance threshold (Supplementary Table). Furthermore, stronger associations of PCHorvath, PCSkinBloodClock, and PCHannum slopes with cancer incidence were seen among participants without family history of cancer (reaching the Bonferroni-corrected significance threshold) than among those with (Supplementary Table). 4 6 4 5 6

The study population was drawn from the ESTHER cohort, an ongoing population-based cohort study conducted in Saarland, Germany. Details of the study design have been described previously^6,30–32. Briefly, 9,940 individuals aged 50–75 years were recruited by their general practitioners (GPs) during routine health screenings between July 2000 and December 2002, and were followed up every two to three years thereafter. At baseline and each follow-up, standardized self-administered questionnaires were used to collect information on sociodemographic characteristics, lifestyle, and dietary habits. General health examination results were documented by GPs using standardized forms. Blood samples were collected during examinations and stored at −80 °C for future analyses. Cancer incidence and mortality follow-up was conducted by record linkage with cancer and population registries. The ESTHER study adhered to the principles of the Declaration of Helsinki and was approved by the ethics committees of the medical faculty of the University of Heidelberg (Application number: S-58/2000) and the medical board of the state of Saarland. Written informed consent was obtained from each participant.

The ESTHER population has been shown to be representative of the general German population of the same age group with respect to key sociodemographic variables and risk factor profiles³³. For this study, a random sample of 1930 participants were selected from the ESTHER study population for epigenome-wide blood DNAm measurements at baseline that were carried out in two different batches. A subsample of 900 participants also had epigenome-wide blood DNAm measurements at 8-year follow-up.

DNA from whole blood samples obtained at recruitment was extracted using a salting-out procedure³⁴ and paired samples from the same individual taken at 8-year follow-up were processed on the same array to minimize technical variation. Genome-wide DNAm was assessed using the Infinium MethylationEPIC (850 K) BeadChip Kit (Illumina, San Diego, CA, USA), according to the manufacturer’s instructions, by the Genomics and Proteomics Core Facility at the German Cancer Research Center (DKFZ), Heidelberg, Germany^30,35,36. During preprocessing, probes with detection p-values > 0.01, those with >10% missing values, and probes targeting the X or Y chromosomes were excluded^30,35,36. After quality control and exclusion of low-quality samples, a total of 1916 participants were retained for subsequent analyses.

Five PC-based epigenetic clocks were calculated in this study, including PC-Horvath Pan-Tissue (PCHorvath), PC-Skin & Blood Clock (PCSkinBloodClock), PC-Hannum DNAmAge (PCHannum), PC-PhenoAge (PCPhenoAge) and PC-GrimAge (PCGrimAge). PC clocks were calculated using the R script and 78,464 CpG sites per sample, as described in a previous report²⁴. Missing CpG values (less than 5%) were imputed using mean substitution. Age acceleration (PCAgeAccel) was defined as the residual from regressing DNAm age estimates on CA¹⁶, with regressions performed separately for baseline and follow-up measurements. Separate residualization was used to accommodate the longitudinal design with repeated measures and to avoid treating multiple observations from the same individual as independent. For individuals with DNAm data measured at both baseline and 8-year follow-up, individuals BA slope of linear regression was interpreted as the average rate of biological aging per year, calculated as the absolute changes in DNAm age over the time period divided by the time interval between baseline and 8-year follow-up. If the PC clock slope is > 0, it indicates an accelerated rate of epigenetic age change.

Incident cases of cancer, including total cancer (ICD-10 codes C00–C97 excluding the code C44 for non-melanoma skin cancer) diagnosed between 2000 and end of 2022 were identified through record linkage with the Saarland Cancer Registry. The history of invasive tumors before baseline was determined by either self-report or record linkage with data from the Saarland Cancer Registry, which started to record cancers conditions in 1970. The distribution of cancer sites among prevalent and incident cases is summarized in Supplementary Table. 1

Several covariates were considered, including information on sociodemographic characteristics, smoking status, physical activity and alcohol consumption, which were obtained from a standardized self-administered questionnaire. Height and weight were assessed and documented on a standardized form by GPs during the health examination and body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared (kg/m²) and categorized according to the World Health Organization classification.

Standard descriptive methods were used to describe demographic characteristics of the study subjects at baseline. Multivariable linear regression was used to examine the association of a previous cancer diagnosis with BA at baseline. The models were firstly adjusted for age, sex, leukocyte composition (estimated by the Houseman approach)³⁷, and batch (Model 1), and we additionally controlled for educational level ( ≤ 9 years, 10-11 years, and ≥12 years), smoking status (never smoker, former smoker, current smoker), alcohol consumption (grams per day), BMI (kg/m2), physical activity (inactive, low, medium, or high) and family history of cancer (yes/no) (Model 2).

To estimate the predictive value of BA and BA slope on cancer incidence, we performed Cox proportional hazard models with the same adjustments of Model 1 and Model 2. The proportional hazards assumption was checked by scaled Schoenfeld residuals plots³⁸. Hazard ratios (HR) with corresponding 95% confidence intervals (CI) per standard deviation (SD) increase in BA and BA slope were calculated for cancer risk in the entire study population and separately for individuals with repeated DNAm data. Cox proportional hazards models were conducted using time-on-study as the time scale. Follow-up time was defined as age at event or censoring minus age at baseline. Baseline age was included as a covariate in models of baseline epigenetic aging measures, whereas age at the second blood draw was used as the age covariate in slope analyses, reflecting the time point at which the aging slope was defined.

To evaluate the exposure-response relationships between slope of BA and overall cancer incidence, restricted cubic spline (RCS) models with three knots at 25th, 50th and 75th percentiles were applied. We generated scatter plots to evaluate the relationship between the PC-clocks trajectories and the time to cancer diagnosis among the individuals with two-wave methylation data. Furthermore, we conducted subgroup specific analyses for the associations of the BA slope with incident cancer by sex, age ( ≤ 60 years / > 60 years) and family history of cancer (without / with) in individuals with multiple measurements of DNAm. For age-stratified slope analyses, participants were stratified according to chronological age at baseline. Baseline age was used to define age strata to ensure stable group assignment throughout the longitudinal assessment period.

All statistical analyses were carried out using the R software environment for statistical computing (version 3.6.0, Vienna, Austria). Bonferroni-correction was performed to reduce the likelihood of type Ⅰ statistical error, and statistical significance was defined by p-value < 0.05 in two-sided testing ( < 0.05/5 = 0.01 after Bonferroni correction for five different BA metrics).