Leukocyte Telomere Length and Its Interaction with Germline Variation in Telomere-Related Genes in Relation to Pancreatic Adenocarcinoma Risk

Study population

Following approval by the Mayo Clinic Institutional Review Board (Rochester, MN), data and biospecimens were obtained from the Mayo Clinic Biospecimen Resource for Pancreas Research, a patient registry supported by the Mayo Clinic Specialized Program of Research Excellence in Pancreatic Cancer (3, 5). The registry utilizes an ultrarapid case ascertainment process, which ensures that approximately 86% of PDAC participants identified at Mayo Clinic are recruited within 30 days of diagnosis, with an average of 2 weeks between first contact and recruitment. The cases included in this study had pathologically (∼95%) or clinically confirmed diagnosis of PDAC, recruited between October 2000 and June 2016. The methods used to identify controls have been described previously (35). In brief, controls were healthy patients recruited from primary care clinics at Mayo Clinic during regular wellness check visits. The controls did not have a prior history of cancer (except nonmelanoma skin cancer) and were recruited between May 2004 and March 2012. The controls used in the present study were selected from a larger pool of controls through frequency matching to cases based on age (±5 years), sex, and race (White and other). In the patient registry, participation rates were approximately 67% for cases and 70% for controls (3, 5, 35). In our experience, nonparticipation by cases is due mainly to the debilitating nature of PDAC and the often rapid fatality associated with PDAC. In all, 1,500 controls were frequency matched to 1,500 cases for the study. All participants provided written informed consent.

Data collection

The participants' completed standardized risk factor questionnaires that sought information on demographics (age, sex, and race/ethnicity), highest level of education, self-reported personal history of diabetes, cigarette smoking history, and usual adult weight and height. Smoking history was categorized as current, former, and never, and self-reported usual adult weight and height were used to calculate body mass index (BMI) in kg/m².

TL measurement

Genomic DNA was isolated from peripheral blood buffy coat leukocytes stored at −80°C until DNA extraction. DNA extraction was performed using the Maxwell RSC Instrument with appropriate kits (Promega) and quantitated using a Qubit fluorometer. Details of the methods used for LTL measurement have been described previously (36). Briefly, quantitative PCR (qPCR) was used to measure LTL, expressed as relative telomere to single-copy gene (T/S) ratio and was based on the average of three measurements per sample. The qPCR-based assay uses one pair of primers to amplify the telomeric hexanucleotide repeats and another pair to amplify a single-copy gene. The primer sequences, written 5′–3′, were:

• telg: ACACTAAGGTTTGGGTTTGGGTTTGGGTTTGGGTTAGTGT

• telc: TGTTAGGTATCCCTATCCCTATCCCTATCCCTATCCCTAACA

• HBBugc: CGGCGGCGGGCGGCGCGGGCTGGGCGGCTTCATCCACGTTCACCTTG

• HBBdgc: GCCCGGCCCGCCGCGCCCGTCCCGCCGGAGGAGAAGTCTGCCGTT

The DNA samples were diluted in dilution buffer, consisting of 10 mmol/L Tris-HCl, 0.1 mmol/L EDTA, pH 8.0, “TE buffer,” with BSA added to 100 ng of DNA per microliter. Dilution buffer without DNA also was assayed in quadruplicate on each 384-well plate as the no template control. A concentrated stock (175 ng of DNA/μL of TE buffer) of pooled samples from several healthy Utah subjects aged ≥65 years (“standard pooled DNA”) was serially diluted into the dilution buffer to produce five DNA concentrations: 27, 9, 3, 1, and 0.333 ng of DNA per 5 μL, in sufficient quantities for four replicate qPCR reactions per DNA concentration for each 384-well qPCR plate. The standard curve serial dilution, always based on the same standard pooled DNA, was freshly prepared for each 384-well qPCR run. Three reference DNAs with known low, middle, and high TLs (T/S values) relative to that of the standard pool DNA (whose T/S value is, by definition, 1.00) were assayed in quadruplicate on each 384-well qPCR plate. Input DNA amount was approximately 3 ng per reaction. DNA samples from the participants were diluted into the dilution buffer in a 96-well source plate, with one subject's DNA sample per well, at a concentration of approximately 3 ng per 5 μL, in sufficient quantities for three replicate qPCR reactions per subject per 384-well qPCR plate. All qPCRs were run on a Bio-Rad CFX384 Real-Time PCR Detection System. Five microliters of 2x Master Mix (2x Klentaq1 Reaction Buffer, pH 7.9, DNA Polymerase Technology; 2 mol/L betaine solution, Sigma-Aldrich; 0.4 mmol/L of each of the four dNTPs, Thermo Fisher Scientific; 2x SYBR Green I, Thermo Fisher Scientific; 100 ng/μL of BSA, New England Biolabs; 2x Titanium Taq DNA Polymerase, Takara Bio USA, Inc.; and 400 nmol/L of each of the four primers) was added to each well that would subsequently be receiving DNA (or dilution buffer without DNA), after which 5 μL of DNA (standard curve, reference, or subjects) or dilution buffer alone was added. Plates were then sealed with adhesive transparent film, shaken to mix master mix and sample, and centrifuged at 700 × g for a few seconds. Thermal cycling was performed as follows: (i) 94°C × 3′, (ii) 98°C × 2″, 49°C × 15″; two cycles, (iii) 98°C × 2″, 65°C × 10″, 74°C × 15″ with signal acquisition, 84°C × 10″, 86°C × 15″ with signal acquisition; 31 cycles, and (iv) melt curve from 72°C to 96°C, 0.5°C steps, each for 5″ with signal acquisition. Analyses were performed using the Bio-Rad CFX Manager 3.1 software with relative quantification by the standard curve method. Following this protocol, LTLs (T/S values) were determined for the 1,500 cases and 1,500 controls; however, 40 case samples and 41 control samples failed assay design or quality control, leaving a final sample of 1,460 cases and 1,459 controls. The intra- and interassay coefficient of variation (CV) based on the assay standards were <3%. The median intra-assay CV for the input DNA was 5.5%. Additional details about intra- and interassay reproducibility of T/S ratio measurements by qPCR has been provided by Cawthon (36).

SNP selection and genotyping

We selected 11 SNPs that have been associated with interindividual variation in measured LTL in genome-wide scans (15–17) or candidate gene studies (18). The SNPs are located in or near the following genes: TERC (rs10936599 and rs1317082), TERT (rs2736100 and rs7726159), NAF1 (rs7675998), OBFC1 (rs9420907 and rs2487999), ZNF208 (rs8105767), ACYP2 (rs11125529), PXK (rs6772228), and CTC1 (rs3027234; refs. 15–18). Genotyping of leukocyte DNA was performed by the Mayo Clinic Genome Analysis Core using the Sequenom multiplex assay (37). The genotyping call rates and concordance with blinded duplicates were each 100%. All allele frequencies were in Hardy–Weinberg equilibrium.

Statistical analyses

Distribution of participant characteristics and selected telomere-related SNPs were compared between the cases and controls using means and proportions. The measured LTLs were categorized into tertiles based on sex-specific distribution among controls, and to maximize sample size and avoid small cell sizes in subgroup analyses. For age-stratified analyses, both age- and sex-specific distributions among controls were used to categorize the LTLs into tertiles. Before performing the main association analyses, we examined the distribution of participant characteristics across LTL tertiles among control subjects using ANOVA for continuous variables and χ² tests for categorical variables. We also verified reported relationships between the candidate SNPs and LTL using linear regression, with adjustment for age (continuous), sex, BMI (continuous), smoking status (never, former, and current), and diabetes (yes, no; both preexisting diabetes and PDAC-induced diabetes; ref. 1). For these analyses, each SNP was modeled separately as predictor and LTL as the outcome and per-allele β-estimates were calculated for alleles reported to be associated with short LTL, using the reported long telomere allele as reference (15–17).

The association between LTL and PDAC risk was examined using unconditional logistic regression to calculate ORs and 95% confidence intervals (CI). The association was examined in multivariate models, adjusting for age, sex, race, diabetes, BMI, and smoking history, and using the longest LTL tertile as reference. Linear trend across tertiles was assessed on the basis of the ordinal method. We also modeled LTL as a continuous predictor variable, adjusting for covariates. Some of the PDAC cases included in the study had already started cancer-directed therapy before blood collection. Because cancer treatment can affect LTL (38), we performed subset analyses among cases with treatment-naïve blood samples (n = 653) and cases who had started cancer treatment before blood collection (n = 773), with each group compared with the same controls. Data on treatment status were not available for 34 participants who were excluded from the subset analyses. Stratified analyses were performed by age (≤65 and >65 years), sex, diabetes (yes and no), BMI (<25, 25–29.9, and ≥30 kg/m²), and smoking history (never, former, and current). To determine whether the association between LTL and PDAC differed by germline variation in telomere-related SNPs, we performed stratified analyses by SNPs found to be significantly associated with variation in LTL among controls based on additive allele modeling. We also performed stratified analyses by the total number of alleles associated with shorter LTL carried by each participant across all 11 SNPs (≤9, 10–11, and ≥12 short allele counts) and number of short telomere alleles based on three SNPs (rs1317082, rs2487999, and rs7726159; 0–2, 3–4, and 5–6 short allele counts) as has been done by others (18). All stratified analyses included assessment of multiplicative interaction. Exploratory analyses were performed with restricted cubic spline functions with LTL as predictor variable in the unconditional logistic regression model to assess the possibility of a nonlinear association between LTL and PDAC risk. Statistical analyses were performed with SAS v9.4 (SAS Institute), and a two-sided P < 0.05 was considered statistically significant.

Data availability statement

The data may be made available to researchers upon request to G.M. Petersen (Petersen.gloria{at}mayo.edu). Ethical and legal restrictions apply to these data.